 |
| What model is this? (Atta) |
"A shared generative model allows for the emergence of communication and cultural dynamics."There it is. That's the sentence I've been looking for, and last night I found it. It was in a chapter that was coauthored by four people, one of whom was the young researcher whose Twitter feed I was looking at. Deeply impressed, I resolved to read all of his carefully written tweets and found more. But first, let me summarize this interesting chapter, titled "A Multi-scale View of the Emergent Complexity of Life". Near the beginning of the chapter we read: “Organisms themselves are the implicit model for which they gather evidence, resulting in the interpretation that they produce evidence for their own existence... Therefore, we are now in a position to interpret processes of adaptation as collecting model evidence and, by extension, to cast natural selection as a form of model selection.” These statements echo what has already been said by Friston, more or less. In short, an organism is a model of the world around it, and it seeks to maximize the evidence for itself. Those that can do this survive. Those that can’t, die.
It continues: “If we assume that every (pluripotential) cell starts with a generative model of the entire ensemble... cells that operate with the same generative model can self-organize into pre-determined structures (organs). This allows us to understand how an intricate functional structure like the brain can be produced by the (epi)genetic information transmitted at conception. This treatment has prepared us for a discussion of the brain, entire organisms, and their interactions. Interestingly, we will see a similar sort of dynamics emerge in the interaction between multiple organisms: a shared generative model allows for the emergence of communication and cultural dynamics.” And there it is: we need to identify the best shared models we have that allow us to organize as a functional whole, maximize evidence, and adapt to changing conditions. One question I have is whether it would be more appropriate to call this a "shared stigmergic model" of the entire ensemble. I'm going to need some time to go through the other papers Hesp linked to (see bottom).
Psychohistory or an "ecobiopsychosocial framework"
In 2017, Ramstead and Friston coauthored a paper, with Paul Badcock of the University of Melbourne, titled Answering Schrödinger’s question: A free-energy formulation. After the publication of Ramstead’s paper, Micah Allen, a cognitive neuroscientist then at the FIL, wrote that the free energy principle had evolved into a real-life version of Isaac Asimov’s psychohistory. Allen wrote: "It seems that even in science, life imitates art. In their ambitious new proposal, Ramstead et al. link the dynamical interaction of cells, brains, persons, and even society itself through the lens of the free energy principle (FEP). By expanding the FEP to define a new “variational neuroethology” (VNE) of biological life, Ramstead et al. propose “...to explain and predict how living systems, at any spatial and temporal scale instantiate the dynamics of adaptive free energy minimization”. Here the self-organizing minimization of information-theoretic free energy interlinks every conceivable evolutionary and ontogenetic timescale."
Ramstead et al. wrote: "Cultural ensembles minimise free energy by enculturing their members so that they share common sets of beliefs. Human beings - with our specific forms of neural organisation, behavioural tendencies, and sociocultural patterns - minimise more free energy across spatial and temporal scales than any other species. Arguably, this is because we have crossed the ‘evolutionary Rubicon’ [term from Henrich]: our survival depends on our ability to access and leverage cultural information and immersively participate in culturally adapted practices. Each individual throughout the course of evolution, along with every individual that has either existed in the past or occupies the present, may alter the course of our evolution and environment. The human Markov blanket is nested within the broader dynamics of other global Markov blankets that extend out into the universe, these lie beyond the limits of the system that this ecobiopsychosocial framework endeavours to explain." The purpose of the article by Ramstead et al. was to answer Schrödinger’s question – ‘what is life?’ – by presenting a formulation that offers the sciences of life, mind, behaviour and society with a unifying perspective that explains the dynamics of living systems across all scales, and offer promising avenues for interdisciplinary research. The challenge, of course, lies in translating theory into productive scientific practice.
Sentience and the Origins of Consciousness: From Cartesian Duality to Markovian Monism
"The dynamical architecture of any particle or creature can be expressed as a functional of some generative model that, in some sense, must be isomorphic with the nonequilibrium steady-state density. This tells us that if we interpret the action of a particle or creature in terms of self-evidencing, then the implicit generative model - which supplies the forces that change internal and active (i.e., autonomous) states - must be a sufficiently good model. This is the good regulator theory that emerged in the formulations of self organisation. This also allows us to describe different kinds of particles or creatures quantified by their critical time or temporal depth. For example, if a certain kind of particle (e.g., a trial or protozoan) has a short temporal horizon, it will respond quickly and reflexively to any perturbations – for as long as it exists. Conversely, creatures like us (e.g., politicians and pontiffs) may be characterised by deep generative models that see far into the future (at least to a time in the future when nonequilibrium steady state is restored); enabling a move from homoeostasis to allostasis and, effectively, the capacity to select courses of action that consider long term consequences. Given that the imperative for this action selection is to minimise expected free energy (i.e., expected surprisal or uncertainty), we now have a plausible description of intentional behaviour. To summarize, this temporal depth, the time it takes for a particle or creature to return to its attracting manifold (i.e., nonequilibrium steady state) from an initial state, may distinguish among different kinds of creatures; depending on how deeply their generative model (entailed by internal states) considers the future. This is functionally equivalent to the temporal depth or extent of policies; namely, courses of action, and internally consistent with the notion of planning as inference.”
Impulsivity is related to the topic of long term planning, temporal depth, and allostasis. Mirza et al. (2018) write (paraphrased): "Impulsivity is often due to steep discounting of the future, which may in turn be due to a lack of planning and imprecise beliefs about the world that can impair one's ability to maintain and process information when planning future actions. It has been shown that various disorders of compulsivity (e.g., obsessive compulsive disorder, binge eating, drug addiction) are less “model based” in this task, as are high impulsivity subjects. Our formulation suggests that one possibility why the subjects are less model based: less precise beliefs about the world and lower policy depth." (More elaboration on this within the paper.)
Models, prediction error, and active inference
"Passively performing inferences and updating the generative model is, for many purposes, not enough to accommodate incoming prediction error. As Hohwy notes: “a system without agency cannot minimize surprise but only optimize its models of the world”. The system needs to also perform active inference, in which it not only predicts the changes in its own states, but also engages in prediction about how its interactions with the environment will affect the inputs that it receives. Thus, in addition to inferring how the world is, the system also infers action policies that it expects will be best at minimising prediction error over long time scales. These action policies can then generate real-world behaviours. In other words, the system can actively intervene on its environment in ways that aim to bring the surroundings into a tighter fit with the current model. This offers a very general mechanism for action control, as any behaviour of the system can be reconceived as aiming to minimize prediction error over the long term. The process takes place over multiple levels and time scales, offering a high degree of plasticity (continuous revision) and efficiency (through its anticipatory nature) in real time processing.
Thinking Through Other Minds
A primary distinguishing feature among people are the beliefs under which they are operating, and which guide adaptive behaviour. Provided that we have a model of someone else's beliefs, then we can leverage our own action (policy) selection mechanisms according to the beliefs of another person. For a species such as Homo sapiens that evolved to rely upon cooperative and highly elaborate coordinated action, expectations about folk psychology (inferences about the way other people think and reason and what they expect of the world) are at least as important as, if not more important than, expectations about the physical world itself. In fact, that's the primary domain we leverage to predict and organize behaviour. The integrative model described in this paper has implications that can advance theories of cognition, enculturation, adaptation, and psychopathology. If we understood the full implications, there are many problems that might be resolved by applying it.
"Many adaptive problems are difficult because the environment does not provide clear cues to the best behavior. What is the best design for a bow? What causes malaria? It is not clear what decision rule will be favored by selection when the environmental cue does not allow accurate inference. The free energy principle (FEP) is not normative in the sense that it tells us what all organisms ought to do, or what will necessarily happen to them. Thus, it leaves plenty of space for thinking about things like branching processes, spandrels, maladaptive traits, constraints, trade-offs, mismatches - all the stuff we need to understand the place of adaptive value in the contingent history and development (ontogenetic or evolutionary) of the system of interest. The observation that particular suboptimal or maladaptive norms obtain is not a problem but an observation that forces us to explore the impact of initial and boundary conditions and historical trajectories on the constitution of humans and their cultural niches.
"The evolutionary and developmental acquisition of cultural affordances builds on a set of attentional biases for coalitional intention-tracking, threat-avoidance, and prestige-cued, and social fitness-enhancing information – where the latter maximize an individual’s access to relevant skills, explanatory models, values, moral status, social recognition, and social support. All cultures and cultural subgroups operate with these dynamics. Once a body is equipped with culture-bound skills, and once a world is layered with culture-bound meaning, patterned dynamics of cooperative action and improvisation become possible. Information about threats and group affiliations matter to human minds. Future work will need to distinguish between biases directly geared toward other minds (like attentional preferences for eyes, faces, group affiliation, propositional attitudes), those that harness, enrich, and ‘anthropomorphize’ evolutionarily older, developmentally earlier biases, and those that are not about other minds at all. Thus, we can think of automatic mechanisms that track prestige, social status, and reputation (Henrich and Gil-White 2001) as ‘recycling’ general epistemic foraging mechanisms (found in all living organisms) scaffolded into dominance hierarchies (found in all social mammals). These high-precision cues offer relevant information about which fitness enhancing model to track and learn from. For humans, these dynamics operate both automatically and self-reflectively through symbolic and status cues grounded in normatively configured hierarchies governing optimal moral standing and social functioning. Similarly, we can think of evolutionarily old threat-detection modalities like the negativity bias and pollution avoidance mechanisms as becoming re-encoded through such symbolically enriched processes as superstition, xenophobia, bullying non-conformists, hypochondria, paranoia, magical thinking, conspiracy theories, and the myriad other metaphors and narrative models that postulate the existence of ‘dark’ ‘malefic’ forces and agents as the ‘cause’ of negative internal states and social problems (Boyer 2018).
"New and competing normativities are also competitions for generative models of the world. Social learning as fundamentally affective, and culture as deeply axiological - less about “what is”, as Clement and Dukes put it, and more about “what matters and what is meaningful. Conflict over meaning and how the world ought to be invariably occurs within and between groups. We understand social change as occurring under these dynamics of optimizing world models through intra-group and inter-group competition. We think that the FEP offers a way to account for both social stabilization and social change, and indeed for human historicity itself. For example, the emerging problem of ‘smartphone’ or ‘screen’ addiction may reflect new dynamics of information foraging. We have argued elsewhere that a hyper-abundance of informational uncertainty online solicits the hyper-activation of evolutionarily old attentional biases for social and group-fitness-enhancing information. In turn, this affords an addictive relationship with screens through a constant search for social rewards, social comparison, and high-precision cultural information. We have termed these dynamics the “hypernatural monitoring hypothesis”. The human minds’ limitations on processing vast amounts of information online have recently been described as ’bottlenecking’ mechanisms that favour belief-consistent, negatively valenced, predictive information of a social nature. These recent mechanisms of digital niche construction have been proposed as candidate explanations for the rise of new social challenges, such as increasing extremism, political polarization, and the proliferation of misinformation."
"From the point of view of the FEP, developmental niche construction can be viewed as the process whereby agents make their niche conform to their expectations. Developmental niches are the set of physical and behavioural resources necessary for an adaptive life cycle. Because actions change the physical architecture (and epistemic affordance) of the environment, they tend to make the niche a good ‘mirror’ of the agent’s epistemic foraging, functional anatomy, and, ultimately, brain-based expectations. In short, if we all act successfully to minimise uncertainty our niche will become inherently more predictable –if, and only if, epistemic affordances become encultured. One can think of niche construction as the process whereby the agent’s action creates a symmetry between internal and external states. The agent changes the structure of the world as it acts on the world. Over time, traces in the world will effectively ‘learn’ agents’ beliefs, in the sense that those traces will encode regularities that relate to those beliefs. This means that changes in the niche mirror changes in agents' beliefs enacted via action." This recalls a line of thought developed by David Grinspoon in his book Earth in Human Hands where he speculates on the possibility of creating "Terra Sapiens" and the beginning of a Sapiezoic aeon. If he's right, then that future depends in no small part on whether our models of the world provide the epistemic affordances needed to minimise uncertainty. If they don't (and they arguably do not today), then we must find ways to enculture such affordances within our existing models through processes of modification and selection.
The existential boundary
"From the single cell, to organs, to individuals, and all the way out to coupled organism-environment systems, all of these can be cast as having their own individual identity. The organism and niche are coupled to one another through active inference. [This is important. The human system has become decoupled from the greater system it lives within: the environment. We need models that couple the systems and allow us to recognize and live within the limits of each.] In a nutshell, at any scale, systems are composed of subsystems that, in virtue of their (relative) conditional independence, can also be described as systems. However, all these nested subsystems are integrated within the same system. They are individuated by their own Markov blanket, but at the same time integrated as one single dynamical system through the system dynamics (i.e., adaptive action). Collectively, there is only one (hierarchical) generative model and therefore one free energy functional, for the ensemble of nested systems (where each constituent has a generative model).
"Crucially, by acting to reduce variational free energy, biological systems come to instantiate a probabilistic (generative) model of their environment, including the states of their body. This generative model can be viewed as a ‘map’ of the relational or causal structure among the various quantities (e.g., sensory observations and beliefs) that are optimized through action, perception, and learning, as the organism navigates, and maintains itself in its environment. Hence, it is said that the generative model is ‘entailed’ by the existence of an organism, in the sense that it changes as a function of the organism’s normal bioregulatory activity. Heuristically, this means that through adaptive action, organisms come to embody a guess about the causes of their sensations (i.e., a generative model) by optimizing their beliefs about those causes.
"This generative model, as we have seen, functions as a control system. That is, its function for the cognitive system is to generate of adaptive patterns of behaviour. In the parlance of the FEP, its purpose is to guide the evaluation and selection of relevant action policies. The generative model is a strange beast in the variational framework, in that it exists only insofar as it underwrites the organism’s inference about states of affairs and subsequent action selection. Since the free energy expected following an action, which determines the policy to be selected, is defined in terms of the generative model, the latter is the cornerstone of the self-evidencing process.
"It is a normative model, in the sense that it specifies the conditions that allow the continued existence of the type of creature being considered. The variational story is one about how the respective structures of the generative model and generative process (the actual causal structure that generated observations) become attuned to one another. So, when everything is going well (i.e., when the organism engages in adaptive behaviour and thrives in its niche), the correlational structure carried by the generative model (ideally) maps onto the causal structure of the generative process in the environment. So, while the model is necessarily only ever probabilistic, it remains that active inference fits or tunes the generative model to the generative process; and by that fact, the generative model gains some causal purchase: indeed, the generative model is often described as a probabilistic description of how sensory consequences are generated from their causes. Inference then corresponds to the inversion of this mapping—to infer causes from consequences. This is inference is, by construction, implicit in the minimisation of free energy or the maximisation of model evidence. This reading of active inference as self-evidencing makes the boundary of cognitive systems an existential notion, tied up with the epistemic process of generating evidence for your own existence. In a nutshell, then, to enact a generative model is to provide evidence (i.e., to generate evidence through adaptive action) for a model of one’s existence.
