Hi Guys, Before I respond directly to John's questions let me first state briefly that I do not represent Numenta in any capacity, either officially or unofficially; I'm simply a hobbyist attempting to do his part to contribute to the acquisition of something I feel has tremendous potential for human beings and is one of the single most important developments humankind can pursue. I think the pursuit of machine intelligence will result in the eventual development of many "cures" for humanity's current condition be they social, economic, medicinal, technological or otherwise. As such, I support any and all efforts toward the achievement of this goal - however I **do** agree with Jeff Hawkins when he says (and I paraphrase) that HTM technology and other technologies which seek to "uncover" the "roadmap" toward intelligence that nature has already provided us; will most likely be the shortest route to achieving that goal.
I do not however think that arguments over which way is the "best" way to get there will serve us in the end - except to realize that all of these conversations take place in an economic climate that can (and has before) had an impact on the momentum with which we achieve this goal (the climate, that is). The quality of that climate and the robustness with which it either offers assistance or dilutes or derails progress is the only aspect of that particular conversation that interests me. I do not think that one can derive truth from endless debate over a topic; as if truth will eventually emerge from the discussion and illuminate us all. Consensus derived truth, or what we all "agree" provides truth or is truthful - doesn't really interest me because the truth will be what it is - what eventual comes to be whether any of us agrees with it, or not; and that's fine with me. Now let me attempt to answer John's question(s). For the first time in human history, we have an algorithm which > models activity in the neocortex and performs with true intelligence > exactly **how** the brain does it (its the HOW that is truly important > here). ...and by the way, this was also contributed by Jeff Hawkins > and Numenta. > "performs with true intelligence" is a pretty bold claim. Yeah I said it. :-) Is it true? Well the answer I have in my head is, who the hell cares? :-P Will its truth have any long lasting impact on forwarding the action toward or away from our goal? My hunch is no... :P However I will try and explain the basis upon which I made this comment. If we define one of the characteristics of intelligence as being the ability for a technique to be broadly applied to multiple classes or kinds of problems without alteration to its basic algorithm, then I feel HTM technology is the closest thing to it (so far, and so far as I know). I think this particular feature is **one** of the primary features that distinguishes it among its contemporary algorithms. Another feature is its adherence to biological plausibility. We must understand that HTM technology is just beginning and doesn't yet look like what it will eventually look like once it has satisfactorily "matured" or "evolved". Numenta's mandate (as far as I can recall) is to learn as much about the neocortex (which represents 70 or 80% of the mass of the human brain [that is an approximation] ), and build intelligent systems, techniques and algorithms from that research. There are still a lot of "parts" to be developed. Does it (right now) represent Artificial General Intelligence? No. I don't even think that is Numenta's goal at all. By "true intelligence", I meant - functions in the way a particular layer in our brain (2/3) functions, and can do useful intelligent things without thousands of cycles of training or altering its basic core to suit one-off problems. And by "true intelligence" I mean that the basic algorithm's operation can be corroborated by actual physical analogs in the human brain. That it "learns" by processing sequences of sparsely distributed patterns and sequences without any explicit storage of symbols or the use of any exotic mathematics or mechanisms which don't exist in the brain. Yes, there is "swarming" which is a technique to arrive at the most optimal configurations for a given dataset and goal, but it doesn't change the basic algorithm which can be broadly applied to many different problems. They're not using one algorithm for this class of problem and another algorithm (with thousands of cycles of training) for that class of problem. I do not think that HTM technology right now is the "end of the road"; they're just getting started. They made a significant discovery with the CLA (Cortical Learning Algorithm); saw it had some commercial potential; and produced some very useful and eye-opening products from it. As far as I know, this is what they will continue to do and I am over joyed that they are here doing it with an unprecedented openness and transparency, which I think is very inspiring! Cheers, David
