The design space of the future for AGI is the many specialized AGI's running on many computers and the feedback from those being functionally interwoven into a new and better AGI.
Dan G ---------------------------------------------------- >From : J.Andrew Rogers <[EMAIL PROTECTED]> To : agi@v2.listbox.com Subject : Re: [agi] an AGI by Minsky and Singh Date : Sun, 12 Jun 2005 10:07:39 -0700 > > On Jun 12, 2005, at 2:02 AM, Ben Goertzel wrote: > > But, your assertion that any competently articulated, competently led > > AGI project should be able to fairly easily raise $5M in venture > > funding is *also* based on a basket of assumptions, which you didn't > > make explicit in your message! > > > Yes, very true. There is no such thing as a context-free opinion. :-) > > However, I also used non-AI development and implementation metrics that > would apply to AI development and implementation to build my > assumption. The "new" part of AGI development is a new design space > from a computer science perspective, but the fundamental mechanics of > implementation of a new design space will not be that different and a > lot of the ancillary stuff is well-described. If it takes a long time > to implement, it will be because parts of the AGI design are poorly > described such that no one knows if/how they will work. > > The only potential money sink that I consider plausible is very large > and exotic hardware, but the necessity of this does not seem apparent > to many people actually working on it. High-end vanilla hardware seems > to be what most people require. > > > > And my suggestion is that the path from here to AGI is almost > > inevitably going to involve a few years of research-oriented > > engineering/experimentation prior to any period of more deterministic > > product-development-like engineering/tuning. > > > The differences in opinion seem to revolve around whether or not useful > products can be spun off the main technology track as the technology is > developed. While I would agree that it can be a diversion of sorts, a > carefully selected mezzanine product target should be reasonably > doable. How feasible this actually is is a function of the > architecture and design to a great extent. > > > > I'm not really sure how you're defining these terms, in this context. > > In terms of creating AGI, as far as I'm concerned, even if you're in > > "late stage development" of your *software system*, until you've > > demonstrated robust human-level AGI behaviors, you're still doing > > speculative research.... This is only fair given the demonstrated > > difficulty of the AGI problem. I apply this to my own work as well > > as yours and anyone else's.... > > > I was referring to the ability to demonstrate robust AGI-ish behaviors > in implementation. It does not have to be a completely implemented or > solved system if one can demonstrate genuinely new capabilities -- this > will have intrinsic business value AGI or not. > > In other words, the pitch should be no less than "we can deliver this > wicked coolness *right now*, and with some additional funding we can > greatly extend the envelope to more wicked coolness". The problem is > that the initial demonstration of "wicked coolness" has to be a clear > differentiator from other half-baked AI ideas, most of which claim to > show some type of vague novelty very early on. It is not easy. > > > > And, VC's criteria for "indistinguishability" in this context are > > generally quite crude... > > > Heh, yes. The problem of education is very real and there is > relatively little one can do about this. Hence the value of having a > bright shiny object for them to fixate on immediately. > > Very few people grok the current theory space (which is somewhat > independent of personal theoretical biases), and unlike nanotech, the > field is neither straightforward or obvious from basic principles that > everyone understands. For almost everyone, it really *is* a crap > shoot. > > cheers, > > j. andrew rogers > > ------- > To unsubscribe, change your address, or temporarily deactivate your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