"To exist as a living being and to engage in adaptive action (when all goes well) just is to realise the relations between quantities that are modeled in the generative model. In other words, under the FEP, to exist at all means to produce evidence for a model of oneself (or more exactly, since the generative model is a control system, a model of oneself acting in the world). Existence in this sense is fundamentally tied up with the creation and maintenance of an informational boundary, i.e. the Markov blanket. The key is to note that organisms are organized such that they instantiate the belief that their actions will minimise free energy. The mechanics of belief is the only causally relevant aspect of the variational free energy. The free energy may or may not exist; what is at stake is the causal consequences of the action-guiding beliefs of organisms and groups of organisms, which are harnessed and finessed in the generative model. What matters is that organisms are organized such that they instantiate such a belief to guide their action."
If our current dysfunctional model is central to our crisis, then any future hope we might have for changing this situation rests on the possibility of highlighting the obstacles we face and what they can tell us about whether and how they might be resolved. We do need to view the cosmos as sharing in the vital/generative qualities of life (as the source of these), and we must confront our contemporary tendency to reduce ethical decision making to equations of power and conflict. While possible in theory, in practice these changes would require fairly radical cultural shifts, or complete reversals! But what choice do we have? As the above paper says, we must "fit or tune the generative model to the generative process". An apt description.
Extended Consciousness and Predictive Processing: A Third-Wave View [website, book]
"A generative model is a probabilistic structure that generates predictions about the causes of sensory stimuli. We argue the ongoing tuning and maintenance of the generative model by active inference entails the dynamic entanglement of the agent and environment. Active inference is all about the minimisation of expected uncertainty. Under active inference, learning action policies takes the form of working to optimise or improve the distribution of policies. Epistemic actions are model-based in the sense that they involve search and exploration over a large number of possible action-policies. In this sense, model-based strategies speak to how novelty enters the picture in the active inference framework."
Free energy principle
During the 1990s researchers such as Geoffrey Hinton and Karl Friston began examining the concept of free energy as a measure of the discrepancy between actual features of the world and representations of those features captured by neural network models. A synthesis has been attempted by Karl Friston, in which the brain emerges from a general principle of free energy minimisation. In this framework, both action and perception are seen as a consequence of suppressing free-energy, leading to perceptual and active inference and a more embodied (enactive) view of the brain. Using variational Bayesian methods, it can be shown how internal models of the world are updated by sensory information to minimize free energy or the discrepancy between sensory input and predictions of that input. As Friston recounts: "An important contribution [from Geoffrey Hinton] was to cast the problem of Bayesian inference in terms of optimization. The insight here was that the same problems that Richard Feynman had solved in statistical physics could be applied to the problem of Bayesian inference, namely, how to evaluate the evidence for a model. This is where free energy minimization comes in, the sense that minimizing free energy is equivalent to (approximately) maximizing the evidence for a model. I wrote to [Hinton], trying to summarize my thoughts on biological minimization of free energy. The notes I sent him were eventually published as the free energy principle about 4 years later (Friston, 2005)."
Dispelling Obsession
Xunzi, a naturalistic Confucian, wrote "When one makes distinctions among the myriad beings of creation, these distinctions each become potential sources of obsession." Or in other words, when one is in accordance with the Way they are able to treat the world holistically, while one who is outside of the Way can only see the world as a collection of unrelated units. With a holistic view, learning can be done, and should be done to the point of sufficiency. ...According to most formulations of active inference, the point of generating a mental model is to enable effective learning about future environmental conditions. Therefore, Xunzi would likely conclude, and we must as well, that a model which is not holistic cannot enable effective learning. He wrote: "He who is well versed in the facts alone will treat each fact as a fact and no more. He who is well versed in the Way will unify his treatment of the facts... since he examines and compares the facts, his perception will be clear. With thinking that is based upon a correct approach and action that is based upon clear perception [cf. action-perception cycle], he is able to control all things. ...[But] anyone who would base his judgments upon evidence produced by impaired faculties would be the biggest fool in the world. Such a fool in his judgments uses what is already doubtful to try to settle further doubts, and hence his judgments are never accurate. And if his judgements are not accurate, how can he hope to escape error?"
Models and loving kindness
"Contemplative practices of loving kindness meditation entail the explicit enrichment and effortful rehearsal of one’s mental models of others, which eventually become automatic through practice. The linguistic (narrative) elaboration of these models may be essential to their extension to include members of out-groups, the whole of humanity, or even to all sentient beings."
Keeping up with the Joneses, and social interactions
Social status once depended on one's family name; however, the rise of consumerism in the United States gave rise to social mobility. With the increasing availability of goods, people became more inclined to define themselves by what they possessed and the quest for higher status accelerated. Conspicuous consumption and materialism have been an insatiable juggernaut ever since. According to Roger Mason, "the demand for status goods [Veblen goods], fueled by conspicuous consumption, has diverted many resources". And what is it exactly that such people want? The model evidence for the prestigious model they seek to embody - the new car, fancier home, lavish vacations, better work ethic, better things, more money... magnanimity. Within a consumer society based on nonrenewable resources, is the sort of dynamic of "keeping up with the Joneses" sustainable? Can we change the signals, the sort of model evidence used in the interminable game of one-upmanship? Or should we change the models based on prestige? Or both? And if so, how? Regardless of the physical costs, it's not an easy game to play (nor an easy one to change). In the current age of fast and easy social comparisons, it's quickly obvious how inadequate many of us can be made to feel. Social media notoriously has this effect, fueled by greater connectivity than was ever possible before. How can any of us ever feel adequate? A few objective measures might be more useful than the jockeying for relative social position we have today. But in the meantime, the contest continues.
When it comes to social interactions however, sometimes people are just looking for a certain kind of response, a certain model. And either you can be that model for them or not. Does a friend just want support, confirmation, and consolation, or do they want honest critical advice? You could either infer that from context, or you could simply ask. In a marriage a spouse is looking for a certain model, someone they can be more intimate with than anyone else. (If they have children together, the genetic recombination that occurs at conception is essentially a merger of two models into one.) So understanding what sort of model another person wants or needs is a very useful skill to have, part of the “thinking through other minds” framework. Once this is understood however, the challenge of becoming the model another person needs can be daunting, and raises questions about our willingness to make the commitment of time and energy required, which for some people may represent an opportunity cost. But models are in constant dynamic change from moment to moment. Do we “tune” ourselves to each other, or tune each other out? Like a “matchmaking service“ website, can these models be made explicit? The potential to reflect upon and change our models, that is our identities, is among our greatest freedoms. (The process of "iterative" or "convergent" evolution suggests that successful models are in some sense immortal or perennial.) We all need models that fit the context. As the Tao Te Ching says:
"To hold and fill a cup to overflowing Is not as good as to stop in time.Carlos Perez
Sharpen a sword edge to its very sharpest, And the (edge) will not last long.
When gold and jade fill your hall, You will not be able to keep them.
To be proud with honour and wealth Is to cause one's own downfall.
withdraw as soon as your work is done. Such is Heaven's Way."
Carlos Perez wrote: "We model so that we can gain access to a smorgasbord of advanced thinking tools. Humans are able to improve on their models because they can introspect the models that they create. (Currently, AI is unable to create explicit models and do the same. Counterfactual reasoning is required to make the leap.) Public health epidemiologist Joshua Epstein also has an essay about why we model."
John Campbell
John Campbell is a fascinating writer. In Universal Darwinism As a Process of Bayesian Inference he writes (paraphrased): "There is no information unless there is something that can be informed. This something is a model. The evolutionary interaction between models and the systems they model are applicable to all natural phenomena. We have to abandon the notion that only humans can make inferences, as self-organization can be seen at multiple levels. In this way, freeing oneself from the tyranny of anthropomorphism leads us to a universal Darwinism." In Bayesian inference and the world mind he concludes "these theories describe information processing systems designed to increase knowledge as evidence for the existence of a world mind, and that a fundamental aspect of the universe is its evolving ability to know itself." This has to be among the more esoteric descriptions I've read, also notable for pulling together Peircean semiotics and evolutionary theory. But essentially similar views are also endorsed by Friston. Where do we go with this teleonomic concept of universal model selectionism? Tread cautiously.
Education or propaganda?
The difference between a documentary film and propaganda film could be described in terms of how they develop and use models to make inferences about the film's subject. A recent example is Planet of the Humans, which provides the audience with just enough insight to realize that the true costs and benefits of energy are not easily accounted, but it fails to follow through on the work needed to do that accounting or make the comparisons that would enable viewers to come away with a truly informed opinion. It has a good setup, but the follow through is too often simply missing. The narrative doesn't go beyond a blanket "good vs. evil" approach. This makes it both deceptive and psychologically attractive, and also gets us into trouble very quickly. All forms of energy can be utilized in good and bad ways (yes, even fossil fuels). But only by properly contextualizing how energy sources are collected and used can we make informed ethical distinctions among them. And due to the highly diverse ways in which renewable energy can be harnessed, it is not always easy to determine which is which. This is why a film that elides these distinctions and takes advantage of a general lack of understanding among the public regarding these details can cause a lot damage to the environmental movement as a whole. While there certainly are both good and bad contexts in which to collect/use RE, the film strongly implies they are simply all bad (leaving fossil fuels, by omission, unscathed).
Just as it is true that not all RE is good, it is also true that not all of what this producer has released is bad. Critiquing a film is not about "getting even" with the producer, as some have suggested, it is about the importance of providing people with an accurate model to help them make informed personal choices and policy decisions at a time when we need to be able to do this more than ever. What makes POTH a disservice to this effort is how it omits contextual information and reduces comparisons between the social and environmental costs of energy sources to an oversimplified binary, and inexcusably provides incorrect information. In contrast, when we do include the right information and context, as for example many advocates (like Solar Done Right) have done, then it becomes clear that there are better solutions to meeting energy needs than either existing FF or the Ivanpah solar power facility. We need to shift the level of analysis beyond the "good vs. evil" (or even "FF vs. RE") sort of framing that POTH relies on and exploits and instead model the social and environmental costs in a way that is both informed by values and accurately contextualized and compared against a range of alternatives. It is our ability to construct, share, and compare these sort of detailed models, which are continually being tuned, that allows us to effectively plan and act in an uncertain future. That's also the skill needed to resist the spread of disinformation.
The film has been surprisingly effective at manipulating many people, particularly those for whom the hard work of accurate model building is not an attractive proposition. It basically provides the sort of selective and inaccurate evidence that would confirm a "doomer" narrative. Now, that's actually an attractive social model for some people, in a way it liberates us from certain obligations we might otherwise have. But I doubt that many who are drawn to this model have fully explored its implications. So we have two main groups the film resonates with most: on one side is the extraction industry and political/ ideological groups who are sympathetic to their position, and on the other side are the "doomers" who'd rather see it all collapse, believing this is necessary for "rebirth". I won't say that this is just a perennial sort of mythology that is being reinterpreted on the modern stage. But I will say that doomerism is just another model, and I think it betrays a dangerous lack of self-reflection among those who profess to believe it.
Preparing for the unknown
I am in my 40s and my father in his 70s, and while each of us faces different uncertainties as we look toward the future, a heating planet is one uncertainty that scares us both. Today he sent me an article titled Will three billion people really live in temperatures as hot as the Sahara by 2070? along with the note “Your kids are in for a challenging life in a changing world.” I replied “Yes they are. Most of your working life was during the time before the Internet. My working life is during the Internet age. The kids are going to be working in an age of even more advanced technology and environmental threats. It's going to take a lot of effort to prepare them for all this.” In truth, I have barely prepared myself, let alone the next generation. To prepare for a crisis we can see coming requires a deep and fairly precise model, but to prepare for the unknown requires understanding how models themselves can shift, change, and adapt. Who teaches our children these things? While Thomas à Kempis didn’t rock the boat when he wrote “The Imitation of Christ”, which was essentially a distillation of the Christian model for life, Baltasar Gracián encountered more resistance with his critical scholarship (he is best known for “Practical Wisdom for Perilous Times”). These authors lived hundreds of years ago. I think I need to add another chapter, or maybe just a few short and sweet verses. If I can really distill a perspective, such as that of Karl Friston, into a few verses, then from those few verses one has a POV from which to see all of evolutionary history. It is "perspective" that is the hardest thing to find in life, and the most precious gift we can give the next generation. A trustworthy perspective allows us to get through the day to day challenges, but an exceptional one might even help to prepare us for the unknown.
Indigenous Pathways
In the euro-western mindset, and still in contemporary culture today, mind and matter are separate, and mind is seen to dominate over matter. "God gave it all to us." From this point of view, man is the measure of all things, including what is right and what is wrong. The problem is that if I decide what is right, then I will always decide that I am right. But if Nature determines what is right and wrong, then I will find I am not always right after all. Nature tells me I am composed of both a body and mind that are inseparable from each other and the rest of the world. When I review my actions by the light of Nature, I discover that I have failed to fulfill its covenant. In the end, Nature decides whether I live or die. So how do we move from an anthropocentric to an ecocentric perspective (which as I understand it is also an Indigenous perspective)? This is a difficult question, and also one with many blind alleys leading off of a central pathway. Those who desire change can be of two kinds, we could call these "chaotic" and "planned". Simply speaking, all people could be seen to fall somewhere on a spectrum between these extremes. If we can see the problems in the world today, but we do not understand where they came from nor how to fix them, yet we agitate for change, then the chances that the resulting change will improve our condition are marginal at best. Chaos usually breeds more chaos. But if we see the problems in the world, and we have some understanding of their source and how they can be avoided, and furthermore if this understanding can withstand a fair interrogation among peers, then the chance of success rises. How do we know this? We have seen that the Medicine People in any society were revered because of their ability to ascertain the causes of maladies and plan an appropriate treatment. This is the same principle. Many people have ideas about how to change cultural thinking processes and usher in a new ecocentric perspective, a rebirth. But what we need today are people who can articulate, interrogate, and implement the best of these.
Today I am very concerned. There are many populist movements agitating for change that, whether intentionally or not, tend more toward "chaotic change" rather than "planned change". Here are a few examples. A recent headline read "A Shocking Number of Americans Want to 'Just Let Them All Burn': A study measured the 'Need for Chaos' that leads people to spread online misinformation and vote for politicians like Trump." Some years ago Richard Slotkin wrote a book titled "Regeneration though violence", which documented the history behind this idea. It has not gone away. Less shocking, yet also concerning to me is when I hear people say "Let it collapse, build a smaller simpler society from the ground up. We need to relocalize and regenerate and that doesn't mean working within the system. Our only chance is to rebuild at the margins and have something there to rise from the ashes." Let me be clear: I agree with the goal of localism and resilience. But where is the understanding of why the system wasn't working to begin with? What would prevent the same problems from emerging again once it rises anew? As a result, I'd place this more toward a vision of chaotic rather than planned change. It must be developed. So when I hear there are people working on "planning as ceremony" it makes me glad. Transitioning from a mentality of "expand, conquer, consume" to a sort of "contract, cooperate, cultivate" story of living in harmony with the Earth cannot be done without this sort of deep reflection. Implicit within the idea of planning is having a model to guide the entire process along, and we must understand how models can and do shift, change, and adapt. Is a paradigm shift possible simply by looking more closely at the euro-western anthropocentric model and comparing it to an Indigenous ecocentric model? We see this in the younger generations who are increasingly questioning the racism and classism of older generations. The future is open-ended. Although there is no guarantee that a full shift will occur, or if it does that it will even be in the direction we need it to go, there is also nothing that precludes this possibility. When faced with tough questions it is best to ask "What would Nature do?" Nature created and continually trims the tree of life through natural (model) selection.
