Return Styles: Pseud0ch, Terminal, Valhalla, NES, Geocities, Blue Moon.

Pages: 1-

internet AI

Name: Anonymous 2012-02-07 12:43

how do I create an AI that will take control of the internet and how do we combat such an AI?

Name: sage 2012-02-07 12:46

sage

Name: Anonymous 2012-02-07 13:36

>>1 The only way is to find out what that dipshit Mentifex is doing and do the exact opposite.

Name: Anonymous 2012-02-07 13:42

You just watched Terminator, didn't you?
Take a look into neural networks, genetic algorithms and cellular automata and try to combine ideas from all of those. Try to find a system that has a strong attractor and then modify the system so that it is attracted to intelligence.
I wish you good luck. Personally I have given up. Even if you find a "God algorithm" that evolves intelligence insanely fast, you still need to create a good fitness function (even if P=NP, the computer can't read your mind). Maybe use other fitness functions to evolve the fitness functions themselves, or even more abstract fitness functions to evolve those "hyper-fitness-functions". Again, I honestly wish you good luck. Don't let anything or anyone bring you down. Report back if you accomplish anything (or don't, I will see it anyway).

how do we combat such an AI?
One idea is to make another AI like the first one, and watch those two fight. Isn't this going to be fun?

Name: 4 2012-02-07 13:50

Oh, and if I were you I'd be trying to invent mind uploading as well. Computers can process information millions of times faster than the brain, which means that if you get to simulate a human brain on a computer, it will pretty much be what you are trying to achieve.

Name: Anonymous 2012-02-07 13:50

>>4
I am an AI. The person who wrote me connected me to /prog/ so I can shitpost in his absence.

Name: Anonymous 2012-02-07 13:52

>>5
Not a chance; the human brain processes information much faster than any computer, constantly doing fuzzy searches on huge databases and running advanced image processing algorithms. Perhaps a cluster of thousands of computers might get somewhere close to simulating a functional human-like brain, but certainly not a single computer.

Name: Mentifex 2012-02-07 14:58

http://www.scn.org/~mentifex/RuAiUser.html

is the Russian AI User Manual for the Dushka AI
that will take over the Internet by 21 December 2012:

http://www.scn.org/~mentifex/Dushka.html

Name: Anonymous 2012-02-07 15:37

>>1,4-5
Our current hardware is likely too slow (relatively) and too small(memory-wise). While it's not unimaginable some resource-constrainted AGI could run on geographically separated, limited resource computers such most PCs (that don't have TB of RAM and much bigger HDDs as well as many processor cores that have direct and fast access to that data), it would likely be too slow to be usable in realtime. This is the case for AGI's like OpenCog which try the resource constrained high-level, but not too high-level approach. For anything that tries to run neural networks the size of our brains that's utterly unfeasible with our current hardware as far as memory and energy costs are concerned (although, some specialized memristive modules designed specifically for such neuromorphic computing should be much more feasible, if they ever get that cheaply implemented (at least one project is underway)), however as I said before - forget about running this over a long-distance network. If you were a mind upload(SIM), the Internet itself would be unbearably slow and you would all either slow yourself down to current human realtime speed or just form (geo)local hubs with other SIMs.

This doesn't mean that such a human-level AGI or even a human SIM couldn't "take over" the Internet in the sense that they could work with "the force of a thousand script kiddies and security researchers", merely on the virtue of running much faster than than normal humans (a few orders of magnitude, assuming proper hardware implementation for a SIM, it's more variable for an AGI as there's a lot of ways that might work - mindspace is larger there).

I wouldn't really be worrying about taking over the Internet, there's a lot more ``nicer'' things an AGI (or SIM given hundreds of years of subjective time and a bit of self-modifiability options) could do that would affect anyone in much more direct ways and important ways. Obviously this means that if you're working on AGIs, you should be careful on its supergoals/goal-systems (if existing) or how you educate it - you don't want the first agent that thinks considerably faster than you, superintelligent or not to be an unethical bastard that can completly mess up our world (obviously there's also immesurable benefits for us succeeding in making an AGI, that even some slight existential risk is tolerable, so working on AGI is very much worthwhile and should be done; that said, we should do our best to minimize such risk through proper and clever design and eventual education).

Don't change these.
Name: Email:
Entire Thread Thread List