Turing meet Darwin, Darwin meet Turing

Started by
34 comments, last by capn_midnight 21 years, 2 months ago
How about using 2 bots. One as a teacher and one as a student.

They both have a goal. Lets say one bot is the head of a four legged robot. And the other bot controls the legs. The goal is to move across a room.

The Head issues signals in the form of sounds ie. BEEP. When the Legs hear the BEEP he tries to move something. If the head is statisfied with it (ie, he got closer to his goal) he gives the Legs a reward. If the legs happens to make them move the wrong way the head punish the legs.

This means there is a controller part of the AI and a worker. They both interact and should develope a language. None of the parts should know what to say before they start. The head just tries different series of beeps in different frequencies and the legs tries doing different things depending on the input. The head knows the goal and can issue punishments and rewards.
Domine non secundum peccata nostra facias nobis
Advertisement
I also think that we should give bots an environment they should survive in. Start with heat and cold. Heat is good but not too much heat, cold is bad period. So the bot would be forced to move somewhere near heat. So we can equate heat with pain. There should be some kind of memory too so the bot remembers cause/effect like going too close to fire means hurt, etc. You can then expand to hunger and drive to stay alive. You might even invent non-human type bot that doesn''t depend on hunger. Later perhaps add self-modifying capabilities to the bot so he can deal with harsher environments better. This would make him think of best way to prolong his existance. Perhaps introduce a friend to the bot so they could collaborate and throw ideas off of each other. Would be interesting to watch how they deal with life threatening situations. Not that I''m a sadist but from purely scientific perspective ofcourse
In a gaming sense, this sort of AI would be really interesting in a RPG - maybe even a MMORPG. Think about it - having NPC''s that form an actual world. People would die of old age, have babies, have their own language and culture. Gossip and rumors spread between the NPC''s would lead to quests for human players. Guilds could be actual real-time organizations.

I think that this genre of games would benefit the most from such AI, and certainly would make them even more fun to play than ever.

"There is no failure without trial; for that matter, there is no success either." - someone
"There is no failure without trial; for that matter, there is no success either." - someone
I finished my first GA program today! It solves a diophantine equation (C1A1+C2A2+...+CnAn=X) where the C''s are given coefficients, X is a given value for the equation, and the A''s are the values of the alleles.

As for the simple life GA, I would like to start with a very simple example. Imagine point-like organisms, that vary in terms of inherent life expectancy (maximum possible age), speed, manuverability, and need for resources (food, metabolism). Food will be generated at random, and the critters will race to grab the food, but more speed is not necessarily better, because a critter that is too fast will overshoot the food. Critters that don''t get enough food will die early, and will have fewer chances to breed. Fitness is based on ability to obtain food, but not on age. Long life ensures that the critter has more chances to breed.

[Formerly "capn_midnight". See some of my projects. Find me on twitter tumblr G+ Github.]

Fup I think wrote an ant farm program with ants racing for food. I''m saying skip the entire evolutionary path and jump straight to the evolved bots. From then on let the bots solve problems on their own. So give them the basic world rules by which they must govern to survive and then let them loose. The bots are mechanical so they don''t need to reproduce they just need to learn new things all the time based on experimentation with their environment. For example, I''m standing in my shower and an idea hit me. What caused me to see the solution? I didn''t have to reproduce for my child to see the idea. The idea just happended. Find out how and put this mechanism into the bot. Ofcourse the hard part is figuring out how I came to the solution, perhaps some kind of catalist spurred neurons to fire just right or something like that.
If you get far enough with linguistics, simple synonyms, you can make thought into this form:

SUBJECT + REASON = VERB

An example: John is tall because of his mother

Subject: John
Verb: is tall
Reason: Because of his mother

So, if you give a code for each subject, verb, then they can produce the reason themselves.

so, if john=asdf and tall=rewq and reason=0987, then it could be stated that:

asdf+rewq=0987

and if we do:

John is not short because of his mother

then, it would be asdf-ard=0987 (ard is short, - is not)

So then, eventually, the bot will have a long list of logical reasoning:

hot if near fire
cold if near freazer
not cold if near fire

so, then if it wanted "hot", it go near fire or whatever code was used.

This topic is closed to new replies.

Advertisement