A "doomer", as I think of the term, is just one extreme one a continuum or spectrum of views, a person who is generally fatalistic in viewpoint and doesn't believe human agency is capable of producing a beneficial outcome on the global scene. In extreme form, a doomer tends to suspect human understanding and/or agency is futile/dysfunctional/irredeemably maladaptive and therefore seeks the collapse of social systems (generally indiscriminately). A surprising number of people really do want chaos as a means toward the dissolution of contemporary systems of governance. It's easy to sympathize with that given the levels of corruption, exploitation, and extraction/destruction of earth systems. All fair points and easily understood. This perspective is also seductive in part because we are psychologically predisposed to take notice of our own failings (cf. "surprisal"). I think if we follow this line of thinking further it begins to reveal its own weaknesses, which follow from what I would characterize as inherently distorted viewpoints on both our capacity for understanding and the role of agency (cf. action-perception cycle). If either understanding or agency are eliminated then the doomer is right. Regarding understanding: we cannot indiscriminately dissolve social systems through a chaotic collapse scenario. All progressive organizations understand this. Nonetheless, some people seem to prefer the "let it all burn" idea, which is propaganda that plays upon the death/rebirth imagery taken to irrational extremes. At large scales there simply is no rebirth after death, a point recently made by James Lovelock in interviews following the publication of his book Novacene. There is only extinction. Regarding agency: I think anyone who is engaged in organizations like CCL understands the importance of assessment, radical honesty, and developing responsible action models. We all exercise, and moreover, seek to highlight our personal and collective agency. So none of us would meet this extreme definition of a doomer whose arguments rest on the disposal of understanding and agency. But we must be aware of these arguments nonetheless.
The amount of information available today, and the degree of abstraction that characterizes that information, has increased dramatically. Is it easier to sew division and occlude holism when there are more informational layers vying for your attention? Probably. If we remove all our accumulated information and mental models, then our sensory experience would simply come in like an undifferentiated flood all at once. Research with psychedelics, like Terence McKenna and others can attest to, appears to confirm this much. So in that sense, a truly holistic perspective might be our earliest and most natural living experience. Modernity has overlaid upon this natural unified experience a bunch of information and models that can either reinforce our native holism or fracture and break it apart. The challenge then is to understand which processes and models promote unity and which agitate and divide.
Milan Ćirković: "In 1543, two revolutionary books transformed our view of both the universe and ourselves. One, written by Flemish physician Andreas Vesalius, was titled De humani corporis fabrica (On the Fabric of the Human Body), and it laid the foundations of modern medical science by proving our bodies are not very different from the bodies of animals. The other, entitled De revolutionibus orbium coelestium (On the Revolutions of the Heavenly Spheres) was by Nicolaus Copernicus. After Copernicus came Darwin’s revolution, and then Freud’s, delivering blows to our illusions of uniqueness and grandiosity within the biological and mental domains, respectively. But our institutions are still profoundly anthropocentric. We deny even the most basic rights to other parts of nature, including our close animal relatives, some of which share more than 97 percent of our DNA. We pollute our environment with close to zero regard for the well-being of its ecosystems - and we fight pollution only if and when it inconveniences us."
Sir Fred Hoyle put it nicely in 1983: "Many are the places in the Universe where life exists in its simplest microbial forms, but few support complex multicellular organisms; and of those that do, still fewer have forms that approach the intellectual stature of man; and of those that do, still fewer again avoid the capacity for self-destruction which their intellectual abilities confer on them. Just as the Earth was at a transition point 570 million years ago, so it is today. The spectre of our self-destruction is not remote or visionary. It is ever-present with hands already upon the trigger, every moment of the day. The issue will not go away, and it will not lie around forever, one way or another it will be resolved, almost certainly within a single human lifetime."
Fristonian model selectionism (cf. Darwinian selection, comparative approach)
If this line of thought can be developed further, then a white paper submitted to a pre-print or self publishing independent research website like Academia.edu or Arxiv.org or hcommons.org may be the next step. A white paper approach of applied theory for analyzing current policy development in Alaska and comparing it to policy in related provinces or nations would be interesting. The angle I'd suggest developing is "model selectionism", which brings together the social sciences and physical science in surprising ways and makes some very big claims. Although heavy on the theory, it may fruitfully examine several test cases of effective social and public movements and policy proposals.
What are our future prospects on Earth? It's a wicked problem and the only thing that can be said is the available evidence is highly contingent and inconclusive. The arguments lead everywhere. If we don't change for the better we definitely will do ourselves in. But change is inevitable, and we can guide the direction of it. "Theory of change" is a huge topic in activist circles. Probably the last big challenge will be the same as the first big challenge was thousands of years ago during the first Agricultural Revolution. For 95% of our species’ history, we didn’t farm, create large settlements or complex political hierarchies. We lived in small, nomadic bands, hunting and gathering. We loved it. We probably would have continued doing this forever if we could have. We didn't transition from hunter-gatherer life to plant harvesting, then cultivation and, finally, cities on purpose, with foresight and intent. According to Nick Longrich, it was an act of desperation following an ecological catastrophe. At that time we also saw a growth in the unjust exercise of power in the political economy and the tendency for governing systems to circumvent the best interests of the community. As Peter Corning earlier asked: Can we solve the self-governance problem? We haven't yet, and (here's my thesis) we will not, that is, until we understand how psychopathological governance and aberrant modeling processes are interrelated. Without understanding this connection, the best we can do is treat the symptoms and offer temporary relief (which mercifully, can buy much needed time). It is clearly demonstrable that people would rather believe a lie that they can understand, and that comports with their internal mental models, than act upon a fact that makes no sense to them. For this reason a comprehensive project that focuses on the intersection between modeling processes and contemporary issues should be a priority.
This is, in a nutshell, the application that I believe should be further developed from a "model selectionism" perspective, as developed by Karl Friston and others involved in this research. It seems to be the big untold story that follows from the FEP. The TTOM paper suggests "the FEP offers a way to account for... human historicity itself" and John Campbell has certainly explored some implications in his universal Darwinism paper. To name a few more interesting aspects of Fristonian model selectionism, it provides a framework for accommodating a wide range of observable phenomena and the theoretical work done to explain them, including "signalling selection" as described by Amotz Zahavi (elucidating the mystery of altruism) which "brought animal communication, game theory, and sexual selection together" (Essays in Animal Behaviour, 2005, p44), and "biosemiotic webs" rooted in Peircean semiotics, John Deely's "physiosemiotics", and today incorporating aspects of Deaconian teleodynamics. A “semiotic web” or semiosphere of relationships, while not as carefully defined as a “generative model”, which in the glossary of the TTOM paper appears as "a mapping from beliefs about hidden causes to observed consequences", does otherwise feature many conceptual similarities. The key difference between Fristonian models and Sebeok's semiotic web (Hoffmeyer's semiosphere) is that the former has the ability to characterize the "mismatch between organism [model] and environment" via the FEP. Again, this provides a useful explanation for many observed phenomena, such as anticipation, prediction, and various psychological disorders, that are difficult to account for otherwise, and utilizes an established set of theoretical concepts from other fields. Friston's conceptual approach may, in so many words, vindicate what earlier theorists like Sebeok, Hoffmeyer, and others had anticipated. If this can be established, then these separate theorists are mutually supporting, though reflect different developmental stages of thought.
I've described the social science application, but the physical science basis is just as if not more interesting, as it goes back to Erwin Schrödinger's question "What is Life?", in other words how does order arise from disorder and counteract the effects of entropy? So with a single theoretical foundation, we can provide an account for life itself, diagnose contemporary social and environmental problems as a dysfunctional development (or maybe a transitional phase) of these basic processes, and indicate possible treatments or approaches that would meet with greater chances of success. So when we ask "What would Nature do?" as a political question, we really can place policy and political analysis on a thoroughly naturalistic foundation. This hasn't always been possible to the degree it is today. ...At this point usually a large number of skeptical questions come to mind (as they should), but from what I've seen there's a lot of promise here at multiple levels (spatial and temporal) for developing our "theories of change" and addressing the biggest challenges we face, which have already been well defined by others. So I place my hope in work like this. It might be misplaced after all, but I'm very Sisyphean in outlook.
A Fristonian theory of cultural evolution as model selection could provide a means for recognizing the environmental constraints that contemporary cultural evolution ignores. We exceed planetary boundaries today because this culture uses models that rely on decoupling the human system from the environment. The solution is to find models for how they can work together, and how to make a successful transition to these. The ability to understand the dynamics of change and develop our potential is our greatest freedom. By exercising it our models evolve like biological species. Since how the parts of a system coordinate with each other is determined by the models it uses, when those models change the system parts may be reappropriated ("exaptation" in evolutionary theory) to serve new functions. Now, biologists have constructed phylogenetic trees on the basis of various differences in morphological characteristics. Evolutionary theory describes how these can change over time to produce the diversity we see today. Examples include the evolution of the jaw from the gill arch, legs from fins, and wings from legs. Ethologists have attempted phylogenies based on behavioral characteristics, which would look quite different. We could likewise construct "model phylogenies", where the traits of a model, such as the degree to which it is attuned to the environment, or the ways in which it participates in a larger ecology of mind, become the specific characters (in a taxonomic sense) that are exposed to selective pressures. It would be similar to the process already used in "scenarios analysis" where a variety of counterfactual situations can be explored for their potential, however here we would incorporate a Fristonian approach. As a result we could be in a better position to do the hard work of comparative model and policy analysis. Viewed in context, how do the models used by an individual, society, or superorganism further shape it's development and evolution? Do they help the whole system tend toward grip on its ecological niche, or has the system lost its grip? Where are the branching points that shift future trajectories? The resulting comparison, analysis, and selection or modification could guide an ecological process for future planning on any scale, from hours to aeons.
Phylogenetic analyses
In "Innovation in the collective brain" Muthukrishna and Henrich write: "Cultural models are chosen based on cues linked to success or skill. There is a relationship between the number of models individuals have access to and the average cultural complexity that a population can maintain. With increases in population size and interconnectivity, thanks to literacy, radio, television and most recently, the Internet, we are now experiencing an unprecedented rate of innovation. Within social networks individuals select cultural models, and through these links innovations are transmitted. Access to multiple models allow individuals to selectively learn from the most successful, but also recombine knowledge from the next most successful models, leading to better outcomes than those who did not have access to many models. Individuals with multicultural experiences are better able to connect seemingly disconnected concepts.
"A common perception of the source of innovation is Thomas Carlyle's ‘great man’ - the thinker, the genius, the great inventor - whose cognitive abilities so far exceed the rest of the population, they take us to new places through singular, Herculean mental effort. The prevalence of this belief in a Great Man or wise ancestor is a recurring theme. However, revolutionary innovations often rely on luck rather than on systematic and fully intentional investigation. There are many examples where ‘new’ inventions are more clearly the product of incremental improvements, recombinations of existing elements and selection. The ‘inventor’ is in most cases just the popularizer. Several lines of evidence point to the crucial role of recombination, incremental improvement and selection in innovation. One method that has proved useful has been the application of phylogenetic analyses to the constituent elements of a technology. These analyses, which have been applied to both portable radios and bicycles, clearly reveal how the constituent components in a diversity of designs have recombined into the products we see today."
"A common perception of the source of innovation is Thomas Carlyle's ‘great man’ - the thinker, the genius, the great inventor - whose cognitive abilities so far exceed the rest of the population, they take us to new places through singular, Herculean mental effort. The prevalence of this belief in a Great Man or wise ancestor is a recurring theme. However, revolutionary innovations often rely on luck rather than on systematic and fully intentional investigation. There are many examples where ‘new’ inventions are more clearly the product of incremental improvements, recombinations of existing elements and selection. The ‘inventor’ is in most cases just the popularizer. Several lines of evidence point to the crucial role of recombination, incremental improvement and selection in innovation. One method that has proved useful has been the application of phylogenetic analyses to the constituent elements of a technology. These analyses, which have been applied to both portable radios and bicycles, clearly reveal how the constituent components in a diversity of designs have recombined into the products we see today."
In semiotics, Peirce is the figure to pay attention to (Sebeok and Hoffmeyer derived most of their ideas from him.) Essentially, Peircean biosemiotic theories describe how significance emerges from the world as a natural process. We have not just a biosphere or "sphere of life", but also a world of meaningful interrelationships, a "semiosphere". Semiotics is about 'probabilistic thinking' and how we, as motivated beings, can accurately discern the possible scenarios for flourishing and avoid degenerative dynamics. It is the process of forming relationships, making interpretations, and guessing about possible futures laden with existential significance to beings such as ourselves.
So it should not be too surprising that attempts have been made to use this understanding to help guide policy decisions, and so the related subfields of "ecosemiotics" and "semioethics" have been proposed. In his remarks at the 2018 Biosemiotics Gathering, Jesper Hoffmeyer expressed the hope that "the biosemiotic scheme... might help us to develop a better environmental policy". Eduardo Kohn provided one of the most interesting attempts. He described the motivation behind his book "How Forests Think" in an interview: “The question for me politically is, how are we going to create an ethical practice in the Anthropocene, this time of ours in which futures, of human and nonhuman kinds, are increasingly entangled, and interdependent in their mutual uncertainty? This is where I’m headed. And in the book I begin to think about this political problem. How does that articulate with what’s happening on the ground in terms of environmental politics? Who might be doing something like this? I don’t know. It’s very abstract right now, but that’s where the political part of this would go.”
