Running with Colin's comments, where ion channels do the real work, synapses are but collections of ion channels, ion channels respond to EM fields, complex synapse behavior is an emergent property of the ion channels that make up the synapses, etc...
In its purest form, an ion channel connects between one limited set of other ion channels (those connected to the same cell) and another less limited set of ion channels (that are part of OTHER cells). Past NN efforts have presumed that neurons are assertions that must be refined, but perhaps this is an emergent property of something more basic? For example, suppose ion channels look at cellular dv/dt (the cellular "noise") and watch for external influences (EM "noise") that tend to temporally precede the noise - in short, "predict" the cellular noise. Where a good influence is found, they would contribute (either positively, or negatively if the correlation is negative) to the cellular voltage - and in the process would tend to move its dv/dt variations to an earlier point in time. As relationships are discovered that are predictive, cells would tend to adopt those relationships to find the most basic of functions that can be computed - leading to understanding of the real-world situations being observed. Design would have to carefully exclude "discovering" relationships that had no temporal lead time, as otherwise ion channels would sense the cells they are part of and the system would lock up. The entire system would optimize, because as a cell that latches onto earlier realization, other cells that haven't done as well would tend to tune themselves up to do better in this constantly evolving "economy" of epiphanies. Something like this ought to work in a dynamic world, but would be DOA in an environment of static tests like NNs are usually tested. The EM field would contain in its "noise" the information of how much ahead of real-time ANY cell is functioning - which would be the thing-to-beat for other cells. The goal of a process control system is to reduce the chaos in a system. For our purposes, let's define "chaos" as "lack of predictability of consequences". A system is "out of control" (in a process control sense) when its medium-term effects can no longer be predicted. In the present model, where it becomes impossible to find leading indicators. Action is called for in these conditions, which becomes "output" for a system. To an outside observer, this would look like a cell that remains quiet until its active inputs go quiet, at which point time it activates. I suspect others here might see other crucial abilities that would be needed for this to work. This all sounds plausible to me. Further, this is analyzable, simulatable, etc., so that development can proceed along an ordinary scientific pathway of theory-design-testing. Note that an actual computer implementation isn't all that much different than NNs, but with the label "synapse" substituted with the label "channel" and the internal operation being based on dynamic dv/dt equivalents rather than absolute values. Note some curious expected operation of the above. Where several ion channels conduct differently-leading EM signals, the gross observation would be that the cell is integrating. Where some ion channels see a negative correlation for less-leading signals, the gross observation would be that the cell is differentiating. Other even more complex functions would spontaneously emerge with this model. Further, there are some strange-looking cells in brains, e.g. glial cells that account for ~90% of all cells. These sorts of things would spontaneously develop as needed with this model. One point of biological confirmation comes from studies of the lobster stomatogastric ganglion that contains <30 cells. These have been completely diagrammed and their operation understood. While these tend to be pretty much identical from lobster to lobster, some have lost a cell somewhere along the way, and have reorganized NOT to simply operate with a cell missing, but instead to do nearly the same overall function but with fewer cells. While this might be a guiding light for AGI methods that ignore biological similes, it shows how finding epiphanies is probably MUCH more computationally intensive than present estimates based on counting synapses - because there are many more ion channels than there are synapses, with each being multiply-connected - at least until it finds its leading indicator. Anyway, let's wring this idea out and see where it leads, as it appears that this approach might crack the whole area of spontaneous discovery and true understanding - and bring the theoretical underpinning Jim was looking for that started this conversation. Thoughts? Steve ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
