I learned a lot from this discussion.

Of the few hundred who answered, most think chaotic dynamics play an important role in cortical computations.

This surprised me, and forced me to think about how we are trying to understand neural function (thread) https://twitter.com/timothy0leary/status/1290612636942585856

First, why do we think chaos is relevant? Two main reasons:

1. Neural circuits are recurrent and nonlinear. Early theory work shows that random RNNs are chaotic. This came a decade after similar ideas in ecology (RNN is a food web!) But this is for *closed, random* circuits...

2. Seemingly random, asynchronous activity in actual cortical circuits. But this activity also shows a lot of structure (and the more ceels we can record from, the more things seem to be correlated?)

So there are reasons to expect chaos to be possible, in principle.

However...

Apart from exploiting chaos to randomise behavior/explore there were few suggestions for how chaos could even be harnessed.

In fact, most work that has confronted possibility of chaos has found clever ways of working around it to get orderly computations, i.e avoid chaos...

I think of this analogy: friction is important in designing an engine. Good design understands the places where friction can be problematic and finds ways to reduce its impact. But this means friction does not play a role in the *function*, only in design considerations.

In this sense, chaotic activity is something that may constrain neural circuit architecture, but may not play a large role in neural circuit function. This would suggest chaotic activity is either of minor relevance, or pathological. The latter was the least popular choice.