Thank you Steve. Telmo, before starting to analyze brain recordings ten years ago, I would have fully reinforced your view regarding Turing-complete machine and brain like simulations.
Here is the main problem, we have two distinct categories: a.Scientists whom imagine how the brain works without having direct knowledge of the real brain, they build models, AI theories or simulate "the brain" b. Scientists whom record, analyze data, build brain computer interfaces and try to provide therapy. Even their “access” to the real brain data is highly limited. Between (a) and (b) it is a huge gap. Building a hybrid model as the first step can change this entire trend which started 60 years ago, a fast access to the “real brain” data will be granted for anyone interested to build models, AI theories or simulate the brain 1. We need a charismatic leader(s) and a board of directors to shape the entire strategy build rapidly the infrastructure of the institute. 2.We need reliable ways to facilitate international collaboration, no one whom likes to do research in AI, DL, AGI,...software/hardware development should be left behind. Any idea how to move fast with (1) and (2)? Dorian On Fri, May 15, 2015 at 7:16 AM, Steve Richfield <[email protected]> wrote: > Telmo, Colin, and Dorian, > > On Fri, May 15, 2015 at 1:34 AM, Telmo Menezes <[email protected]> > wrote: > >> Hi Dorian, >> >> On Thu, May 14, 2015 at 7:17 PM, Dorian Aur <[email protected]> wrote: >> >>> >>> Here is the rationale: >>> a.Why use a digital computer to simulate/map or emulate the whole brain >>> • It cannot express all forms of computation that are built within >>> biological structure (see neuroelectrodynamics); >>> >> >> All computable functions can be computed by any Turing-complete machine. >> It is not possible for a biological substrate to perform computations that >> are impossible to a digital computer. >> > > THIS particular point is stupid for BOTH sides to argue. It weakens BOTH > sides, and leads to needless stupid divisions of efforts and organizations. > > Simplistic "Turing-complete" machines have some SERIOUS utility issues for > many applications, probably including AGI. Sure they can in theory simulate > anything, and in so doing they can compute anything, but there are VAST > areas where they slow to a standstill when faced with real-world-sized > problems, because for many computations the time they take to compute grows > considerably FASTER than the complexity of the systems they are emulating. > For example, simulating electric circuits with programs like SPICE, > execution time grows as n log n on ALL present-day computers except > analog computers, and for us, n is REALLY BIG. Further, if Colin's EM > field theory is correct (I think it is), growth in execution time may be as > bad as n^2, with n representing ion channels rather than neurons, so n would > be really astronomical. This would represent a computational brick wall, > probably exceeding the efforts of civilization to EVER surmount, so any > successful development would have to sidestep such insanity. > > I suspect that Colin, in seeing a situation where everything affects > everything, thinks it cannot be simulated or otherwise computed on a > conventional computer. For small-scale research purposes I suspect this is > simply a reflection of Colin's disconnection with simulation technology. > However, for large-scale AGI-like systems, *Colin is more probably right > than wrong* that simplistic "Turing-complete" computers are and *probably > always will be inadequate for the task*. > > No, this doesn't mean that we can't build AGI's, it just means they won't > have Intel processors in them, because there ARE prospective architectures > that can compute things that are WAY beyond what any imaginable extension > of a PC can do. > > The apparent reason why there isn't presently much effort developing such > systems is that they won't run existing benchmarks any faster. What we need > are better benchmarks, like the ability to simulate large circuits, rather > than faster computers to run present-day stupid little benchmarks that are > COMPLETELY disconnected from AGI requirements. > > To illustrate one prospective architecture, a large FPGA-like chip that > wired up programmable analog components instead of simplistic digital > components could be used as a processor that takes a BIG bite into such > applications. Development cost estimates run into the hundreds of millions > of dollars, so start collecting your lunch money. > > So everyone, unless you can somehow argue that there is any good reason to > even discuss this particular point BEFORE we know MUCH more than we now > know about how a *successful* AGI might function internally, let's all > agree on sane behavior and stop arguing this particular point where BOTH > sides are more probably wrong than right, at least at some level. > > Steve > =========== > >> computers are already conscious. Maybe some people are not. How would you >> know? >> >> >>> • No reliable model for brain diseases. >>> b. Why not shape a biological structure, connect it with a digital >>> computer use machine learning (e.g DL) and perform all kinds of >>> computations - Can we build a conscious machine >>> http://arxiv.org/abs/1411.5224. >>> • Naturally, emotion, consciousness ....are expressed >>> >> >> How? What is special about biological substrates that is required to >> generate emotions or consciousness? Do you have any theory? >> >> >>> • Can be used as a model for therapy for about 600 brain diseases >>> • Can be connected to a laptop, iPhone uses digital and biological >>> computation together which can make any digital computer highly interactive >>> • Far less amount of funding required. AGI can become fast an academic >>> discipline, it can attract funding not only from private companies >>> >>> My previous answers on FB >>> >>> 5.Does an AGI need to be conscious? >>> Yes, it has to be conscious otherwise AGI can be dangerous (see 9). >>> >> >> Conscious natural intelligences can be extremely dangerous. The danger of >> AGI is related to our inability to understand the consequences of providing >> an utility function to a super-intelligence (since we are not intelligent, >> we cannot foresee which steps the AGI might take to increase utility). >> There is also the danger of bootstrapping and evolutionary process, then >> AGIs can start to compete for our resources, etc. >> >> >>> 6.Can AGI be creative? >>> >>> If we build hybrid systems AGI can become creative >>> >> >> By what mechanism? >> >> >>> 7.Will AGI have emotions? >>> >>> Biological structure embedded in the hybrid system will allow any AGI >>> system to experience emotions >>> >> >> This is magical thinking. >> >> >>> 8.How far off is AGI? >>> With current technology the first prototype can be implemented in less >>> than 5 years, far less than the BIG detour (2001 - 2015) >>> >>> 9.Will AGI be dangerous? >>> >>> The system needs to be conscious about its actions, otherwise it can be >>> dangerous >>> An example : the missile crisis in Cuba, less intelligent actions can >>> lead to an apocalypse for everyone ( it should be embedded in consciousness) >>> >> >> What about the ISIS people? Are they conscious? What prevents an entity >> from being conscious but having different values from yours? >> >> Best, >> Telmo. >> >> >>> It's time for action >>> >>> >>> Best, >>> >>> Dorian >>> >>> >>> >>> *Note:* EM interaction establishes communication in case of a more >>> powerful form of computation, five years ago we call it - >>> neuroelectrodynamics. A classical model or a quantum model can be used >>> to describe a natural phenomenon, they are our models . Almost >>> everything can be approximated ,simulated on digital computers only if one >>> has the algorithm. The simulation in this case requires a huge cost, it is >>> highly inefficient and in addition many characteristics developed within >>> biological structure are completely lost. Current trend in AGI can continue >>> another 5-10 years however a general loss of credibility will follow - a >>> less "intelligent" path. Saving the AI/AGI idea should be a priority, we >>> do have the technology to keep alive, grow and "connect " neurons and any >>> already developed algorithm (e.g AI algorithm) can be used since the >>> digital computer will be an important part of the hybrid system. >>> >>> >>> >>> >>> >>> >>> On Thu, May 14, 2015 at 7:27 AM, Steve Richfield < >>> [email protected]> wrote: >>> >>>> Ben, >>>> >>>> I don't know what Alan's problem is, but it appears he doesn't >>>> understand forums in general, and this forum in particular. >>>> >>>> As Alan's first objection to threads that has been running for several >>>> days, Alan rises up to request that the subject be killed!!! This is >>>> absurd. >>>> >>>> The whole purpose of threads is for people to follow the ones they are >>>> interested in, while ignoring the others. Apparently Alan is unable to >>>> participate in this very simple process. >>>> >>>> The bases for Alan's request are also absurd as explained below. >>>> >>>> On Thu, May 14, 2015 at 6:07 AM, Alan Grimes <[email protected]> >>>> wrote: >>>> >>>>> I'm about three days away from formally requesting a killthread on this >>>>> EM fields crap. >>>>> >>>> >>>> Ben, you might want to think about moderating Alan. >>>> >>>>> >>>>> 1. Electromagnetism has been Well Understood (tm) for about 140 years >>>>> now. >>>>> >>>> >>>> So what. This doesn't seem to be an issue. >>>> >>>> >>>>> 2. By [nearly] all accounts, EM fields in the brain are secondary to >>>>> its >>>>> operation. >>>>> >>>> >>>> What accounts? >>>> >>>> >>>>> 3. Neural Science is a well established field that runs parallel to AGI >>>>> and, yes, they do VERY careful science. >>>>> >>>> >>>> You obviously have never worked in a neuroscience lab. However, others >>>> in this discussion, including myself, HAVE worked in these labs and know >>>> the severe limitations of what people think they know about how neurons >>>> work. >>>> >>>> >>>>> 4. AGI is not, formally, a science. >>>> >>>> >>>> I can't speak for the others here, but I suspect that most people here >>>> agree, but believe that it should become a science once we know enough to >>>> talk about the prospective internals of an AGI system. >>>> >>>> >>>>> It is a branch of engineering. >>>>> >>>> >>>> B.S. If this were true, computers would have been thinking for decades >>>> by now. There is presently NO recognizable science supporting AGI. AGI has >>>> yet to rise to being science, let alone rising to be engineering based on >>>> science. >>>> >>>> 5. In the interests of getting things done, simplifications have to be >>>>> made wherever possible. >>>>> >>>> >>>> So what? This doesn't seem to be an issue. The issue here is >>>> determining what is essential, and what can be "simplified". There are many >>>> opinions, including yours, none of which have significant evidence to >>>> support them. >>>> >>>> >>>>> 6. We are not trying to simulate a brain, we are trying to identify >>>>> what >>>>> characteristics are actually required to create a thinking machine. >>>>> >>>> >>>> Agreed. So what? >>>> >>>> >>>>> 7. The standard of evidence, at this point, to indicate some kind of >>>>> non-Turing computation is required to produce thinking is >>>>> extraordinarily high at this point. >>>>> >>>> >>>> "Turing computation" isn't really a well defined term, e.g. does it >>>> include analog computation? >>>> >>>> I have posted in the past regarding the potential need for >>>> bidirectional computation in AGI, which can be simulated on Turing systems >>>> with a loss in speed which is proportional to the logarithm of system size. >>>> If bidirectional computation proves to be needed, than Turing systems may >>>> indeed NOT be up to AGI. Fortunately there are non-Turing approaches to >>>> bidirectional computing. >>>> >>>> Note that Colin's proposal also includes bidirectional computing, >>>> though we haven't yet discussed that. >>>> >>>> There is a pretty strong case for bidirectional computing, so don't >>>> clutch your Turing machine too closely. >>>> >>>> 8. Once AGI is created it is highly probable that it could be further >>>>> enhanced by means of mystical physics, ie quantum fields, and stuff, >>>>> but >>>>> right now it's only a distraction. >>>>> >>>> >>>> ONLY if "mystical physics" proves to be unnecessary. I have seen NO >>>> hard evidence either way. >>>> >>>> >>>>> 9. The brain may indeed utilize mystical physics to some extent, we >>>>> should be extremely cautious about brain emulation, even if you want to >>>>> stick your head in the sand about the identity issue. >>>>> >>>> >>>> We are a loooooong way from brain emulation, but it would sure be nice >>>> to be able to emulate a single neuron that can do ALL of the things our own >>>> neurons do - fast learning, abandoning useless functions, reducing power >>>> demands for slow/rare phenomena, etc. - all things that an AGI will also >>>> have to do. >>>> >>>> 10. I have some pretty strong hypothesii about how the brain works but >>>>> I'm frustrated by my inability to test those hypothesii for lack of a >>>>> simulation environment or a robot. >>>> >>>> >>>> Join the club. Oh, I see you already have. >>>> >>>> I don't have either. I have been >>>>> stuck at this state of not having a testing platform for ten years >>>> >>>> >>>> Only ten years? I can see you are a newbie at this. I have had this >>>> same frustration for >40 years. >>>> >>>> >>>>> and >>>>> it's driving me nuts!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! >>>> >>>> >>>> THIS explains a LOT ... B-:D> >>>> >>>> >>>>> (this is what >>>>> my minecraft post was about...) I saved up about $12,000 out of a >>>>> required $16,000 to get a Nao but then I've been unemployed for three >>>>> years and have no job prospects in this awful economy. =((((((( >>>>> 11. Meanwhile, I have not been chewing up list bandwidth talking about >>>>> how great my untested theory is or spending much time deriding other >>>>> list participants. >>>>> >>>> >>>> There ARE other paths, e.g. invent something relating to AGI, get a >>>> patent, find someone to promote your invention, find a VC, start a company, >>>> etc. >>>> >>>>> >>>>> BTW: Hplus-talk mailing list seems to be down and the admin forwarder >>>>> is >>>>> down too. >>>>> >>>>> IQ is a measure of how stupid you feel. >>>>> >>>> >>>> Aha, you very obviously do NOT feel stupid at all here, so, by your own >>>> measure, your IQ must be VERY low. >>>> >>>> OK, sorry (but not very sorry) I beat you up here, but understand that >>>> it is often quite difficult to examine possibilities that violate your >>>> world model, which is obviously your difficulty here. Just because >>>> something is obviously "crap" doesn't mean that it is crap. If you can't >>>> deal with this, then stand aside for others here who CAN deal with it. >>>> >>>> Steve >>>> >>>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>>> <https://www.listbox.com/member/archive/rss/303/17795807-366cfa2a> | >>>> Modify <https://www.listbox.com/member/?&;> Your Subscription >>>> <http://www.listbox.com> >>>> >>> >>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>> <https://www.listbox.com/member/archive/rss/303/25129130-ee4f7d55> | >>> Modify <https://www.listbox.com/member/?&;> Your Subscription >>> <http://www.listbox.com> >>> >> >> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >> <https://www.listbox.com/member/archive/rss/303/10443978-6f4c28ac> | >> Modify <https://www.listbox.com/member/?&;> Your Subscription >> <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> > <https://www.listbox.com/member/archive/rss/303/17795807-366cfa2a> | > Modify > <https://www.listbox.com/member/?&;> > Your Subscription <http://www.listbox.com> > ------------------------------------------- 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