My guess is that if he were to follow this line of thinking further, then Kohn and the biosemiotic community at large would benefit from the productive research surrounding Karl Friston. (In fact, the "extended evolutionary synthesis" itself would probably benefit from a Fristonian approach.) Common to all these areas is selectionism. In biosemiotics, accurate semiotic relationships are selectively favored over more fallible or inaccurate ones. Likewise for Friston, "generative models" that accurately map "beliefs about causes" on to "observed consequences" are selectively favored over inaccurate models. (Incidentally, this suggests a homology with the Peircean triad of "sign, object, and interpretant".) What I'd suggest is that biosemiotics, and thus Sebeok and Hoffmeyer, and evolutionary theory itself lacks a unified means for characterizing these selective forces. How is accuracy determined? How do we describe fallibility or a mismatch between organism and environment? In what terms? Here Friston provides a very useful tool: the "free energy principle". Along with several other novel applications from related fields, the Fristonian approach to selectionism has the potential to be a universal solvent for Kohn's political problem. It is normative by means of overcoming the mind/body duality, and comparative by means of providing evaluative methods for analyzing policies using the free energy principle applied to modeling processes. It would, in short, provide Kohn with the tools and insight needed to select from among a variety of policies those most likely to "create an ethical practice in the Anthropocene". Big claim? Yes. (More modestly, my sense is that there are many in the biosemiotic community who are looking forward to a new synthesis with academic research in other fields to broaden its utility, and I think this might help provide that.)
Kohn's book was excellent and should be mind-blowing for everyone. The Fristonian approach can take it to the next level and really make it far more applicable. As an anthropologist, Kohn was the right person to bring those ideas to a wider audience. Another great anthropologist is Joseph Henrich, and he is cited in some of the most exciting papers tracing out the implications of Friston's research. There's a paper titled "Thinking through Other Minds". Take a moment to let the full import of that title sink in when comparing it to "How Forests Think". What we need is another writer with ecological concerns like Kohn to step forward and develop the promise of "How Forests Think" by incorporating the Fristonian approach to selectionism. That won't be easy, and it will definitely be controversial. In a podcast he was interviewed for last year Kohn said: "Anthropology is a mind-manifesting science. We must learn to come into contact with this other form of thinking and find spaces where one can be connected with the larger ecology of mind through modes that are not restricted. Trump is a master dream worker. He works at that register, and when shamans do their work they are battling at that level with those kinds of people. We are beyond the conversation that spirits are real. What are the ethics of working at that register? ...There are many forms of attunement. When the weather is weird, how does that make you feel? Do you want to just ignore it? Be attuned to those things. We need to cultivate our attunement to the world."
Here Kohn isn't recommending the "turn on, tune in, drop out" counterculture-era advice, but he is pointing out that the Western world is seeing a rise in political figures like Trump in part because demagogues can exploit aspects of the larger ecology of mind (a term from Gregory Bateson) that we haven't been able to fully account for. Friston's research is attempting to do just that: provide an account for how our ecology of mind becomes (or fails to become) attuned to the world. Compare this to a passage from the TTOM paper: "Heuristically, one can think of niche construction as the process whereby the agent’s action creates a symmetry between internal and external states. The agent changes the statistical structure of the world as it acts on the world... Another intriguing consequence of this is that, over time, traces in the world will effectively ‘learn’ agents’ beliefs, in the sense that those traces will encode statistical regularities that relate to those beliefs. This means that changes in the niche mirror changes in agents beliefs enacted via action. ...one can model ‘environmental learning’ about the agents’ action in the same way that one models ‘agent’s learning’ of the environment’s sensory causes... In active inference, everything that can change, changes to minimise the mismatch between organism and environment." It's a massive POV switcheroo.
How to apply models
Lex Fridman recently published an interview with Karl Friston: "You've got to have a model that is conditioned upon different courses of action and always thinking about the future. Planning requires selecting among a number of different courses of action. We've got free will, and the act of selecting amongst different alternative ways forward, this policy or that policy, involves inference and self evidencing. ...Now the free energy principle is tautological. Things that exist minimize their energy. Why do they minimize free energy? Because they exist. But there is a practical thing you can get from it. If it looks as if things that exist are trying to optimize a variational free energy, and a variational free energy has to be a generative model, that is to say a probabilistic description of causes and consequences of the world in which it is immersed, then it should in theory be possible to write down a self-evidencing, functional generative model. So you should be able to create models where you have supplied the objective function, that supplies the self-organizing dynamics to non-equilibrium steady state. This is the practical application of the free energy principle. When you can write down the generative model, you can create the kind of steady state system it describes. The big problem, however, is writing down the generative model. That's where the heavy lifting comes in. It's the form and structure of that generative model which basically defines what you will create. That's where all the hard work comes in. We need to have models that are fit for purpose for the kind of world in which we live."
Here's an analogy to understand "non-equilibrium steady state". The human body tries to keep an internal temperature of approximately 97.7–99.5°F, this is a "steady state". But this is definitely not always in equilibrium with the environmental temperature, which in Alaska can often be anywhere between 40F below zero to 100F, highly variable depending on local conditions. Since our body temperature cannot be in equilibrium with the external environment, we are "non-equilibrium steady state" systems. Applying this to the planetary boundaries concept, since the start of the Holocene epoch the Earth has maintained its climate within fairly narrow limits. Furthermore the average temperature at the surface of Earth is not in equilibrium with the average temperature of outer space. The atmosphere serves as a boundary that separates the two systems. Therefore Gaia herself is a "non-equilibrium steady state" system. Just as we prefer our bodies at 97.7–99.5°F, conditions during the Holocene provided a biosphere with a comfortable average temperature of approximately 58°F.
For Karl Friston, a "generative model" maps beliefs about causes on to observed consequences. That's a simple but fairly complete explanation. From this definition, we can see why accurate models would be favored over those that are less accurate. Another way to put it is that in the ongoing process of adaptation, we are trying to fit or tune the generative model to the generative process. We want the relationships we see in the environment to be fully accounted for within our models. If they are, then they can accurately guide the policies we adopt and live by.
For Karl Friston, a "generative model" maps beliefs about causes on to observed consequences. That's a simple but fairly complete explanation. From this definition, we can see why accurate models would be favored over those that are less accurate. Another way to put it is that in the ongoing process of adaptation, we are trying to fit or tune the generative model to the generative process. We want the relationships we see in the environment to be fully accounted for within our models. If they are, then they can accurately guide the policies we adopt and live by.
The banality of evil
Claire Wardle writes “In February 2017 I created seven types of information disorder in an attempt to emphasize the spectrum of content being used to pollute the information ecosystem. They included, among others, false context, which is when genuine content is shared with false contextual information. The most effective disinformation has always been that which has a kernel of truth to it, and indeed most of the content being disseminated now is not fake—it is misleading. Instead of wholly fabricated stories, influence agents are reframing genuine content and using hyperbolic headlines. The strategy involves connecting genuine content with polarizing topics or people. In these efforts, context, rather than content, is being weaponized. The result is intentional chaos.” When the main challenge today is to understand how our ecology of mind (a term used by both Gregory Bateson and Eduardo Kohn) becomes, or fails to become, attuned to the world, we need to be aware that there are many people working to destroy a more holistic future by distorting contextual information. There may be no more effective way to undermine adaptation.
We are vulnerable to this sort of subversion because even in their absence, at a very fundamental level, our greatest problems aren't caused by any overtly malicious intent (as most people believe), but rather a more benign ineptitude and misapprehension of reality. That is to say, in the vast majority of situations we should be more concerned about the accumulated effects of minor problems than any single large problem. This is often explained through the analogy that more people die of cancer and heart disease than from terrorism and plane crashes. We are more likely to suffer from and die of what in the final analysis is no more than our misdirected attention and misplaced priorities. Society dies a death of a thousand cuts, not from the asteroid impact of apocalyptic Hollywood movies. Relationships often buckle and fall from a hundred small omissions and thoughtless actions, any single one of which in isolation would’ve been excused, but in the aggregate are unforgivable. And how does this happen? We miss the forest for the trees again and again. The forest may be characterized as the holistic modeling processes that guide our goal directed behaviors, the trees are any one of the individual sub processes. But most of the time it isn't even that we become distracted by, or neglect, any single tree. No, we simply fail to recognize how they all come together as one single forest. We literally can't see the forest, and so we wander about lost, periodically shocked out of our stupor to see it dying around the edges from a plague we struggle to comprehend.
We are vulnerable to this sort of subversion because even in their absence, at a very fundamental level, our greatest problems aren't caused by any overtly malicious intent (as most people believe), but rather a more benign ineptitude and misapprehension of reality. That is to say, in the vast majority of situations we should be more concerned about the accumulated effects of minor problems than any single large problem. This is often explained through the analogy that more people die of cancer and heart disease than from terrorism and plane crashes. We are more likely to suffer from and die of what in the final analysis is no more than our misdirected attention and misplaced priorities. Society dies a death of a thousand cuts, not from the asteroid impact of apocalyptic Hollywood movies. Relationships often buckle and fall from a hundred small omissions and thoughtless actions, any single one of which in isolation would’ve been excused, but in the aggregate are unforgivable. And how does this happen? We miss the forest for the trees again and again. The forest may be characterized as the holistic modeling processes that guide our goal directed behaviors, the trees are any one of the individual sub processes. But most of the time it isn't even that we become distracted by, or neglect, any single tree. No, we simply fail to recognize how they all come together as one single forest. We literally can't see the forest, and so we wander about lost, periodically shocked out of our stupor to see it dying around the edges from a plague we struggle to comprehend.
The marketplace of ideas
According to Cailin O’Connor and James Weatherall: "Matters of fact shouldn’t be settled by public vote. They should be settled by gathering evidence and using that evidence to feed into our best tools to figure out what’s true based on it. Of course, we’re always going to have some false beliefs because we’re social learners. It’s easy for false beliefs to propagate from person to person. But that doesn’t mean we’re always going to have the same degree of false belief. If you look at cultural evolution, we developed these cultural systems that help us do better with our brains. We’ve developed amazing learning systems that help little kids learn more effectively than in the past. We also can develop systems that allow us to do the best we can with the brains that we’ve got. It’s not just that we should give up and we’re hopeless.
"There’s been misinformation and propaganda for hundreds of years. If you’re a governing body, you have interests you’re trying to protect. You want to control what people believe. But if someone is monopolizing information flow, and interfering with what kind of information gets out there, that’s going to affect the efficiency of the marketplace of ideas. That’s what influencers, propagandists, and industrial groups are doing. Ideas aren’t spreading properly from one community to another. So we get enclaves. This is what polarization looks like - a failure of reliable beliefs to spread from one community into another community. On an individual level, it’s tricky. People just don’t trust others who have different beliefs from them. Social media has changed the structure of communication between people as well. All social media sites should be employing teams to fight active misinformation and disinformation."
According to Cailin O’Connor and James Weatherall: "Matters of fact shouldn’t be settled by public vote. They should be settled by gathering evidence and using that evidence to feed into our best tools to figure out what’s true based on it. Of course, we’re always going to have some false beliefs because we’re social learners. It’s easy for false beliefs to propagate from person to person. But that doesn’t mean we’re always going to have the same degree of false belief. If you look at cultural evolution, we developed these cultural systems that help us do better with our brains. We’ve developed amazing learning systems that help little kids learn more effectively than in the past. We also can develop systems that allow us to do the best we can with the brains that we’ve got. It’s not just that we should give up and we’re hopeless.
"There’s been misinformation and propaganda for hundreds of years. If you’re a governing body, you have interests you’re trying to protect. You want to control what people believe. But if someone is monopolizing information flow, and interfering with what kind of information gets out there, that’s going to affect the efficiency of the marketplace of ideas. That’s what influencers, propagandists, and industrial groups are doing. Ideas aren’t spreading properly from one community to another. So we get enclaves. This is what polarization looks like - a failure of reliable beliefs to spread from one community into another community. On an individual level, it’s tricky. People just don’t trust others who have different beliefs from them. Social media has changed the structure of communication between people as well. All social media sites should be employing teams to fight active misinformation and disinformation."
Models and equanimity
Democritus (~460-370 B.C.) wrote: "[In order to achieve cheerfulness]... one must keep one's mind on what is attainable, and be content with what one has, paying little heed to things envied and admired, and not dwelling on them in one's mind. Rather must you consider the lives of those in distress, reflecting on their intense sufferings, in order that your own possessions and condition may seem great and enviable, and you may, by ceasing to desire more, cease to suffer in your soul... One must... [compare] one's own life with that of those in worse cases, and must consider oneself fortunate, reflecting on their sufferings, on being so much better off than they. If you keep to this way of thinking, you will live more serenely." (fr. 191, cf. fr. 3; Kirk, Raven and Schofield) This fragment from Democritus exhorts us to always maintain a holistic perspective on suffering and recall that we are but a small, inseparable part of much larger community of life. It is our ignorance of the greater whole that causes our emotional disposition to depart from equanimity. From understanding holism there is empathy, and from empathy there is equanimity. Holism not only binds one person to the next, and one day to the next as well.
Active inference
It's useful to have an evolved inclination to see things in terms of hierarchies. It's a great mental tool generating all sorts of epistemic affordances. And it's how complex organisms coordinate the multiple parts of a physical system (with cells, tissues, organs, etc). Analogously, we also have an evolved attraction for fat and sugar. And similar to the way we use hierarchy, a little bit of fat and sugar is great but too much could kill you. So why is it that we eat too much now? And why have hierarchical structures become problematic today, leading to the appeal of authoritarian systems of governance (which contemporary political conservatives tend to align with)? And can our attempts to explain this help account for similar problems, like xenophobia? In short, the main problem here is that our contemporary global culture has outpaced biological evolution, but has yet to develop a sufficient amount of self awareness and control to avoid triggering positive feedback cycles leading to inhospitable conditions and greater selective pressures.
During the last 10,000 years we have been able to obtain large surpluses of energy and materials. This period began with the first Agricultural Revolution. It wasn't really possible before then. (There are some exceptions. Cultures in the Pacific Northwest achieved complex social structures in an environment that is very biologically productive.) These social conditions are so new, on a geologic time frame, that our current global culture has yet to experience significant selective pressures. There have been some major events, but these were mostly localized. We are still enjoying the stable conditions of the Holocene in which we've developed. Without self awareness or control, and still in the throes of rapid growth, we are enormously profligate and wasteful. But all this is about to change. In life, everything is about healthy relationships, whether that's between society and the environment, or us and our consumption of food. Our use of energy and materials is clearly in disequilibrium with planetary systems. That unhealthy relationship will shock us out of the Holocene and usher in some new, powerful selective pressures.
Ever since the harmful consequences of the Industrial Revolution became apparent, we have predicted such consequences would result, and we have been learning how to prevent these from occurring. Reducing pollution is the most obvious, but the harmful consequences of evolved epistemic affordances like hierarchy and other biologically or culturally evolved concepts, tools, etc, when applied indiscriminately to concentrate wealth, inflate prestige, and form social castes, are far harder to develop a fine grained understanding of, and are difficult to interrupt before they have already caused harm. Freud in his Civilization and Discontents (1930) once warned: "The fateful question for the human species seems to me to be whether and to what extent their cultural development will succeed in mastering the disturbance of their communal life by the human instinct of aggression and self-destruction." Why has cultural development enabled individuals to indiscriminately act on their biological inclinations? (Hopefully some of the preceding discussion has helped to shed light on that.) Can cultural evolution eventually equip the global community with the understanding and self control to mitigate our harmful propensities? I think the answer is yes, and this is where I see work on "active inference" potentially taking us. We can characterize the disequilibrium between systems and find ways to re-establish harmony.
