"Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science"
Abstract: Brains, it has recently been argued, are essentially prediction machines. They are bundles of cells that support perception and action by constantly attempting to match incoming sensory inputs with top-down expectations or predictions. This is achieved using a hierarchical generative model that aims to minimize prediction error within a bidirectional cascade of cortical processing. Such accounts offer a unifying model of perception and action, illuminate the functional role of attention, and may neatly capture the special contribution of cortical processing to adaptive success. The paper critically examines this 'hierarchical prediction machine' approach, concluding that it offers the best clue yet to the shape of a unified science of mind and action. Sections 1 and 2 lay out the key elements and implications of the approach. Section 3 explores a variety of pitfalls and challenges, spanning the evidential, the methodological, and the more properly conceptual. The paper ends (sections 4 and 5) by asking how such approaches might impact our more general vision of mind, experience, and agency.
This blog reports new ideas and work on mind, brain, behavior, psychology, and politics - as well as random curious stuff. (Try the Dynamic Views at top of right column.)
Monday, February 20, 2012
Our brains as prediction machines - a unified view of mind and action
Anything Andy Clark writes is totally worth reading (I used his charming essay "I am John's brain" when I first began teaching my "Biology of Mind Course" at the University of Wisconsin in the 1990's), and so I pass on this manuscript of an article on which comments are currently being solicited. It is a fascinating read, lucidly and clearly written.
Thank you for making the article available. I would like to submit some comments.
ReplyDeleteI have an alternative approach to models of action and perception that stands in sharp contrast to those of Prof. Clark and others whose approach is chiefly computational. My approach is stated on a web page and in an essay titled "How to solve free-will puzzles and overcome limitations of platonic science" at http://www.quadnets.com/puzzle.html
My approach is based on thermodynamics, which I taught and studied as a graduate student in a materials sciences laboratory and which I have investigated for many years since. I challenge the approach taken by Clark (without discussing his work) and others he cites that is based on statistical mechanics, also known as thermostatics. Statistical mechanics only applies to systems at equilibrium or undergoing continuous quasi-static changes. Statistical mechanics does not apply to a high-dissipation system like the human brain that is burning through some 20% of the entire stream of bodily energy. It is such energy that is tracked in images like MRI's. Nor does statistical mechanics apply to discontinuous transformations like those that occur in the generation of snowflakes (one of my favorite physics problems) and that I suggest are occurring in our brains and driving our muscular movements.
In contrast to selections based on predictive coding, as proposed by Prof. Clark, I suggest selections based on memories of the past. Such memories are signaled, indexed and encoded by images (also known as experiences or qualia) that are generated during our activities. I suggest that brains try to match successful activities in the past rather than trying to make successful predictions about the future. That is, in my approach, the function of imagery is to enable matching by memory. In contrast, I do not see any function for imagery in the predictive approach of Prof. Clark.
Thank you again for the posting.
Bob Kovsky
http://www.quadnets.com
http://www.kovsky.com
Very interesting! Thank you.
ReplyDelete