I enjoyed reading Andy Clark's book "Surfing Uncertainty", it provided an excellent account of how we form and use models of the world. But more importantly, it highlighted why these are critical for cultural evolution. As we navigate into an increasingly uncertain future of existential threats, these processes become more important than ever before. My New Year's resolution was to "begin a project of identifying the models we use, what they consist of, where these coincide with and diverge from each other, and how they influence our relationships." It was a search for models, model predictions, and model evidence. Of the three, perhaps it is the uncovering of evidence that occupies the majority of our thoughts, actions, and creates feelings of anxiety or satisfaction. This book helps to place these endeavors on a firm foundation. Paraphrasing Clark:
"Our best takes on the world are given material form and made available (in that new guise) as publicly perceptible objects - words, sentences, equations. An important side-effect of this is that our own thoughts and ideas now become available, to ourselves and others, as potential objects for deliberate processes of attention. This opens the door to a whole array of knowledge-improvement and knowledge-testing techniques, ranging from simple conversations in which we ask for reasons, to the complex practices of testing, dissemination, and peer-review characteristic of contemporary science. Courtesy of all that material public vehicling in spoken words, written text, diagrams, and pictures, our best predictive models of the world (unlike those of other creatures) have thus become stable, reinspectable objects apt for public critique and systemic, multi-agent, multi-generational test and refinement.
"These models of the world, rooted in patterns of organism-environment interaction, serve as the basis for cumulative, communally distributed reasoning, rather than just providing the means by which individual thoughts occur. They are structured, meaningful realms apt for perception, thought, imagination, and action. They deliver a grip on the patterns that matter for the interactions that matter. All this hints at the shape of a truly fundamental and deeply unified science. But there remain challenges that need to be addressed. Among these is the extension of these accounts into the intuitively 'higher level' domains of long-term planning, cognitive control, social cognition, conscious experience, and explicit, linguistically inflected, reasoning. Also challenging will be the implied reconstruction of motivation, value, and desire in terms of more fundamental processes of uncertainty, inference, and prediction. Understanding the interplay between culture, technology, action, and cascading neural prediction is surely one of the major tasks confronting twenty-first-century cognitive science.
"Combining fast, cheap modes of response, such as cached strategies, with more costly, effortful mental simulations reveals that both are simply extreme poles on a continuum that may appear in many mixtures and combinations determined by the task at hand and the need to balance efficacy and efficiency. They emerge and dissolve in ways determined by the progressive reduction of precise, high-quality, prediction error - the all-purpose adhesive that binds elements from brain, body, and world into temporary problem-solving wholes." As Karl Friston says, "It's a dance, a dialogue." Our task, then, is to keep the beat, sing in tune, and move in synchrony. ...I must admit however, that in this regard, my skills are not so well refined as those of my wife. I need practice.
Below are several quotes (at times paraphrased for brevity) from the book itself, with corresponding pages on which they appear:
122 'Active Inference' names the combined mechanism by which perceptual and motor systems conspire to reduce prediction error using the twin strategies of altering predictions to fit the world, and altering the world to fit the predictions.
123 Thinking, predicting, and doing are all part of the same unfolding of sequences. (Hawkins and Blakeslee)
129 Behaviors are brought about by the interaction of our beliefs with the environment. Reward and pleasure are consequences of some of those interactions, but they are not the cause of those interactions. Instead, complex expectations drive behavior, causing us to probe and sample the world in ways that may often deliver reward or pleasure. In this way, reward is a perceptual (hedonic) consequence of behavior, not a cause.
137 Causality, as present in the real world, is reversed in the inner world. A mental representation of the intended effect of an action is the cause of the action: here it is not the action that produces the effect, but the (internal representation of the effect) that produces the action. (Pezzulo et al)
138 Perception, cognition, and action are manifestations of a single adaptive regime geared to the reduction of organism-salient prediction error.
139 Actions are consequences of expectations encoded in models.
169 What we actually perceive are our brain's models of the world. Not the world itself. You could say that our perceptions are fantasies that coincide with reality (Frith)
171 Our brains are a tool for encountering a world of significance, populated by human affordances.
190 From incomplete and fragmentary data, one generates hypotheses (or models) for the true nature of the world, which are then tested against and modified in light of further incoming sensory stimulation. (Anderson)
195 What we perceive is the structured external world itself. But this is not the world 'as it is', rather it is a world parsed according to our organism-specific needs and action repertoire. If a label is required, it may most safely be dubbed 'not-indirect perception'.
