I guess I’m the latest arrival to this excellent website. I ‘found’ it via a New York Times article and it’s been the most depressing and exhilarating hour of browsing since. Exhilarating because you guys at Numenta are in an exclusive group who are almost certainly getting it right (!); and profoundly depressing because I’ll never catch up!
A bit of personal background.
I’m a chemist by background (Oxford 1975 – 79) but I gave up being a proper chemist in 1979 and have been in the finance/venture capital world for nearly 40 years now. However, for the last 9, maybe 10, years I’ve been marginally obsessed by how real brains think in the sense of processing sensory information and then acting. The ‘obsession’ actually goes back further; it started at about 6am on a wet November morning in 1976 when I had to get up out of a warm bred to prepare punched cards for a Fortran program which calculated frequencies for the vibrational modes for CO2; my over-riding thought at the time was that there had to be some better way to model systems than punched cards on a digital computer. It’s what a cold, wet November morning does to a young undergraduate I suppose, especially as he’s leaving the girlfriend de jour behind in the bed.
At the time I put a bit of preliminary thought into non-binary logic and an oscillatory wave approach to information processing and got absolutely nowhere; I’d done a bit of quantum theory but the maths requirements, especially Fourier manipulations, were too difficult. After university, I gave very little thought to alternative logic systems and over the next 30 years raised a family and got on with a career. Then in 2008 there was quite a bit of speculation in the popular science magazines about next generation AI and I started, in a clumsy way, thinking about thought processes again.
I’m still not sure I’ve got anywhere! But back in 2009 I did a calculation of the sort that only chemists do which was estimating the atomic weight of the average neuron from which I deducted the very approximate contribution from water and lipids and so got to an estimated number of very big complex protein molecules - possibly as much as 200,000 per cell as an average over all neurone and glial cells. It begged the question of what (as in WTF) were they all doing because the electrochemical switching duties of a well-connected pyramidal cell are probably quite significant and the big proteins surely have to be involved. Further, the chemical environment such a cell lives in might include several 10’s of types of neurotransmitter delivered to it at various times and under varying circumstances. Definitely not digital then.
Then in 2012 I read Professor Dennis Bray’s excellent book ‘Wetware’ and realised that the concept of the brain cell (of whatever type) having quite a sophisticated, internal information processing role was beyond doubt. Obvious really because, after all, amoebas can ‘learn’ in a basic way.
Where I’ve got to in my amateurish way is that a brain is essentially a complex community of specialist amoeba-like cells.Their interactions once in place in the matrix are incredibly subtle and the net result is possibly not even a Turing machine…? I tentatively think they assemble into a co-operative matrix through a series of iterative trial-and-error cycles and fundamental to this is the feedback processes which eventually result in a Darwinian elimination of the ‘wrong’ links. It’s a long process (years) for the sophisticated brain to evolve in this way but I think it’s the dynamic process underpinning the design of an efficient architecture for decision making by the organism which owns the brain.
Anyway, I’m glad I found you. Please let me know if any of the team are coming to London, or anywhere in the UK, to do a bit of proselyting/lecturing!