Tim, Trying to see the forest from under the pine needles on the forest floor is difficult. I wrote a paper, copies of which are on various servers around the Internet, entitled *The Itanium Effect* that seeks to see our present situation from a higher point of view. There are easy orders of magnitude just waiting to be harvested, the most important of which are architectures that are fundamentally fault tolerant. Once on a fault tolerant platform, designs can be made independently of yield considerations, like massive configured-on-the-fly data pipelining, super-wide memory busses, etc.
My paper advances an approach that is not only fault tolerant, but which will continue producing correct results, even in the presence of new faults while computation is proceeding. All this while requiring little additional hardware and being VERY power efficient. Nothing really new is needed for such operation - just a slightly unusual architecture made up of fairly ordinary pieces. Also, as probably the only one on this forum owing their very own analog computer, I have looked at the arguments for analog computation. The issue boils down to the prospective need for bidirectional computation like in electric circuits, where information flows in BOTH directions, e.g. voltage out, and impedance in. This sort of computation is clearly being done in the brain and is probably needed for intelligence, yet is quite difficult for unidirectional computers (e.g. digital computers) to simulate, while bidirectional computing is quite natural for analog methods. For unidirectional computation having clearly delimited inputs and outputs, something new seems to be needed for analog computation, as biological systems can integrate (in the calculus sense) over VERY long times by accumulating particular ions, whereas electronic integration is NOT stable over the long term. Also, as you make analog devices smaller and smaller, they get noisier and noisier, which would seem to limit their application in the world of VLSI ICs. There ARE technologies partially addressing these issues, e.g. current-mode communication, holding capacitors with CMOS switches, etc., but the REAL analog vs. digital decider seems to be the prospective need for bidirectional computation. I have posted about the need for bidirectional computation in the past, but apparently no one here wants to even think about such things, because such a need would pretty much doom AGI on conventional digital computers. I want to look at making digital computers suitable for bidirectional computing with special ALU architectures, etc., which appears to be possible, but again, if you can't do it on a PC (except VERY slowly in simulation), then nearly everyone here simply goes on to the next posting. I think this is a make-or-break issue for AGI, but I am apparently alone in this belief. *Steve* On Tue, Feb 7, 2017 at 8:51 AM, TimTyler <[email protected]> wrote: > Hi. Here's a recent essay relating to machine intelligence. > > It is titled: "Architectural limits to computation": > > http://matchingpennies.com/architectural_limits_to_computation/ > > The topic is hardware limitations and bottlenecks. Three > of these are listed: > > * Lack of parallelism; > * Global synchrony; > * Digitization; > > In the essay, I argue that the demands of machine intelligence > systems is likely to supply pressure to overcome these issues. > > I also point out the the effect of architectural changes to > overcome these issues would be large - and that if they all > happened it would be a spectacular revolution in our ideas > about how to build computers. > > Probably the most controversial point is the third one - > where I weigh in on the issue of whether there will be a > resurgence of analog computing associated with neuromorphic > hardware (or other similar machine learning hardware). > > This seems plausible to me - but the larger issue of whether > most future computers will be digital, analog or hybrid > designs seems unclear. > > -- > __________ > |im Tyler http://timtyler.org/ > > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/10443978-6f4c28ac > Modify Your Subscription: https://www.listbox.com/member > /?& > Powered by Listbox: http://www.listbox.com > -- Full employment can be had with the stoke of a pen. Simply institute a six hour workday. That will easily create enough new jobs to bring back full employment. ------------------------------------------- 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