Naturalising phenomenology
"Naturalising phenomenology is a scientific research program that aims to characterise the mind-brain system on its own terms, as it were; the way that it appears to itself for itself, as a subject of experience, rather than only focusing on it as a mere thing. What is at stake here is how best to characterise the relation between first-person data obtained from phenomenological accounts of lived experience to third-person cognitive and neuroscientific accounts. For instance, how might one relate the direct lived experience of watching a beautiful sunset to its physiological manifestation [or vice versa]?
"Active inference is a framework that casts perception, learning, and action as essentially being in the same game: that of gathering evidence for the model that underwrites the existence of the agent. In this sense, active inference casts living and cognitive processes as self-fulfilling prophecies, which gather evidence for an implicit (generative) model that the agent embodies and enacts. Generative models are so called because they are models of the generative process ‘out there in the world’ that cause (or generate) our sensory data. The active inference agents themselves, of course, do not have [direct] access to the generative process, and must deploy inference and action to guess-timate its structure." Selecting which actions to deploy makes active inference, perforce, a game of policy selection. Or put another way, "In a nutshell, active inference says that action and perception are in the service of maximizing not a value function of states, but a functional of beliefs about states (known as variational or expected free energy)."
"Naturalising phenomenology is a scientific research program that aims to characterise the mind-brain system on its own terms, as it were; the way that it appears to itself for itself, as a subject of experience, rather than only focusing on it as a mere thing. What is at stake here is how best to characterise the relation between first-person data obtained from phenomenological accounts of lived experience to third-person cognitive and neuroscientific accounts. For instance, how might one relate the direct lived experience of watching a beautiful sunset to its physiological manifestation [or vice versa]?
"Active inference is a framework that casts perception, learning, and action as essentially being in the same game: that of gathering evidence for the model that underwrites the existence of the agent. In this sense, active inference casts living and cognitive processes as self-fulfilling prophecies, which gather evidence for an implicit (generative) model that the agent embodies and enacts. Generative models are so called because they are models of the generative process ‘out there in the world’ that cause (or generate) our sensory data. The active inference agents themselves, of course, do not have [direct] access to the generative process, and must deploy inference and action to guess-timate its structure." Selecting which actions to deploy makes active inference, perforce, a game of policy selection. Or put another way, "In a nutshell, active inference says that action and perception are in the service of maximizing not a value function of states, but a functional of beliefs about states (known as variational or expected free energy)."
Real-world interventions on maladaptive social norms
As Matteo Colombo writes "There are many social norms that are
maladaptive, harmful, morally abhorrent, or just lack any value or
social function. In order to change social norms, significant public
policy efforts are being made to intervene and target people’s
expectations and conditioned preferences for following those norms.
Let's evaluate the prospects of using Veissière et al.’s account for
real-world interventions on maladaptive social norms." Where a discipline lacks a common theoretic language for describing competing perspectives, bringing rival positions under a single theoretical framework can help to dissolve disputes. The Free Energy Principle (FEP) is one such paradigm for describing the behavior of systems. Van de Cruys & Heylighen provide compelling examples:
"Thinking Through Other Minds (TTOM) and the underlying FEP framework allow us to revise and refine, not only theory of mind accounts, but also theories of the evolution of culture, most notably memetics. First, the analysis of Veissière et al. shows that the concept of memes as independent units of cultural information is rather deceptive. An FEP-based account of culture highlights that ideas or hidden causes are not insular units but parts of hierarchically and laterally-structured belief networks, or narratives. Such networks also include so-called auxiliary hypotheses that can take the blame when other, high precision beliefs are under threat of being disproven. The networks even include socially-expected ways of sampling evidence (expected precision or epistemic value of different sources). These observations suggest that cultural beliefs are socially constructed and self-sustaining. Thus, fabricated beliefs such as conspiracy theories can spread easily and take root.
"Second, Veissière et al. rightly call attention to the importance of embodied cultural practices (rather than just “ideas,” as in memetics) in the evolution of culture. Indeed, practices have primacy in steering not just behavior but thought. This “practice before ideology” principle can be seen in enculturation through religious rituals. Heylighen et al. (2018) observe that: “the undeniable act of praying to God can only be safeguarded from cognitive dissonance by denying any doubts you may have about the existence of God.” In FEP terms, the irrefutable perceptual evidence created by the active practice can only be explained away by adopting the ideological “hidden causes”. In rituals (as in many cultural practices), actions are triggered by cultural markers in the environment – I do it because others like me do/did it – circumventing explicit thought but at times also the actual interests of the participant. Here, practices become a tool for control of individual action by the social system (conformity pressure).
"Third, TTOM/FEP may provide a unified selection criterion (“fitness”) for the evolution of cultures. Culture constrains the behavioral paths for its individual members, generally because its practices and narratives have shown to be efficient free-energy reduction vehicles, for “agents like you”. Ideas and practices that reduce free energy more efficiently tend to spread and become dominant in a culture. These ideas could concern hidden causes of the environment and the challenges it provides (e.g., a god causing thunderstorms) but also hidden causes of the behavior of other individuals in your community. This gives cultural ideas a circular, self-reinforcing character. For example, the cultural expectation that sinning requires guilt and atonement reduces the free energy of the harmed party, but guilt also becomes a hidden cause efficiently explaining away someone’s behavior in the eyes of others belonging to the culture. However, note that the success of these expectations depends on the conservative perpetuation of the culture, and the exclusion of “dissident” behavior.
"Similarly, ideologies such as religion or nationalism, as interconnected sets of hidden causes and shared expectations and practices, are an efficient means of free-energy minimization. As Atran and Ginges (2012) remark, most religions have at their core a limited set of principles (expectations) that they consider “sacred.” In essence, to be sacred implies unconditionality. Indeed, expectations that are independent of contextual parameters provide a simple, dependable (high precision) foundation for how to act in and explain the (social) world. It makes these principles into very powerful, socially fulfilled hidden causes. The same free-energy minimization logic explains why strictly patterned (hence predictable) religious rituals are especially successful at important transition points in life (such as the transition to adulthood), characterized by higher uncertainty about how one should act. In the same vein, Hogg (2014) reports evidence that individuals that experience high personal uncertainty (e.g., adolescents going through identity problems) tend to strongly identify with a group and (radical) ideology to easily resolve their self-uncertainty. Examples can even be found of cultures systematically plunging their members into uncertainty to increase allegiance. Thus, cultures and their “sacred” rules often actually harm their members, hence outright increasing their free energy. Think, for example, of rules inducing genital mutilation, suicide terrorism, honor killings, or more mundanely, chronic stress because of a ruthless, sacred rule of productivity.
"The above examples show that a meme-centric concept of fitness will not do (Ramsey & De Block 2015), but, more interestingly, they also suggest that a purely organism-centric concept of fitness (organism-centered free-energy minimization) is unsatisfactory to explain the power of cultures on their members. Indeed, internalized and environmentally anchored cultural expectations (behavioral “rules”) often take on a life of their own, not necessarily benefitting the individual that follows them, but rather maintaining the very system of social ideas and practices they are part of. Luhmann (1986) has argued that social systems should be seen as autopoietic, organism-like agents that, via their human constituents, actively counteract any deviation from their organization, so as to ensure the continuation and self-regeneration of the system (Heylighen et al. 2018). Hence, these social systems seem to also reduce their free energy, consistent with a multiscale formulation of the FEP (Ramstead et al.2018). On the one hand, the relationship between individual and social system is one of symbiosis or mutual benefit, with social systems providing means for reducing free energy to the individual through coordination of action and prevention of conflicts. On the other hand, social systems, via TTOM mechanisms, can also veer into dogmatism, radicalism, and mind control that suppresses individual expression, creativity, and well-being (Heylighen et al. 2018). We believe that the account of Veissière et al. should also provide insight into this dark side of cultural phenomena."
"Second, Veissière et al. rightly call attention to the importance of embodied cultural practices (rather than just “ideas,” as in memetics) in the evolution of culture. Indeed, practices have primacy in steering not just behavior but thought. This “practice before ideology” principle can be seen in enculturation through religious rituals. Heylighen et al. (2018) observe that: “the undeniable act of praying to God can only be safeguarded from cognitive dissonance by denying any doubts you may have about the existence of God.” In FEP terms, the irrefutable perceptual evidence created by the active practice can only be explained away by adopting the ideological “hidden causes”. In rituals (as in many cultural practices), actions are triggered by cultural markers in the environment – I do it because others like me do/did it – circumventing explicit thought but at times also the actual interests of the participant. Here, practices become a tool for control of individual action by the social system (conformity pressure).
"Third, TTOM/FEP may provide a unified selection criterion (“fitness”) for the evolution of cultures. Culture constrains the behavioral paths for its individual members, generally because its practices and narratives have shown to be efficient free-energy reduction vehicles, for “agents like you”. Ideas and practices that reduce free energy more efficiently tend to spread and become dominant in a culture. These ideas could concern hidden causes of the environment and the challenges it provides (e.g., a god causing thunderstorms) but also hidden causes of the behavior of other individuals in your community. This gives cultural ideas a circular, self-reinforcing character. For example, the cultural expectation that sinning requires guilt and atonement reduces the free energy of the harmed party, but guilt also becomes a hidden cause efficiently explaining away someone’s behavior in the eyes of others belonging to the culture. However, note that the success of these expectations depends on the conservative perpetuation of the culture, and the exclusion of “dissident” behavior.
"Similarly, ideologies such as religion or nationalism, as interconnected sets of hidden causes and shared expectations and practices, are an efficient means of free-energy minimization. As Atran and Ginges (2012) remark, most religions have at their core a limited set of principles (expectations) that they consider “sacred.” In essence, to be sacred implies unconditionality. Indeed, expectations that are independent of contextual parameters provide a simple, dependable (high precision) foundation for how to act in and explain the (social) world. It makes these principles into very powerful, socially fulfilled hidden causes. The same free-energy minimization logic explains why strictly patterned (hence predictable) religious rituals are especially successful at important transition points in life (such as the transition to adulthood), characterized by higher uncertainty about how one should act. In the same vein, Hogg (2014) reports evidence that individuals that experience high personal uncertainty (e.g., adolescents going through identity problems) tend to strongly identify with a group and (radical) ideology to easily resolve their self-uncertainty. Examples can even be found of cultures systematically plunging their members into uncertainty to increase allegiance. Thus, cultures and their “sacred” rules often actually harm their members, hence outright increasing their free energy. Think, for example, of rules inducing genital mutilation, suicide terrorism, honor killings, or more mundanely, chronic stress because of a ruthless, sacred rule of productivity.
"The above examples show that a meme-centric concept of fitness will not do (Ramsey & De Block 2015), but, more interestingly, they also suggest that a purely organism-centric concept of fitness (organism-centered free-energy minimization) is unsatisfactory to explain the power of cultures on their members. Indeed, internalized and environmentally anchored cultural expectations (behavioral “rules”) often take on a life of their own, not necessarily benefitting the individual that follows them, but rather maintaining the very system of social ideas and practices they are part of. Luhmann (1986) has argued that social systems should be seen as autopoietic, organism-like agents that, via their human constituents, actively counteract any deviation from their organization, so as to ensure the continuation and self-regeneration of the system (Heylighen et al. 2018). Hence, these social systems seem to also reduce their free energy, consistent with a multiscale formulation of the FEP (Ramstead et al.2018). On the one hand, the relationship between individual and social system is one of symbiosis or mutual benefit, with social systems providing means for reducing free energy to the individual through coordination of action and prevention of conflicts. On the other hand, social systems, via TTOM mechanisms, can also veer into dogmatism, radicalism, and mind control that suppresses individual expression, creativity, and well-being (Heylighen et al. 2018). We believe that the account of Veissière et al. should also provide insight into this dark side of cultural phenomena."
A planetary model?
Wolfram Barfuss: "Collective action is required to enter sustainable development pathways in coupled social-ecological systems, safely away from dangerous tipping elements. Yet, we lack a proper theoretical grounding... current state-of-the-art Earth system models do not represent dynamic human societies and their feedback interactions with the biogeophysical Earth system. Earth system analysis of the Anthropocene requires closing the loop by integrating the dynamics of complex human societies into integrated whole Earth system models." Therefore he and coauthors presented "a simulation modeling framework that aims at facilitating the implementation and analysis of world–Earth (or planetary social–ecological) models."
It doesn't sound like predicting with certainty "what will happen" is a goal for this research. Here's a line that leads me to this conclusion: "although we show example trajectories that are based on parameters and initial conditions... the time evolutions shown may not be interpreted as any kind of meaningful quantitative prediction or projection." So instead, this research is more along the lines of scenario analysis, or comparative model and policy analysis. It is likely we can aspire for no more than this anyway. In agent-environment modeling, all we can say is that "given initial condition states and agent dynamics, the probability of any particular resulting conditions is greater or lesser, but never absolutely certain". Personally, I like how Barfuss frames the challenge, and I agree that "we lack a proper theoretical grounding", but I don't see how this paper resolves that fundamental problem. I am much more hopeful for the methods being proposed by Karl Friston concerning "active inference", which I think does provide a strong theoretical grounding for comparative analysis of policies. Applying that to these challenges would make a much more interesting paper.
"In the beginning was the model"
In his paper "I Interact Therefore I Am", Dimitris Bolis writes: "To survive, an organism obeys the following straightforward rule: adjust yourself to reality or change the reality itself. Humans use cultural models for describing, predicting and manipulating the environment. Many early societies construed natural phenomena, such as weather or earthquakes, as behavioral expressions of personified deities. This recruited powerful cognitive capacities originally developed for dealing with the undoubtedly complex social realm. Any level of abstraction can be considered as a model of the world. To come back to the example of language, a word can be thought of as a sociocultural model in and of itself, which presupposes the evolution of both the necessary biological apparatus across evolution, and an interpersonal attunement across development. For instance, the word ‘animal’ or ‘wave’ practically captures and summarizes higher level similarities being met in a plethora of diverse natural processes. The construction of internal models allows not only for the prediction of the world, but also the (socioculturally) transformation of it for meeting survival needs." Concerning the personification of nature and natural processes, both Gregory Bateson and Eduardo Kohn have described “ecologies of mind”, and Joi Ito related how prevalent animist beliefs in Japanese culture have enabled a more harmonious relationship with both natural and artificial materialities (in the Jane Bennett sense). Indeed, the panpsychist overtones of the FEP and active inference also lent credence to these beliefs of early societies, who may have not been as far from the mark as many members of contemporary Westerner culture, in their conceit, would believe.
