The discovery that molecular convergence can be widespread in a genome is "bittersweet,” Castoe adds. Biologists building family trees are likely being misled into suggesting that some organisms are closely related because genes and proteins are similar due to convergence, and not because the organisms had a recent common ancestor. No family trees are entirely safe from these misleading effects, Castoe says. “And we currently have no way to deal with this.”
This paper casts doubt on the clear line of evolution of features, clouding the certainty that this or that feature is part of a regular progression.
His nodes are not always intended to represent actual neurons (or bunch of them) but also… simply… concepts. I guess. You know, his original “required nutrient state” as driver of a “hunger” concept, and “nutrient state” as an inhibitor… need not to be actual cells, or sensors. I can conceptually imagine a cell able to detect current nutrient states, and comparing that to an evolution-tuned, internal “requirement level” to decide whether it ought to fire or not. Know what I mean? This whole ‘one plain arrow + one dashed towards same “hunger” node’ part of the schema could just be… one cell.
I understand this could have been quite confusing if thinking about them in terms of neurons.
yes its clear that he talks about concepts. I always try to understand it in terms of neurons but this could lead to problems sometimes as you said.
Thank you a lot for your response. You helped me with understanding this quite a lot:)
Hello everyone. I just joined the forum and wanted to say how excited and flattered I am by all of the attention you’ve paid to my work. There is a lot to read through, here and in the other thread, so it will take me some time to absorb it and formulate some replies. But I will try to reply to the questions raised (at least to those where I think I know the answer) in the near future.
I should admit up-front that I’m not familiar with the HTM theory, so I can’t comment on the comparisons that some of you have made between it and my work. But I will read some of the papers and try to bring myself up to speed. It is always nice to find kindred spirits!
[Same post as on the “affordance competition” thread]
Yes, my diagram was meant to be pretty abstract, so those excitatory and inhibitory arrows are not necessarily neurons (In fact, that whole scheme could be implemented by many unicellular animals). But for ancestral eumetazoans, with that ANS/BNS distinction, there is some specialization. The state control system roughly corresponds to the hypothalamus (or more precisely, its “terminal” or rostral segment), in which the signalling is primarily hormonal. That’s probably the case for the “tegmental neuropile” of lancelets, and of course our hypothalamus does a lot of hormonal control. In fact, chemical sensation might not even need “sensors” because chemicals directly influence metabolism and “downstream” systems as long as you have the right receptors in your membrane.
In any case, in modern animals neuropeptide Y (NPY) does appear to be a big part of a hunger signal, and the shift from locally exploiting to long-range exploring does appear to be governed by dopamine. Strictly speaking, originally it didn’t have to be an explicit competition, but rather a dopamine-dependent influence on how far you move before turning: As dopamine levels go up, you turn more, so you stay more local. Eventually, however, I believe the distinct needs of exploiting versus exploring motivated explicit specialization of different systems within the “peduncular” or caudal segment of the hypothalamus, which expanded into the telencephalon. Within it, the ventrolateral pallium (future piriform, insular, and neo-cortex) specialized for exploiting locally (learning local appetitive and aversive cues), whereas the medial pallium (future hippocampus) specialized for exploration (learning landmarks, odor gradients, etc). So at that point, some of those arrows correspond to neural circuits. And they make predictions - high tonic levels of dopamine should excite the ventrolateral circuits and inhibit the medial circuits.
Indeed, it is sometimes challenging to determine whether shared features are due to common ancestry or whether they are due to convergent evolution. The key is to look at a lot of species whose phylogenetic relationships are known. In the past, phylogenetic relationships were determined by shared features, rendering the whole exercise dangerously circular. But now we have genetic methods that are not only independent of any morphological issues, but can be validated by gigantic data sets (especially by looking at “noncoding” regions of the genome). So we can derive phylogeny first, even with some estimates of timing, directly from the genome, and then only later do we start looking at the features. Again, as we look at more and more species, the question of homology versus convergence becomes easier to determine. That is not to say that pitfalls don’t remain - that’s why we need to be careful.
The case of echolocation is actually a pretty easy example. Since bats and dolphins are very far from each other and very few of the intervening cousins have anything like echolocation, it’s clearly a case of convergence. What is interesting is that the mechanisms that evolved in those independent branches are so similar. This is actually remarkably common, simply because we all live in the same world with the same physics.
Cisekp, I was surprised to learn from you that the neo-cortex is not new. Can you tell us a bit about how it has changed particularly in terms of the number of layers, now six, only five when marine mammals returned to the sea. Thanks your work is uniquely informative.
From that figure alone (with which I’m not familiar, but still…) I would tend to refrain from drawing the conclusion that 4, 5, 6 layers was a timely evolution which somehow “did not get to 6 at the time of cetacea divergence”.
Making more sense is neocortex at the mammalian (not placental, mind you… mammalian. That means kangoos and, wow, echidna) last-common-ancestor.
And from that quite well established common ground, some divergence for cetaceans (even if it somehow seems to “revert” to 5, this does not sound much of a “reversion” to me).
I think it was Jeff in some of the videos, pointing out that there seemed to be “two similar interleaved circuits” among layers of the neocortical sheet. Which, to me, would call for a one-time “oops, replicated that once too many, boss” kind of error.
I think a lot of people are converging to similar ideas, and I hope that’s a sign that we’re all going in a good direction. The idea of behavior as a control loop is actually extremely old. Dewey talked about it 120 years ago, and since then we had Herbert Mead, Maurice Merleau-Ponty, Norbert Wiener, H.R. Ashby, Bill Powers, Humberto Maturana, James Gibson… Those are just the ones I know about, but I bet that precursors to these ideas go further back. (Sometimes I wonder whether Democritus said it in 400 B.C. - that guy seemed to have almost everything right…)
I wonder if they reverted to 5 layers because they sleep one hemisphere at a time. J Fuster mentioned that either L2 or L3 is dedicated to contralateral connections, at least in PFC. So, if the opposite hemisphere always sleeps, contralateral connections may not be needed. Just a guess.
I read the abstract. That is really exciting as it is a close match for my work. It’s consistant with psychology and Sapolsky’s views.
So I try to translate into plain english for social media. I call this a functionally based model grounded in biology.
Note that it is polarized.
The mention of psychology sounded off to me although I think I get the basic point. It’s been a problem.
In reducing this I sometimes combine action with intent. Intent can be viewed at a neurological and also cellular level. I was trying to establish a direct tie to stimulus / response and homeostasis.
This as all as a developer doing fact finding for design research. I can assure you there were very few facts available in the 1980’s.
I have a requirement where I express an opinion now and then do a revision of it having listened to him.
Coming upon this thread a year later, have the folks here had thoughts regarding how to implement these loops from an HTM engineering perspective? The more topics I read under the ‘neuroscience’ and ‘theory’ tags, the more it seems we need far more than just the basic building blocks of HTM (SDRs, SPs, TMs, TPs) to really get anywhere with brain-inspired AGI.
About feedback loops and equilibrium, found these highly underrated videos on YouTube yesterday evening, and crashed through them, starting from oldest to newest. What it tells me is that maybe HTM could be improving how it handles inhibition and the entire feedback process, as well as taking its basic algorithm and creating the other pieces of the brain as they interact and coordinate with neocortex and thalamus.
Certainly, it brought more of the interconnected nature and inhibition into my perspective which had perhaps been missing/lacking previously.