On Sunday 15 January 2006 04:41 am, [EMAIL PROTECTED] wrote: > Searching is a part of AI... But is not deep logic like Chess... > Is IBM Deep Blue just a look up machine or really perceiving and logical > reasoning with an output of action.. the next move. Deep Blue, the Chess Expert, was purely an expert system. The tour-de-force in that was the hardware it was running on. Blue Gene is the current successor. (Remember, IBM started as a hardware company. Software will be sort of secondary, even when given LOTS of thought and polish.)
I'm expecting IBM to keep pushing the hardware, with explosively spectacular software to show just what it's latest hardware can do. You can't do proteosome matching with a simple search, that requires 4-D modeling of molecular strains. (You need to predict the ways a particular collection of amino acids will fold...and what catalysts can change the folding patterns.) This is a complex job requiring both searching and modeling skills. (Those are two of the skills that I see as basic to an AI.) When you have those two skills operating in a domain a complex as protein folding, then it becomes natural to work on adding in a goal-seeking component. That gives you three of the basic four skills. The fourth one is trickier, and I don't even know quite how to talk about it. It involves judging how "good" something is. In every category that I'm aware of, those creating things that I think might develop into AIs, but who aren't intentionally setting out to create an AI, are creating a module that could form a basis of a mental substrate. Thus Google is working around a searching and pattern matching approach. Well, that's where they start, and no, that alone won't be a full AI. But once that have that working, they have a useful tool that can be extended by adding other capabilities. And that is to me one of the most probable ways for an AI to emerge. It can also be a rather dangerous way, as if you aren't intending to get an AI, you may well not consider some consequences of some of your basic decisions. At some point you will just have added enough capabilities that it will "wake up"...and it's motives will be whatever you have happened to cause them to me. In Google's case it's likely to be answering questions...with a bias towards searching for previous solutions. Not a particularly bad choice...but not exactly a Friendly AI, either. But do notice that it could equally be a company with a goal of "make our company famous and successful", which could be disastrous, depending on just how they define "famous" and "successful". At a minimum that goal would seem to entail causing the AI to seize control of the company. In my mental model, any of these four skills can be the bedrock onto which the others are added. Then you work on various elaborations and mixtures of those skills to arrive at the complex that I see as a general intelligence. Note that where ever you start, that will form a bias around the way that you store your data. (I.e., you need a uniform way to access your memories, and the access mechanism will be highly dependant on the basic reasoning module...because that's how your first memories are laid down.) > ... > I do believe the Financial and Economics expert AI system will be of > enormous value and one of the best performance test of any AI System. An expert system can serve as a base, I suppose, but a more generalized physical model would be a better choice. As I understand it, expert systems are fragile and don't expand well beyond their initial domain. (Then again, don't assume that the financial system can be reasonably predicted. Certainly it's possible that it can, but certain experiments that have challenged that assumption haven't done too badly. See "A Random Walk through Wall Street". It's possible that with our current system, no individual can predictably do much better than chance. Certainly being a successful trader for one year isn't a good predictor of being a successful trader the next. Of course, it's possible to do much worse that average...) >... > > Dan Goe ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
