Someone asked on my latest video:
Is bursting related to the p300?
No. I believe p300 is a large scale EEG signal related to decision making. The p300 signal is an empirical result.
Bursting in a minicolumn is a local effect, meaning it occurs on a minicolumn by minicolumn basis. It is not related to decision making.Instead it reflects an unpredicted input. It is a proposed mechanism that solve a theoretical need, but was also suggested by several empirical observations.
I asked it!
But, what if we had a great prediction mismatch?
Wouldn’t it produce bursting in many minicolumns at the same time?
It could produce bursting in different regions in different columns as a chain reaction or massive event.
Bursting in a minicolumn activates all the plausible hypothesis given a SP input, without previous context after all.
Those initial hypothesis disambiguate step by step as more sequential information arrives.
Couldn’t this be a mechanism which help decision making?
Sure. A significant mismatch draws your attention to the mismatch signal and this could have many downstream effects, such as help with decision making. But just looking at the HTM sequence memory, there is no decision making going on. As you point out, bursting is equivalent to forming a union of hypotheses consistent with the input.The sequence memory tries to narrow down this union of hypotheses with each subsequent input.
There is a nice review by Peter Hagoort (Reflections on the Neurobiology of Syntax)
In this document he draws a set of hypothesis about the role of certain regions of cortex -especially Broca’s area- with the focus put on event-related potentials (ERPs).
From my point of view, the hypotheses drawn in this review are consistent with your approach and your book On Intelligence.
Isn’t in some respects forming a union of hypothesizes making a decision? Because you’re narrowing your options by asking a better question…
Could it be that perhaps more generally, ERPs are measuring bursting behavior, especially since it’s usually measured over repeated trials, which would average out other brain activity due to the sparse data structure?