In his paper "I Interact Therefore I Am", Dimitris Bolis writes: "To survive, an organism obeys the following straightforward rule: adjust yourself to reality or change the reality itself. Humans use cultural models for describing, predicting and manipulating the environment. Many early societies construed natural phenomena, such as weather or earthquakes, as behavioral expressions of personified deities. This recruited powerful cognitive capacities originally developed for dealing with the undoubtedly complex social realm. Any level of abstraction can be considered as a model of the world. To come back to the example of language, a word can be thought of as a sociocultural model in and of itself, which presupposes the evolution of both the necessary biological apparatus across evolution, and an interpersonal attunement across development. For instance, the word ‘animal’ or ‘wave’ practically captures and summarizes higher level similarities being met in a plethora of diverse natural processes. The construction of internal models allows not only for the prediction of the world, but also the (socioculturally) transformation of it for meeting survival needs." Concerning the personification of nature and natural processes, both Gregory Bateson and Eduardo Kohn have described “ecologies of mind”, and Joi Ito related how prevalent animist beliefs in Japanese culture have enabled a more harmonious relationship with both natural and artificial materialities (in the Jane Bennett sense). Indeed, the panpsychist overtones of the FEP and active inference also lent credence to these beliefs of early societies, who may have not been as far from the mark as many members of contemporary Westerner culture, in their conceit, would believe.
The danger of psychopathological models
Carl Jung wrote “Our personal psychology is just a thin skin, a ripple on the ocean of collective psychology. The powerful factor, the factor which changes our whole life, which changes the surface of our known world, which makes history, is collective psychology, which moves according to laws entirely different from those of our [individual] consciousness.” In an interview in 1959 on "Face to Face" Jung said: "We need more understanding of human nature, because the only real danger that exists is man himself. He is the great danger, and we are pitifully unaware of it. We know nothing of man, far too little. His psyche should be studied, because we are the origin of all coming evil." The difference between individual and collective psychology is entirely analagous to that between personal and suprapersonal models, a distinction notably made by Thomas Metzinger in a recent interview, and Karl Friston (in "The Markov Blankets of Life"), but also present in Maslow's hierarchy of needs when he described the "self actualizing" person as "focused on problems outside themself... in general these tasks are nonpersonal or "unselfish", concerned rather with the good of mankind in general". This is the distinction between individualistic and holistic models. When Jung pleads for a greater understanding of human nature, he knows that psychopathological models, left unexamined, are indeed changing the surface of our world today, in the form of societal and environmental disruption.
Carl Jung wrote “Our personal psychology is just a thin skin, a ripple on the ocean of collective psychology. The powerful factor, the factor which changes our whole life, which changes the surface of our known world, which makes history, is collective psychology, which moves according to laws entirely different from those of our [individual] consciousness.” In an interview in 1959 on "Face to Face" Jung said: "We need more understanding of human nature, because the only real danger that exists is man himself. He is the great danger, and we are pitifully unaware of it. We know nothing of man, far too little. His psyche should be studied, because we are the origin of all coming evil." The difference between individual and collective psychology is entirely analagous to that between personal and suprapersonal models, a distinction notably made by Thomas Metzinger in a recent interview, and Karl Friston (in "The Markov Blankets of Life"), but also present in Maslow's hierarchy of needs when he described the "self actualizing" person as "focused on problems outside themself... in general these tasks are nonpersonal or "unselfish", concerned rather with the good of mankind in general". This is the distinction between individualistic and holistic models. When Jung pleads for a greater understanding of human nature, he knows that psychopathological models, left unexamined, are indeed changing the surface of our world today, in the form of societal and environmental disruption.
Extended planning horizons
In Sophisticated Inference Friston et al. utilize the notion of 'sophistication' from the economics literature, which refers to the degree to which an agent has beliefs about beliefs. For example, there are agents with beliefs about the counterfactual consequences of action, who consider 'what would happen if I did that'. This is a 'level one' sophistication. Then there are agents who consider 'what would I believe about what would happen if I did that'. This is a 'level two' sophistication. And so on. These levels correspond to our 'planning horizon'. If the planning horizon is insufficient to enable us to contemplate distal (and potentially preferable) outcomes, we can easily get stuck in local minima as we pursue our goals. However with a greater planning horizon, local minima are vitiated, and we are able to plan and execute the shortest path to our ultimate goal, which often involves excursions through state (and belief) space that point away from it.
In Sophisticated Inference Friston et al. utilize the notion of 'sophistication' from the economics literature, which refers to the degree to which an agent has beliefs about beliefs. For example, there are agents with beliefs about the counterfactual consequences of action, who consider 'what would happen if I did that'. This is a 'level one' sophistication. Then there are agents who consider 'what would I believe about what would happen if I did that'. This is a 'level two' sophistication. And so on. These levels correspond to our 'planning horizon'. If the planning horizon is insufficient to enable us to contemplate distal (and potentially preferable) outcomes, we can easily get stuck in local minima as we pursue our goals. However with a greater planning horizon, local minima are vitiated, and we are able to plan and execute the shortest path to our ultimate goal, which often involves excursions through state (and belief) space that point away from it.
Integrated World Modeling Theory (IWMT)
Adam Safron writes: "The extreme generality of the free energy principle and active inference (FEP-AI) requires emphasis. Not only do nervous systems entail predictive models, but so do entire populations of organisms and their extended phenotypes as teleonomical predictions with respect to evolutionary fitness. In this way, FEP-AI provides a formalism where persisting dynamical systems can be understood as self-generating models, grounded in first principles regarding the necessary preconditions for existence in a world governed by the 2nd law.
"Each level in hierarchical predictive processing models the level below it, extending down to sensor and effector systems, with all these models being integrated when they are combined into larger generative models (e.g., brains and organisms). A hierarchy of Markov Blankets (MBs) constitute a hierarchy of selective pressures, with dynamics on one level being selected by the next level of organization. These informational shielding properties of MBs connect with debates regarding units of selection in evolutionary theory, in that only organismic phenotypes — and sometimes groups of organisms — are “visible” to natural selection with respect to phylogeny. Broadly speaking, nervous systems can be straightforwardly understood as generative probabilistic graphical models (PGMs). From this perspective, nervous systems can be viewed as modeling the world, to the extent neural dynamics reflect patterns in the world. The Bayesian brain hypothesis proposes this mutual information takes the form of probabilistic mappings from observations to likely causes, and that these inferences may approach bounded optimality with respect to ecological decision-theoretic objectives over phylogenetic and ontogenetic timescales.
"Coupling systems form larger systems via mutual entrainment. From an FEP-AI perspective, this coupling relationship is one of mutual modeling and collaborative inference. This generalized synchrony has also been characterized in thermodynamic terms, where systems spontaneously self-organize into resonant modes with the environments with which they couple where coordinated dynamics have been observed to contain mutually predictive information. While FEP-AI views all systems as models, only some of these models afford adaptivity, and only some systems also have models. Living organisms possess specific sub-systems capable of supporting generative models with temporal depth and counterfactual richness. These sub-systems are called brains, and they allow organisms to navigate exchanges with their environments by modeling not just present world configurations, but also possible world configurations predicted based on future (counterfactual) actions.
"Brains acquire especially powerful predictive modeling abilities when they are organized according to multiple layers of hierarchical depth. This deep organization allows these systems to model not only transient events at lower levels, but also their organization into more temporally extended sequences at higher levels. Further, deep internal dynamics create a potential for functional decoupling between modeling and the unfolding of particular sensorimotor engagements, thus enabling counterfactual simulations with temporal “thickness”/“depth”, which when conscious enable imagination and explicit planning. These capacities afford the possibility of constructing rich causal world models and the preconditions for coherent conscious experience."
Adam Safron writes: "The extreme generality of the free energy principle and active inference (FEP-AI) requires emphasis. Not only do nervous systems entail predictive models, but so do entire populations of organisms and their extended phenotypes as teleonomical predictions with respect to evolutionary fitness. In this way, FEP-AI provides a formalism where persisting dynamical systems can be understood as self-generating models, grounded in first principles regarding the necessary preconditions for existence in a world governed by the 2nd law.
"Each level in hierarchical predictive processing models the level below it, extending down to sensor and effector systems, with all these models being integrated when they are combined into larger generative models (e.g., brains and organisms). A hierarchy of Markov Blankets (MBs) constitute a hierarchy of selective pressures, with dynamics on one level being selected by the next level of organization. These informational shielding properties of MBs connect with debates regarding units of selection in evolutionary theory, in that only organismic phenotypes — and sometimes groups of organisms — are “visible” to natural selection with respect to phylogeny. Broadly speaking, nervous systems can be straightforwardly understood as generative probabilistic graphical models (PGMs). From this perspective, nervous systems can be viewed as modeling the world, to the extent neural dynamics reflect patterns in the world. The Bayesian brain hypothesis proposes this mutual information takes the form of probabilistic mappings from observations to likely causes, and that these inferences may approach bounded optimality with respect to ecological decision-theoretic objectives over phylogenetic and ontogenetic timescales.
"Coupling systems form larger systems via mutual entrainment. From an FEP-AI perspective, this coupling relationship is one of mutual modeling and collaborative inference. This generalized synchrony has also been characterized in thermodynamic terms, where systems spontaneously self-organize into resonant modes with the environments with which they couple where coordinated dynamics have been observed to contain mutually predictive information. While FEP-AI views all systems as models, only some of these models afford adaptivity, and only some systems also have models. Living organisms possess specific sub-systems capable of supporting generative models with temporal depth and counterfactual richness. These sub-systems are called brains, and they allow organisms to navigate exchanges with their environments by modeling not just present world configurations, but also possible world configurations predicted based on future (counterfactual) actions.
"Brains acquire especially powerful predictive modeling abilities when they are organized according to multiple layers of hierarchical depth. This deep organization allows these systems to model not only transient events at lower levels, but also their organization into more temporally extended sequences at higher levels. Further, deep internal dynamics create a potential for functional decoupling between modeling and the unfolding of particular sensorimotor engagements, thus enabling counterfactual simulations with temporal “thickness”/“depth”, which when conscious enable imagination and explicit planning. These capacities afford the possibility of constructing rich causal world models and the preconditions for coherent conscious experience."
Baltieri et al. write (again, paraphrasing): "Generative models are epistemic tools for an observer to specify the properties of a system. They are part of an effective formalism connecting and extending ideas such as the good regulator theorem (Conant and Ashby), perceptual control theory (Powers), and the notion of entailment in theoretical biology (Rosen). In this light, active inference plays an important instrumental role for the overarching attempt of unifying these results in the study of adaptive agents. Under this general framework, notions such as feedback, stability, inference, attractors, uncertainty, and dynamics can be seen from different, but complementary, perspectives that allow for more complete descriptions of agents and agency, where actions and policies of an agent are described as an inferential process biased towards ensuring that its normative constraints, such as its survival and very existence, are met.
A Markov blanket (or Douglas Adam's towel)
By virtue of being a Markov blanket, things can be distinguished from their environment. If something doesn’t have a Markov blanket, it doesn’t exist as an independent entity. The Markov blanket is a permeable interface between the inside and the outside, enabling a two-way exchange. With this you can go a lot further than 20th-century physics, which was all about equilibrium statistics and thermodynamics. Implicit in equilibrium physics is the notion that you’ve got an isolated or a closed system immersed in a heat bath — without ever asking where the heat bath came from. It just implicitly assumed a Markov blanket. But now, with Markov blankets, you can address open systems that are far from equilibrium — non-equilibrium steady states that persist despite the fact that they are in exchange with their environment. We are the perfect examples of things that seem to persist over time, despite a fluctuating and capricious world out there. Any system that survives reality, that is engendered by a world that is at least somewhat sympathetic to its existence, must embody or contain a model, and by virtue of that model regulate its exchange with the world. The inside has to have some form of synchronization with the outside.
By virtue of being a Markov blanket, things can be distinguished from their environment. If something doesn’t have a Markov blanket, it doesn’t exist as an independent entity. The Markov blanket is a permeable interface between the inside and the outside, enabling a two-way exchange. With this you can go a lot further than 20th-century physics, which was all about equilibrium statistics and thermodynamics. Implicit in equilibrium physics is the notion that you’ve got an isolated or a closed system immersed in a heat bath — without ever asking where the heat bath came from. It just implicitly assumed a Markov blanket. But now, with Markov blankets, you can address open systems that are far from equilibrium — non-equilibrium steady states that persist despite the fact that they are in exchange with their environment. We are the perfect examples of things that seem to persist over time, despite a fluctuating and capricious world out there. Any system that survives reality, that is engendered by a world that is at least somewhat sympathetic to its existence, must embody or contain a model, and by virtue of that model regulate its exchange with the world. The inside has to have some form of synchronization with the outside.
Scientists’ warning on affluence
Wiedmann et al. provide a very good overview of our current situation: "Long-term and concurrent human and planetary wellbeing will not be achieved in the Anthropocene if affluent overconsumption continues, spurred by economic systems that exploit nature and humans." Furthermore "the digital revolution—and more broadly the Fourth Industrial Revolution has only led to more consumption and inequality." Near the core of this problem lies a failed social model: "With every actor striving to increase their position relative to their peers, the average consumption level rises and thus even more expensive positional goods become necessary, while the societal wellbeing level stagnates. An individual’s happiness correlates positively with their own income but negatively with the peer group’s income, and unequal access to positional goods fosters rising consumption. This endless process is a core part of capitalism." As Francis Heylighten and others have noted, these processes harm people through "chronic stress because of a ruthless, sacred rule of productivity."
Key questions: "Which circumstances will allow for and support widespread shifts in lifestyles? How can these be motivated and sustained? What are the institutional, cultural and individual barriers to adopting lifestyle changes and how can they be overcome? What is the role of social groups, organisations and bottom-up movements? Can we learn from societies, e.g. indigenous and pre-industrial societies, which managed to live without economic growth?" Identified transition strategies: To address overconsumption and wealth concentration, we will need to "strengthen equality and redistribution through suitable taxation policies, basic income and job guarantees, set maximum income levels, expand public services and roll back neoliberal reforms." Also we can establish the "avoid-shift-improve framework" outlined by Creutzig et al. including: increasing the lifespans of goods, use telecommunication instead of physical travel, share and repair instead of buying new, and house retrofitting; shifting away from resource and carbon-intensive goods and services, e.g. mobility from cars and airplanes to public buses and trains, biking or walking, heating from oil heating to heat pumps, nutrition, where possible, from animal to seasonal plant-based products." Big picture: "To avoid further deterioration and irreversible damage to natural and societal systems, there will need to be a global and rapid decoupling of detrimental impacts from economic activity." Or, what I think is a more preferable way of putting it, there will need to be a recoupling and attunement between the human and natural systems.
Wiedmann et al. provide a very good overview of our current situation: "Long-term and concurrent human and planetary wellbeing will not be achieved in the Anthropocene if affluent overconsumption continues, spurred by economic systems that exploit nature and humans." Furthermore "the digital revolution—and more broadly the Fourth Industrial Revolution has only led to more consumption and inequality." Near the core of this problem lies a failed social model: "With every actor striving to increase their position relative to their peers, the average consumption level rises and thus even more expensive positional goods become necessary, while the societal wellbeing level stagnates. An individual’s happiness correlates positively with their own income but negatively with the peer group’s income, and unequal access to positional goods fosters rising consumption. This endless process is a core part of capitalism." As Francis Heylighten and others have noted, these processes harm people through "chronic stress because of a ruthless, sacred rule of productivity."
