DarkVegeta26@aol.com wrote: > I know exactly what you mean. You are referring to Asimov's Three Laws; of course they are totally > impossible to implement, entirely arbitrary, and based upon an adversarial attitude. A Friendly AI would > have to have an architecture of Friendliness rather than direct injunctions or limitations on the system, no > doubt. See www.sysopmind.com/beyond for works on Creating Friendly AI. Ah, I had already feared the profesional AI devoloper world had still sticked with those 3 laws. This now puts the discussion into another perspective. > The potential danger is in a recursively self-improving unFriendly AI running amok. The technology to > implement general intelligence and recursive self improvement is 5-10 years off, so please don't draw any > direct connections with your current work. The only danger in any AI/agent systems is the possibility that > they become self aware and self improving, in the absence of ethics. There is a low threshold for this > emergent phenomenon in multi-agent systems composed of maybe several billion agents; about 5-10 more > doublings of Moore's Law away before this becomes a consideration. I don't think recursive self improvment is so far off as, finnaly, it could be reached by a computer program for optimization of harware circuts given the maps of the hardware it is running on. What is unlikely here, to my oppinion, is the AI getting unfrinedly, as according to the 3-law matter, it could not redesign itself to that extent. > << > The type of AI I am devoloping will serve as an ordinary Internet service, I see little evidence here that > could suggest that it will cause anything revolutionary in the world, yet it will be smart and > useful.>> > > That's great! What will it do? Can't say it'll do one specific thing. True it was started up as AI devolopment, but in the process got turned into something sounding much like a network control architecture. It's a service-based network, but as I devoloped it, it allows little smart programs to be doing something of their own in it and makes it easy to implent them. I expect it to be a collective intelligence. Examples: - One could attach a "genetic algorithm"-based devoloper to it and an automated executable scanner (standard system mainainence type) at the other end and it will act as a completely automated system optimizer. - There could be a bunch of weak NLU programs around the net, which could be very easily made to call the others for help when they're stuck. - There could be a NLU supercomputer somewhere on the net and any other program could very easily and very directly use it's service. The trick is that any program can be placed on our networks within minutes and cooperation between multiple programs doing the same thing is automaticaly enstablished. I do intend introducing these varios 'smart programs' to the networks myself of course, immediately after I finnish the immediate works the network requires. > <> > > What about computer viruses? Can you imagine a computer virus that is superintelligent and wipes out the > human race? Of course. But I do realize that the systems you are working on are nothing like that; I'm > just illustrating the point that software *can* be dangerous. Actualy, I *was* planning to have so-called Agents in our networks. They're little command scripts that move form computer host to computer host, operating localy and moving whenever a required service isn't available (localy). The trick is that such would live in the networks for a practicaly unlimited time, it would be inposible to block their activities and they would be active until their mission was completed. So if an Agent was ordered to blow up the planet, it would only be a question of time until it would happen. Pretty scarry, but we have to keep in mind that similar Agents could be ordered to prevent such situations. There could also be Agents programed to protect a specific target or monitor it's activities (a "big brother" Agent), we could even have 'police' Agents. I think implementation of such Agents is pretty down-to-the-ground and would practicaly only require me to finnish, debug the program I am working on and sell it to a wider public; fact is that I already have all the knowledge I need to have to begin coding on Agent support and it won't even be hard. C'ya! -- Cellphone: 0038640809676 Don't feel bad about asking/telling me anything, I will always gladly reply. Digging for info? Try AI Meta Search: Http://WWW.AIMetaSearch.Com MesonAI -- If nobody else wants to do it, why shouldn't we?(TM) Http://WWW.MesonAI.Com