239 This is a world in which unexpected absences are every bit as salient as that which is real and present. Predictive processing offers an approach to understanding agency, experience, and human mattering.
244 Predictive processing combines fast, cheap modes of response with more costly, effortful strategies, revealing these as simply extreme poles on a continuum of self-organizing dynamics.
246 "Principle of Ecological Balance" and "Ecological Psychology"
250 Agents try to "guess the world" to help them achieve their goals while avoiding fatally surprising encounters.
254 The model-free response is inherently backwards-looking, associating specific actions with previously encountered rewards. It is condemned to repeat the past, releasing previously reinforced actions when circumstances dictate. A model-based system, by contrast, is able to evaluate potential actions using (as the name suggests) some kind of inner surrogate of the external arena in which actions are to be performed and choices made. Animals that deploy a model-based system are thus able to 'navigate into the future' rather than remaining 'driven by the past'. ...'Model-based' and 'model-free' modes of valuation and response simply name extremes along a continuum and may appear in many mixtures and combinations determined by the task at hand.
254 Agents seek both to maximize the accuracy of their predictions while minimizing the complexity of the models they use to generate those predictions. It's a delicate accuracy/complexity trade-off.
261 Deploying a simple cached strategy, a more costly mental simulation, or exploiting the environment itself as a cognitive resource are thus all strategies apt for context-sensitive recruitment using the predictive processing apparatus. The formation and dissolution of extended (brain-body-world) problem-solving ensembles balances efficacy and efficiency. They emerge and dissolve in ways determined by the progressive reduction, in environmental context, of precise, high-quality, prediction error - the all-purpose adhesive that binds elements from brain, body, and world into temporary problem-solving wholes.
266 Some systems have a tendency to destroy their own fixed points (attractors), actively inducing instabilities in ways that result in 'peripatetic or itinerant (wandering) dynamics'. Such systems would appear to pursue change and novelty 'for their own sake'.
278 Our best takes on the world are given material form and made available (in that new guise) as publicly perceptible objects - words, sentences, equations. An important side-effect of this is that our own thoughts and ideas now become available, to ourselves and others, as potential objects for deliberate processes of attention. This opens the door to a whole array of knowledge-improvement and knowledge-testing techniques, ranging from simple conversations in which we ask for reasons, to the complex practices of testing, dissemination, and peer-review characteristic of contemporary science. Courtesy of all that material public vehicling in spoken words, written text, diagrams, and pictures, our best predictive models of the world (unlike those of other creatures) have thus become stable, reinspectable objects apt for public critique and systemic, multi-agent, multi-generational test and refinement. Our best models of the world are thus able to serve as the basis for cumulative, communally distributed reasoning, rather than just providing the means by which individual thoughts occur.
288 Language allows ideas to be preserved and (in some sense) migrate between individuals. Such migrations may allow the communal construction of delicate and difficult intellectual trajectories and progressions. An idea which only Joe's experience makes available, but which can flourish and realize its full potential only in the intellectual niche currently provided by the brain of Mary, can now realize that potential by journeying between those agents.
289 Agents help construct the very worlds they model and inhabit, corresponding closely to the notion of 'enacting a world'.
291 Predictive processing invokes models that aim to engage the world, rather than to depict it in some action-neutral fashion. They are rooted in the patterns of organism-environment interaction. The role of these models is to deliver an efficient, context-sensitive grip upon a world of multiple competing affordances for action. ...A grip on the patterns that matter for the interactions that matter. A model that helps maintain the integrity and viability of a system by enabling it to minimize prediction error and thus avoid compromising (possibly fatal) encounters with the environment. A model is a structured meaningful realm apt for perception, thought, imagination, and action. (293)
294 The generative model that issues sensory predictions is nothing but a multi-level, multi-area, multi-scale, body-and-action involving grip on the unfolding sensory stream. Understanding the interplay between culture, technology, action, and cascading neural prediction is surely one of the major tasks confronting twenty-first-century cognitive science.
297 There are hints here of a new understanding of what it means to encounter the world in perception. This will be an understanding in which experience, expectation, estimated uncertainty, and action are inextricably intertwined.
297 All this hints at the shape of a truly fundamental and deeply unified science of the embodied mind. But there remain challenges that need to be addressed. Among these is the extension of these accounts into the intuitively 'higher level' domains of long-term planning, cognitive control, social cognition, conscious experience, and explicit, linguistically inflected, reasoning. Most challenging of all, perhaps, will be the implied reconstruction of motivation, value, and desire in terms of more fundamental processes of prediction, Bayesian inference, and self-estimated uncertainty. (299)
No comments:
Post a Comment