Key questions: "Which circumstances will allow for and support widespread shifts in lifestyles? How can these be motivated and sustained? What are the institutional, cultural and individual barriers to adopting lifestyle changes and how can they be overcome? What is the role of social groups, organisations and bottom-up movements? Can we learn from societies, e.g. indigenous and pre-industrial societies, which managed to live without economic growth?" Identified transition strategies: To address overconsumption and wealth concentration, we will need to "strengthen equality and redistribution through suitable taxation policies, basic income and job guarantees, set maximum income levels, expand public services and roll back neoliberal reforms." Also we can establish the "avoid-shift-improve framework" outlined by Creutzig et al. including: increasing the lifespans of goods, use telecommunication instead of physical travel, share and repair instead of buying new, and house retrofitting; shifting away from resource and carbon-intensive goods and services, e.g. mobility from cars and airplanes to public buses and trains, biking or walking, heating from oil heating to heat pumps, nutrition, where possible, from animal to seasonal plant-based products." Big picture: "To avoid further deterioration and irreversible damage to natural and societal systems, there will need to be a global and rapid decoupling of detrimental impacts from economic activity." Or, what I think is a more preferable way of putting it, there will need to be a recoupling and attunement between the human and natural systems.
The price of an uncertain future
Gideon Lewis-Kraus reports: "On January 20, both the United States and South Korea confirmed their first cases of Covid-19; Taiwan reported its first case the next day, and Singapore followed two days later. Epidemic parity began and ended there. By the end of March, those three Asian countries had largely contained at least the first wave of their outbreaks. These governments had earned their citizens' trust with a record of public investment and accountability. With a functional government, universal health care, and a much stronger social safety net, people can trust that seeking treatment won't bankrupt them, and that they can stay home from work without finding themselves unable to pay for their daily needs." This the right way to prepare for a pandemic: improve public trust in a social safety net that will reduce personal risk and uncertainty.
Olivia Remes writes: "About one in 14 people around the world are affected by anxiety disorders at any given time. Previous studies have linked anxiety to inflammatory processes in the body and suppression of the immune system, and increased risk for a host of other negative outcomes, such as cancer, heart disease, diabetes, and thyroid conditions. But there are certain measures we can take to alleviate feelings of anxiety and improve our overall health. The mind and body are intricately connected: one influences the other. Therefore, engaging in regular physical activity, getting enough sleep, and mindfulness meditation can all have positive effects on mental health, making you feel less stressed and anxious." We can't fully eliminate anxiety, but we can trace it's origin to fear, whether rational or not, about what the future may bring. Societies that are future oriented can do a lot to reduce public risk and improve health outcomes for their citizens.
Gideon Lewis-Kraus reports: "On January 20, both the United States and South Korea confirmed their first cases of Covid-19; Taiwan reported its first case the next day, and Singapore followed two days later. Epidemic parity began and ended there. By the end of March, those three Asian countries had largely contained at least the first wave of their outbreaks. These governments had earned their citizens' trust with a record of public investment and accountability. With a functional government, universal health care, and a much stronger social safety net, people can trust that seeking treatment won't bankrupt them, and that they can stay home from work without finding themselves unable to pay for their daily needs." This the right way to prepare for a pandemic: improve public trust in a social safety net that will reduce personal risk and uncertainty.
Olivia Remes writes: "About one in 14 people around the world are affected by anxiety disorders at any given time. Previous studies have linked anxiety to inflammatory processes in the body and suppression of the immune system, and increased risk for a host of other negative outcomes, such as cancer, heart disease, diabetes, and thyroid conditions. But there are certain measures we can take to alleviate feelings of anxiety and improve our overall health. The mind and body are intricately connected: one influences the other. Therefore, engaging in regular physical activity, getting enough sleep, and mindfulness meditation can all have positive effects on mental health, making you feel less stressed and anxious." We can't fully eliminate anxiety, but we can trace it's origin to fear, whether rational or not, about what the future may bring. Societies that are future oriented can do a lot to reduce public risk and improve health outcomes for their citizens.
Induced reasoning via choice blindness
Strandberd, Johansson, and Pärnamets write: "The current run of hyper-polarization in politics is a result of our larger societal views of what it is to hold an attitude. It can be especially harmful when society holds a standard where particularly firm opinions are held in reverie, and where undecideds and moderates are derided as “wishy-washers” and “flip-floppers". In this time of information bubbles, fake news, political acrimony, and gridlock, it is encouraging that a brief choice blindness paradigm (CBP) intervention can nudge people to find support for positions other than those originally held. This opens up new perspectives for understanding across the political divide and serves as a reminder that people can demonstrate flexibility when they are induced to reason about complex political issues." You can see CBP in action in this short video, or read more about it in an article related this research.
Strandberd, Johansson, and Pärnamets write: "The current run of hyper-polarization in politics is a result of our larger societal views of what it is to hold an attitude. It can be especially harmful when society holds a standard where particularly firm opinions are held in reverie, and where undecideds and moderates are derided as “wishy-washers” and “flip-floppers". In this time of information bubbles, fake news, political acrimony, and gridlock, it is encouraging that a brief choice blindness paradigm (CBP) intervention can nudge people to find support for positions other than those originally held. This opens up new perspectives for understanding across the political divide and serves as a reminder that people can demonstrate flexibility when they are induced to reason about complex political issues." You can see CBP in action in this short video, or read more about it in an article related this research.
How to break the causal chain
The past causes (or generates) the future. It makes certain outcomes more likely than others. As a consequence, biological systems need to use a generative model of their environment, that is to say, a probabilistic structure that generates predictions about the causes of sensory stimuli. It's how they are able to survive. And they need to be confident that they got it right so they don't make mistakes. Take for example a mosquito buzzing around me. It wants to bite me, but I don't want to let it. So I swipe at it, though it quickly sees my hand and dodges at the last minute. How was that possible? Through the mosquito's generative model. Or for another example, consider the chance of developing high blood pressure. If a person understands the causes that can lead to hypertension, then there is the potential to mitigate this health risk. But it requires making changes to their generative model, changing their perceptions and actions in a variety of situations. But this is not easy! Their current model has already been hard wired to be virtually unconscious, conditioned habits and responses triggered by numerous environmental and interoceptive features. Changes in our models, to be successful, require that we make corresponding structural changes in our habits and environments. New habits need a supportive environment to take root and grow in as these are coupled systems with mutually predictive information. Change, even when it comes to serious health issues, is not inevitable, especially when it requires conscious intervention to behavior choices. For this reason, pharmacological or surgical solutions are sometimes preferred even though a simple lifestyle adjustment could achieve the same result. When less thought is involved, the need for high commitment and conscious will power is avoided. Nonetheless, we must learn more about how to change, even though it may be hard. Even a mosquito must temporarily forego its meal if it wants to live to bite another day. As Bruineberg and Rietveld noted in a paper, "Given that we live in a highly dynamic world, being metastably poised to switch to multiple new relevant action-readiness patterns is itself an adaptive trait, and necessary for an agent to tend toward grip on its interactions with the environment." So we have at least two adages to follow: "adapt or die", and "don't get too comfortable". Is it time to "switch it up"?
Shifting baselines syndrome
“Cultures hang on to knowledge of things that are changing or gone longer,” J.B. MacKinnon says, “if those things are the kinds of things that they pay attention to.” On this topic David Roberts writes: "North America’s indigenous cultures still carry an enormous amount of accumulated knowledge that can help reveal what’s been lost. That kind of historical consciousness — a day-to-day awareness of the obligations that come with being a good ancestor — has faded. Today, the window of experience that humans find emotionally and cognitively salient has become too narrow to take in long-term changes in ecological systems. Put more bluntly: The public has stopped noticing that it’s getting warmer. And so, little by little, a hotter, more chaotic, and more dangerous world is becoming normal to us, as we sleepwalk toward more tragedies. How can the public be convinced that they should expect, and demand, something better?"
This relates to value #6 of my recently published Ecologue: the planning horizon or depth of our mental models can be very short or very long. If we are only concerned with "satisfying the next material desire" then that would suggest a very short planning horizon. But right now we need to use deep models to evaluate ecological health to avoid "shifting baselines syndrome". I agree with Roberts when he concludes "It cannot ultimately fall to ordinary people to hold baselines stable. Considering the welfare of coming generations and making decisions with the long view — those are things leaders are supposed to do." This is value #3 of the Ecologue: we certainly must remind leaders at all levels of government and within civic institutions that they are role-bearing persons. They will be held accountable. We also cannot forget that we are responsible for filling many of the most consequential leadership positions in society. "When the people lead, the leaders will follow". Ultimately, preserving ecological health falls on the shoulders of all of us in one way or another.
“Cultures hang on to knowledge of things that are changing or gone longer,” J.B. MacKinnon says, “if those things are the kinds of things that they pay attention to.” On this topic David Roberts writes: "North America’s indigenous cultures still carry an enormous amount of accumulated knowledge that can help reveal what’s been lost. That kind of historical consciousness — a day-to-day awareness of the obligations that come with being a good ancestor — has faded. Today, the window of experience that humans find emotionally and cognitively salient has become too narrow to take in long-term changes in ecological systems. Put more bluntly: The public has stopped noticing that it’s getting warmer. And so, little by little, a hotter, more chaotic, and more dangerous world is becoming normal to us, as we sleepwalk toward more tragedies. How can the public be convinced that they should expect, and demand, something better?"
This relates to value #6 of my recently published Ecologue: the planning horizon or depth of our mental models can be very short or very long. If we are only concerned with "satisfying the next material desire" then that would suggest a very short planning horizon. But right now we need to use deep models to evaluate ecological health to avoid "shifting baselines syndrome". I agree with Roberts when he concludes "It cannot ultimately fall to ordinary people to hold baselines stable. Considering the welfare of coming generations and making decisions with the long view — those are things leaders are supposed to do." This is value #3 of the Ecologue: we certainly must remind leaders at all levels of government and within civic institutions that they are role-bearing persons. They will be held accountable. We also cannot forget that we are responsible for filling many of the most consequential leadership positions in society. "When the people lead, the leaders will follow". Ultimately, preserving ecological health falls on the shoulders of all of us in one way or another.
Judea Pearl and Karl Friston
I was interested in seeing a response from the community of researchers loosely organized around Friston's "free energy principle" to Pearl’s well known 2001 paper, “Bayesianism and Causality, or, Why I am Only a Half-Bayesian”. In particular, I was curious whether his current thoughts on the limitations of Bayesian statistics are in any way relevant to current active inference formulations. I hadn't seen this explicitly addressed elsewhere, but I thought it might be important, so I asked him about it. He addressed my question by referencing an earlier post by Andrew Gelman that summed up his concerns. In response to that I wrote "Ah, so your concern is, succinctly, that elementary facts cannot be expressed in the vocabulary of probability calculus. This inherent limitation seems to be recognized within the active inference/ FEP community as well. There, "hidden causes" are merely inferred from observed consequences. This leaves significant room for error. Whereas you are concerned with factual accuracy, Friston is concerned with survival and adaptation. [An accurate understanding of causal structure is, apparently, not required for successful adaptation. So long as it enables context-sensitive "grip" on the environment, it is sufficient. It is notable how frequently active inference papers refer to the importance of mapping causality within generative models, but this is likely an exaggerated shorthand. In Jelle Bruineberg et al.'s paper "The anticipating brain is not a scientist" he notes that free energy minimization isn't the goal of organisms, it is simply the means by which grip is achieved.] As I understand it, the assumption, laid out in Ramstead et al.'s paper, is that it is possible to infer causes from consequences by simply inverting a probabilistic, correlational model, once it has been tuned to the hidden causal processes in the environment. You suggest we haven't yet been able to algorithmatize that." Pearl responded: "That something is possible is an aspiration, not assumption. As to going from probabilities to causal processes, it is in general impossible (provably) unless we start with causal assumptions, which one should explicate before starting to search for the impossible." Pearl is strict in his use of terminology! I should have said something more along the lines of: "The 'aspiration' is that it is possible to infer 'causal assumptions' from consequences by inverting a probabilistic, correlational model, once tuned to the hidden causal processes in the environment." But Pearl's comment did illuminate an interesting point: To go from probabilities to causal processes, we must start with causal assumptions. And how do we get causal assumptions? From a generative model. So combining your paper with Pearl's terminology I get something like: A generative model maps sensory data of observed consequences (represented in probability calculus) and when this map is inverted it allows an agent to make 'causal assumptions' about hidden causal processes. Now we are at the same place as Pearl, who recently said "Causal analysis is easy, but requires causal assumptions (or experiments) and those assumptions require a new mathematical notation, and a new calculus." I think this reconciles my confusion and gives me confidence that Pearl and the FEP community are on the same path.
I was interested in seeing a response from the community of researchers loosely organized around Friston's "free energy principle" to Pearl’s well known 2001 paper, “Bayesianism and Causality, or, Why I am Only a Half-Bayesian”. In particular, I was curious whether his current thoughts on the limitations of Bayesian statistics are in any way relevant to current active inference formulations. I hadn't seen this explicitly addressed elsewhere, but I thought it might be important, so I asked him about it. He addressed my question by referencing an earlier post by Andrew Gelman that summed up his concerns. In response to that I wrote "Ah, so your concern is, succinctly, that elementary facts cannot be expressed in the vocabulary of probability calculus. This inherent limitation seems to be recognized within the active inference/ FEP community as well. There, "hidden causes" are merely inferred from observed consequences. This leaves significant room for error. Whereas you are concerned with factual accuracy, Friston is concerned with survival and adaptation. [An accurate understanding of causal structure is, apparently, not required for successful adaptation. So long as it enables context-sensitive "grip" on the environment, it is sufficient. It is notable how frequently active inference papers refer to the importance of mapping causality within generative models, but this is likely an exaggerated shorthand. In Jelle Bruineberg et al.'s paper "The anticipating brain is not a scientist" he notes that free energy minimization isn't the goal of organisms, it is simply the means by which grip is achieved.] As I understand it, the assumption, laid out in Ramstead et al.'s paper, is that it is possible to infer causes from consequences by simply inverting a probabilistic, correlational model, once it has been tuned to the hidden causal processes in the environment. You suggest we haven't yet been able to algorithmatize that." Pearl responded: "That something is possible is an aspiration, not assumption. As to going from probabilities to causal processes, it is in general impossible (provably) unless we start with causal assumptions, which one should explicate before starting to search for the impossible." Pearl is strict in his use of terminology! I should have said something more along the lines of: "The 'aspiration' is that it is possible to infer 'causal assumptions' from consequences by inverting a probabilistic, correlational model, once tuned to the hidden causal processes in the environment." But Pearl's comment did illuminate an interesting point: To go from probabilities to causal processes, we must start with causal assumptions. And how do we get causal assumptions? From a generative model. So combining your paper with Pearl's terminology I get something like: A generative model maps sensory data of observed consequences (represented in probability calculus) and when this map is inverted it allows an agent to make 'causal assumptions' about hidden causal processes. Now we are at the same place as Pearl, who recently said "Causal analysis is easy, but requires causal assumptions (or experiments) and those assumptions require a new mathematical notation, and a new calculus." I think this reconciles my confusion and gives me confidence that Pearl and the FEP community are on the same path.
