Well, theoretically, once we have something about as smart as us, we can just feed it massive amounts of processing power until it figures out how to make itself smarter. Theoretically.
I just made an asteroids clone that uses imitation learning to copy player behaviour. It actually works pretty well, and I'm not even doing any really advanced stuff yet. So it's definitely got potential applications for vidya gaems.
Sepples is oddly lacking in ML libraries, though. Matlab, Java, and Python all have better.
Well, there is self-learning maps for analysing data, but I can't think of any really commercial AI ideas one could use in software alone.
Name:
Anonymous2012-08-29 19:24
Human language translation
Machine vision
Domain expertise decision support
Data mining
Name:
Anonymous2012-08-29 19:31
>>18
All could be used for evil. Imagine the Jews, who are small like rats and in need of machinery to oppress goyim. This machinery needs sensors and manipulators, which in effect need AI. Language translation would serve to detect any antisemitic writings, while machine vision will discern a goy from a Jew.
AI basically died in the seventies for a reason. Everyone was like "hurr we will replicate human intellect durr" and then it ended up not being doable and now you are usually shoved up some tl;dr if-else routines as "intelligence". Fuck that shit, there is no commercial future in AI. Don't get close to the field unless you want to be an academic who wants to get away with not actually doing anything.
Name:
Anonymous2012-08-31 7:49
>>25
The computer gaming field appreciates new advances in AI technology.
>>27
This is why they appreciate the advances made in AI. ATM, the AI technologies aren't sufficiently satisfying to the users. This means that improvements in this field would also translate into a more satisfying computer controlled entities.
I always wondered in those racing games, when you knock a CPU car off the road, how do they know to get back on the track, the right way around and into the race? Is it neural nets?
>>40
what >>42 says. A lot of game ai can be modeled with simple state machines. When a car has lost control and exits the track, it can transfer its state to the ``get back on the track state''. Then once it is back on the track it can go into the ``go as fast as possible without hitting the walls on the turns'' state. And then when it approaches the player from behind, it can choose to go into the ``try to fishtail the target'' state, where the target has been assigned to you. When the driver is in each of these states, optimal driving patterns can usually be calculated using some physics. Something this simple can end up producing a very brutal and unbeatable opponent. It actually takes more effort to make an AI that drivers dumber so that the player wont be completely frustrated.