The brain is the result of evolution. A lot of evolution. Most neuroscientists don't really think about this fact. Should we? On this episode we talk about two papers---one focused on brains and the other on AI---that argue that following evolution is the path to success. As part of this argument, they make the point that, in evolution, each stage along the way needs to be fully functional, which impacts the shape and role of the brain. As a result, the system is best thought of as a whole---not chunked into perception, cognition and action, as many psychologists and neuroscientists are wont to do. In discussing these arguments, we talk about the role of representations in intelligence, go through a bit of the evolution of the nervous system, and remind ourselves that evolution does not necessarily optimize. Throughout, we ask how this take on neuroscience impacts our own work and try to avoid saying "represents".
We read:
Resynthesizing behavior through phylogenetic refinement
Intelligence without Representation
And we mentioned previous episode topics:
The Concept of Coding
Reinforcement Learning, Biological and Artificial
To listen to (or download) this episode, (right) click
here or use the player below
As always, our jazzy theme music "Quirky Dog" is courtesy of Kevin MacLeod (incompetech.com)
Dear Grace, Josh, Alex, and Yann,
ReplyDeleteThanks for featuring my paper in your podcast. I’m very glad that you liked it, and that you completely understood everything I wanted to say, as well as its context. You also raised some excellent questions. Probably the most important one, I think, is what Grace asked near the end: What do we get from this approach? How does it help us answer the questions we started with?
One answer is similar to the comment (by Josh, I think) that this approach helps us re-evaluate whether the questions we’re asking are really the right ones. An example is “how attention works”. I believe that thinking about sensory overload in Times Square get us going in the wrong direction, and that a better place to start is with the necessity of action selection given basic constraints on behavior such as only being able to do one thing at a time (e.g. winner-take-all in the approach circuit). In a paper with some colleagues, coming soon, we argue that from an evolutionary perspective “attention” is a flawed concept, which should be abandoned in favor of concepts that better capture what’s really going on in the brain (Hommel, Chapman, Cisek, Neyedli, Song & Welsh “No one knows what attention is”). Of course we’re not the first to suggest this...
Ok, telling someone they’re asking the wrong question might seem counterproductive, and maybe even rude. But there’s a constructive side as well, in the form of a research strategy: Instead of focusing on complex functions in modern species, one can focus on specific evolutionary transitions. For example, if you’re interested in the cerebellum, then at least four transitions seem most important (What follows is based on the work of David Bodznick, John Montgomery, Curtis Bell, Naresh Ramnani, and others): First, there is the emergence of cerebellum-like structures in early vertebrates – these cancelled out the consequences of one’s actions on sensory input through common mode suppression and adaptive filtering. Second, in the transition from jawless to jawed vertebrates there’s the elaboration of the cerebellum proper, which introduced climbing fibers – permitting error-based learning of predictive forward models. Next, in early mammals, climbing fibers spread up to the Purkinje dendrites – perhaps conferring additional specificity to the learning rule. And then, in the transition to humans, there was an expansion of those parts of the lateral cerebellum that are interconnected with the expanding prefrontal cortex – perhaps extending predictive forward models to more abstract domains of behavioral control. In each case, you can think of the transition in terms of both anatomy and function. Of course, testing these ideas will still require laboratory experiments of specific cells in specific species, but an evolutionary perspective can help us decide which experiments are most valuable (e.g. for understanding the cerebellum, we need more studies of predictive control in sharks!). But most importantly, the theories the experiments help to develop can be informed by that larger context. As computational neuroscientists, we don’t need to build models of just one behavior in one species – we can (and should) read all the papers that are relevant, and build models in a stage-like process that aims to capture the gradual extension of behavioral functions over evolution along a specific lineage. Many of Stephen Grossberg’s models do this, and I think it’s an approach that can be part of any computational neuroscience research program. For example, for the cerebellum, the stuff I mentioned above does exactly that.
Anyway, thanks again for reading my paper and discussing it with such depth! Now off I go to listen to your episode about Brette’s paper...
Thanks so much, Paul! Glad to hear we did your work justice. Looking forward to reading the attention work.
DeleteBest!
Grace
I recently discovered (in twitter) this podcast. I really really liked this episode. As Paul commented, my fav part was the question at the end: "What do we get from this approach" and the answers (both in the podcast and from Paul here in the comments :)
ReplyDeleteI am very excited to listen to previous episodes! Congrats on a great show and thank you for sharing it!