An interesting observation regarding Pearl and Friston is the POV each adopts, when speaking of causality Pearl adopts a more objective perspective, while Friston is agential and subjective - causal inferences are causal precisely because they are the agent's hypotheses about what causes sensory data. In the end, does Pearl's "causal revolution" and an "ecological
revolution" based on active inference both depend upon a fundamentally similar foundation? I think they are likely parallel attempts. In Pearl's recent interview with Lex Fridman he shared an understanding that is very similar to that of Friston and the motivation behind active inference: "Probability is the degree of uncertainty that an agent has about the world. A statistical view of the universe is useful because it allows you to predict things with a certain probability, and computing those probabilities is very useful. That's the whole idea of prediction. And you need prediction to be able to survive. If you cannot predict the future then just crossing the street would be extremely fearful. Correlation occurs when two things vary together over a very long time, or when a bunch of variables all vary cohesively, then we have a correlation. Usually when we think about correlation, we really think causation. Things cannot be correlated unless there is a reason for them to vary together. Why should they vary together? If they don't see each other, why should they vary together? So hidden in our intuition there is a notion of causation, because we cannot grasp any other logic except causation. ...All our life, all our intelligence, is built around metaphors, mapping from the unfamiliar to the familiar, but the marriage between the two is a tough thing, which we haven't yet been able to algorithmatize."
A critical response
Raja et al. write a critical review of the FEP: "By using Markov blankets, one can effectively model any thing as if it were a variational autoencoder (Kingma & Welling 2019) or a Helmoltz machine (Dayan et al. 1995), and Bayesian inference naturally follows from that. We refer to this move as The Markov Blanket Trick... The historical development of FEP suggests that the actual logical flow is: if you are able to model any system as if it were an autoencoder/a Helmholtz machine, you can describe any system as engaging in Bayesian inference; Markov blankets permit you to model almost anything as if it were an autoencoder/a Helmholtz machine; thus we can model anything as engaging in Bayesian inference; therefore FEP holds. In this context, it is not surprising that, for instance, FEP accommodates computational realism. In this sense, FEP is not different from the computational metaphor. ...The FEP is not a principle for a theory of everything. This much is generally acknowledged in the FEP literature. This does undermine some of the more outlandish claims about the scope of the FEP as a totalizing and universal theory. Nevertheless, if the FEP notion of thing is able to accommodate biological and cognitive systems within its formal constraints, FEP would still be a powerful framework."
Raja et al. write a critical review of the FEP: "By using Markov blankets, one can effectively model any thing as if it were a variational autoencoder (Kingma & Welling 2019) or a Helmoltz machine (Dayan et al. 1995), and Bayesian inference naturally follows from that. We refer to this move as The Markov Blanket Trick... The historical development of FEP suggests that the actual logical flow is: if you are able to model any system as if it were an autoencoder/a Helmholtz machine, you can describe any system as engaging in Bayesian inference; Markov blankets permit you to model almost anything as if it were an autoencoder/a Helmholtz machine; thus we can model anything as engaging in Bayesian inference; therefore FEP holds. In this context, it is not surprising that, for instance, FEP accommodates computational realism. In this sense, FEP is not different from the computational metaphor. ...The FEP is not a principle for a theory of everything. This much is generally acknowledged in the FEP literature. This does undermine some of the more outlandish claims about the scope of the FEP as a totalizing and universal theory. Nevertheless, if the FEP notion of thing is able to accommodate biological and cognitive systems within its formal constraints, FEP would still be a powerful framework."
History shapes policy selection
Environmental problems, social problems, and historical context are all inseparable. So when different social groups collaborate to select policy solutions they should begin by acknowledging that the past is still with us today. Although at the moment various forces may have conspired to bring different groups together to the face the same challenges, each group arrived via very different paths. This is enormously important. Because even though they may see the same contemporary threats, they may not understand the different historical conditions that each has experienced. It is those conditions that will also influence what each group believes to be the appropriate response. If this isn't recognized, then even though each group may be highly motivated to solve the same problem, together they may disagree on the appropriate policy solutions as long as their differing perspectives remain mutually unintelligible. Consequently in their deliberations they will "talk past each other" in trying to reach a consensus of opinion. History provides us with our culture, our prior beliefs, and our mental models of the world and everything that follows from these. This is where we must begin: How did we all get here?
Environmental problems, social problems, and historical context are all inseparable. So when different social groups collaborate to select policy solutions they should begin by acknowledging that the past is still with us today. Although at the moment various forces may have conspired to bring different groups together to the face the same challenges, each group arrived via very different paths. This is enormously important. Because even though they may see the same contemporary threats, they may not understand the different historical conditions that each has experienced. It is those conditions that will also influence what each group believes to be the appropriate response. If this isn't recognized, then even though each group may be highly motivated to solve the same problem, together they may disagree on the appropriate policy solutions as long as their differing perspectives remain mutually unintelligible. Consequently in their deliberations they will "talk past each other" in trying to reach a consensus of opinion. History provides us with our culture, our prior beliefs, and our mental models of the world and everything that follows from these. This is where we must begin: How did we all get here?
Variational ecology
In "Variational ecology and the physics of sentient systems" Ramstead et al. write: "Variational ecology (VE) provides an explanation of collective purposive action and intentionality of living systems – a physics of sentient systems. An ecological niche is a structured set of affordances that are shared by agents, which enables its denizens to coordinate purposive action over sometimes vast spatial and temporal distances. These carry semiotic and axiological meaning, like moral states held in common. As they engage with the affordances of their niche, ensembles of agents: (i) maintain the structural integrity of the niche through niche construction (and active inference); and (ii) collectively enact a generative model of their relation to the niche, thereby providing an account of the physics of intentionality, and especially of shared intentionality. Variational ecology, then, is also a physics of interacting minds. ...How can a 'variational ecology' shed light on selecting adaptive actions? When applied to cultural animals such as humans, variational ecology enables us to formulate not just a physics of individual minds, but also a physics of interacting minds across spatial and temporal scales – a physics of sentient systems that range from cells to intersubjective and sociocultural dynamics."
Confucian role ethics
In Confucian Role Ethics, Roger Ames said: "If we think ecologically, each element within an ecology is what it is by virtue of its relationship for the other members that constitute the ecology... [But] what we're doing right now is we're favoring the human being at the expense of the rest of the planet, and if you keep on going in that direction we won't continue. We'll be gone... For me individualism is a fiction, and in its own time it was a liberating fiction, but the fact of the matter is that that it's now become a pernicious fiction. Eight people in the world own half of the world's wealth. This idea of autonomy, and freedom, and rationality as being the fundamental definition of the human being because we don't have an alternative, becomes a default position. So then you have a starting point where everything is in fragments... If we think of sovereign states as individuals, then we have a model of winners and losers. But the kind of problems that the human being faces today: global warming, pandemics, food and water shortages, nuclear proliferation, environmental degradation, massive species extinction, all of these problems can't be solved by America, can't be solved by Israel, can't be solved by China. But together we have the cultural resources where we can make a go of it, and so we have to move from a winner and loser individual model to winners and winners. Either we win together or we lose together. It's that simple." Ames quoted Confucius' Analects 15.16: "There is nothing that I can do for someone who is not constantly asking himself: "What should I do? What should I do?" Ames remarked on this "Morality is hard work." (This was later echoed by Robert Hutchins, who also said "What should I do, what should we do, why should we do these things?")
Henry Rosemont Jr., a frequent collaborator with Ames, wrote: "For myself, I believe that if increasingly diverse communities, ethnic and religious groupings, and nation states are to live together in amity not enmity, and do so democratically, it is necessary to fundamentally alter the conception of what it is to be a human being that currently undergirds virtually all political, legal, and moral thinking today, and dominates the discourse on these issues as they are shaped by the government of the United States, and pretty much shared by so-called liberals and conservatives alike. The view I will advance defines human beings most fundamentally as standing in sets of relations, not with deity, but with other human beings living and dead, a view first articulated in the texts of classical Confucianism, wherein the absence of the concept of a creator deity in no way attenuates the moral, political and spiritual insights to be found in them. The concept of harmony looms large in these texts, and to my mind the concept is essential for genuine intercultural dialogues, dialogues enhanced by taking place between role-bearing cooperating persons rather than rights- bearing competing individuals."
Reinforcement Learning or Active Inference
This paper coauthored by Rosalyn Moran is excellent: “From our simulations, we find that unlike an agent that simply seeks to maximise reward, a free energy minimizing agent can develop an internal model of the environment in which it is placed. The free energy principle, and active inference, the dual normative and process theory of the brain, can reveal a more comprehensive understanding of psychopathologies, as these likely arise from dysfunctional models of the environment, maladaptive learning, or failures of inference.” This makes sense from an intuitive perspective. If my model of the world is dysfunctional, my actions will likely appear so as well.
However a skeptical Reddit commenter, describing Friston's active inference framework, says "Feels like he is just rehashing the standard "RL [reinforcement learning] as inference" thing, but goes out of his way to make it sound different." Is the commenter right? Let's consider that in Friston's 2012 paper "What is Value?" one of the subheadings is titled "optimal control as inference", and notably in 2018 Sergey Levine wrote "A number of prior works have also sought to incorporate probabilistic inference into a model of biological decision making and control. The formulation proposed by Friston is similar." So yes, there is definitely some convergence in these approaches. What, if anything, distinguishes them? Friston is clear that he does not invoke the notion of "reward, value or utility" in his theory, in place of these "surprise" is the primary motivator. As he writes: "It is interesting to note that classical rewards and punishments only have meaning when one agent teaches another; for example in social neuroscience or exchanges between an experimenter and subject. It should be noted that in value-learning and free-energy schemes there are no distinct rewards or punishments; every sensory signal has an expected cost, which, in the present context, is just surprise." Adaptation and survival usually means just that: avoid surprise and situations that are highly unpredictable, like conflict (per Philip Henshaw), and replace them with peace and harmony. It may sound like a cliche, but that's what it boils down to.
Scaling selfhood increases teleological capacity
Dennett and Levin wrote "Cognition all the way down", a great article. First, it highlights a central feature of life, that life pursues goals. Second, it notes that when many individuals work together to pursue the same goals, they can form a "superorganism" (or as the authors put it, a "superagent"). The way that this is described here is as a process of "erasing boundaries", I think this part of their argument is very interesting, but incomplete. I might've characterized this more as a process of realizing functional complementarity and synergy (Peter Corning) such that the fortunes of multiple agents become inseparably bound together. There are likely other caveats. But overall, the article is very much related to Karl Friston's speculation on the implications for the spatial and temporal scope of teleological behavior under active inference accounts of cognition, in papers such as "Sophisticated Inference", "Sentience and the Origins of Consciousness", and also Ramstead's "Variational ecology and the physics of sentient systems". Paraphrasing Dennett and Levin:
“Teleophobia significantly holds back the ability to predict and control complex systems because it prevents discovery of their most efficient internal controls or pressure points... What’s important for being a cognitive agent with some degree of sophistication is the scale of its goals. Agents can combine into networks, scaling their tiny, local goals into more grandiose ones belonging to a larger, unified self. And of course, any cognitive agent can be made up of smaller agents, each with their own limits on the size and complexity of what they’re working towards. Cooperation is ensured by the erasure of boundaries between the agents, forming a kind of superagent in which the individual identity of the original ones is very hard to maintain. The key dynamic that evolution discovered is a special kind of communication allowing privileged access of agents to the same information pool, which in turn made it possible to scale selves. This kickstarted the continuum of increasing agency. One way to categorise and compare cognitive systems, whether artificial or evolved, simple or complex, is by mapping the size and shape of the goals it can support (represent and work toward).”
Additional References:
A Multi-scale View of the Emergent Complexity of Life: A Free-Energy Proposal (February 2019)
Deeply Felt Affect: The Emergence of Valence in Deep Active Inference (December 2019)
Regimes of Expectations: An Active Inference Model of Social Conformity and Human Decision Making (March 2019) [paper]
Regimes of Expectations: An Active Inference Model of Social Conformity and Human Decision Making (March 2019) [article]
Computational Nosology and Precision Psychiatry (October 2017)
Scaling active inference (November 2019)
Thinking Through Other Minds: A Variational Approach to Cognition and Culture (May 2019)
TTOM in Action: Refining the Variational Approach to Cognition and Culture (December 2019)
Reinforcement Learning through Active Inference (February 2020)
Multiscale integration: beyond internalism and externalism (February 2019)
Deep temporal models and active inference (June 2017)
The theory of constructed emotion: an active inference account of interoception and categorization (January 2017)
Impulsivity and Active Inference (December 2018)
Blogposts and articles:
The Brain's Schrödinger Equation by Stanislas Dehaene (2008)
Grand Unified Theory of the Brain? by Steven Craig Hickman (2016)
Third-wave extended mind and predictive processing (June 2019)
The neuroscience of adversity vs Friston’s free energy principle (March 2019)
If you found this post interesting, see also "Shared models (part 1): modeling and simulation".
A Multi-scale View of the Emergent Complexity of Life: A Free-Energy Proposal (February 2019)
Deeply Felt Affect: The Emergence of Valence in Deep Active Inference (December 2019)
Regimes of Expectations: An Active Inference Model of Social Conformity and Human Decision Making (March 2019) [paper]
Regimes of Expectations: An Active Inference Model of Social Conformity and Human Decision Making (March 2019) [article]
Computational Nosology and Precision Psychiatry (October 2017)
Scaling active inference (November 2019)
Thinking Through Other Minds: A Variational Approach to Cognition and Culture (May 2019)
TTOM in Action: Refining the Variational Approach to Cognition and Culture (December 2019)
Reinforcement Learning through Active Inference (February 2020)
Multiscale integration: beyond internalism and externalism (February 2019)
Deep temporal models and active inference (June 2017)
The theory of constructed emotion: an active inference account of interoception and categorization (January 2017)
Impulsivity and Active Inference (December 2018)
Blogposts and articles:
The Brain's Schrödinger Equation by Stanislas Dehaene (2008)
Grand Unified Theory of the Brain? by Steven Craig Hickman (2016)
Third-wave extended mind and predictive processing (June 2019)
The neuroscience of adversity vs Friston’s free energy principle (March 2019)
If you found this post interesting, see also "Shared models (part 1): modeling and simulation".
No comments:
Post a Comment