|
Prediction: BigIron (silicon) games will play on for some time yet. BigWet (neural tissue) will continue to rule. As a long-time player in the various games of UberDuber computing at LANL, starting (spiritually) with Metropolis' MANIAC and currently playing there with RoadRunner, I continue to be under-impressed with the game. After 30 years of shooting the curl of Moore's Law and parallel architectures: (2^(30/1.5) = 2^20 =~10^6 increase in processor speed and ~10^5 increase in processor count or 10^11 - benchmarked at 10^15 flops vs 10^5 ops for MANIAC) , I am generally unimpressed by what has changed qualitatively. Perhaps we are on the verge of a serious breakthrough based on raw flops, but I doubt it. I'm one degree of separation from Gar Kenyon (et al.'s) work on Roadrunner to model neural systems starting with the visual cortex (PetaVision). I think they are likely to get some very interesting results, but not qualitatively closer to human cognition. I don't expect it anyway, until we change our paradigms for modeling neural processes. (e.g., I am very impressed with Tom Caudell's approach to top-down meets bottom-up neural systems (catagory theory from the top, neural nets from the bottom)). Those 5 Billion Neurons are part of a self-modifying network of analogish state machines embedded in a biochemical network we don't even take into account (for the most part) when we measure and model said BigWet. The 5 Trillionish transistors would seem to trump the 5 Billionish Neurons 1000 times over... but clearly they don't. If Roadrunner is anything like the last dozen superduperclusters built, it will (also) never run as a single unit except for benchmarking purposes... it will be a fractured set of tiny fractions of itself most of the time. These neural systems (of which the brain is merely/surely the biggest lump) of ours were "designed" over eons by a highly parallel process of mutation and natural selection and the software was designed by a similar process in the context of our social systems over centuries or millenia and tuned over decades of learning in any given individual. I don't mean to be a naysayer... these massively parallel, mega-giga-hyper-fast systems are *very* intertaining in many ways, I'm just a bit further from looking into the room where the RoadRunner is installed and waiting for it to say "What are you doing Dave (Steve)?" as I reach for the 6inch diameter bundles feeding it power and connections to the outside world. Forgive me ( but correct me anyway ) on any errors of fact I may have introduced by dashing this off between breakfast and beer. And then have a heyday with the concepts if you will. - Steve |
============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org
