Now that we have the PPU, what is next??

That would be awesome :eek: , its like, smelling the aroma of another country, making you feel like your there, lol. Or like, the outdoors or something, lol.
 
A card with AI functions would be nice, to allow for more lifelike NPCs and smarter level bosses. It could route travel paths and determine whether the computer really could see the player at any given time. Possibly perform some level of weapon aiming. as long is it was flexible enough that all games didn't start seeming more alike than they seem now.

Games won't really seem lifelike until you have to stop and exchange insurance information after wrecking.
 
something to reduce all the noisy fans.....

please dont tell me water cooling... i'm not gonna run water over a very electrical computer.. i dont want my house to burn down because my kid or girlfriend dropped my computer

oh and another vote for hamster brain CPUs
 
I think AI advances will be up to reasearch...not dependent on hardware...probably not even in the industry...in academia moreso.
 
Sly said:
Of course, i have no freaking clue how you're supposed to code that without using fixed (e.g.scripted) parameters.
Click to expand...


Cause & Effect. Have the AI find out what happens. The only scripted part of the AI system would be the physics and things like that. It would know about gravity, mass, weight, light, sound, etc... But, it would have to be "taught" about other things. Such as: collision detection, balance, movements, predicability; things that don't fit into the normal "logical" world. Basically, it would need to learn chaos. Things don't always happen the same way twice.

I'm working on an AI experiment currently, and am coming up with some amazing new ways to work with things. There is some amazing stuff out there if you do some research. It does take some massive horsepower to do GOOD AI. Not only at the CPU level. You need lots of memory bandwidth (to take all the inputs and make split second decisions based on those inputs). What I'm working with is a lot slower than what it should be, but its a small cluster of PC's to control various inputs (senses) and react to them depending on a certain "mood" and "personallity". It's the personallity that is hard to implement. As it is programming, I can get a mad, easily angered response, or I can get a generally happy response, just by changing a little bit of code.

I don't see an addon AI card with these types of AI. Possibly on a game level, but I can see that done on a multicore CPU very easily. Like it was mentioned before, Oblivion did very well with the AI, there. For gaming. FPS's are another story. I've had to TK a computer player because he was just standing in a corner at my base. Never moved the entire game. Yea, great AI there. :rolleyes:

I see maybe a 3D/VR style addon card coming next. I'd pay for that kind of immersion, if it was good. >120 degree visuals, Dolby Digital surround sound, fluid framerates, top notch graphics. With SLI video cards doing 2 images, I don't see why this isn't being done. 1 for each eye, high res... Hmmm.... Untapped market.
 
Any advances in AI will be algorithmic, not hardware-based. All the various AI techniques I learned about in my AI class were sound, but VERY difficult to apply to something besides games. You want an SVM that recognizes zip codes? How bout a minimax algo for chess? Those aren't too terribly hard, compared to the limitless possibilities that you'd run into for a modern, smart, human AI. You can't just make a decision tree and handcode 30000 different possibilities for all the little things that a real person would do. Perhaps something like cellular automata might be of some help, along with neural nets so we can just train the algo for weeks/months until it shows some reasonable intelligence.

I'm going to guess that that's what we'll see for AI at least sometime in the future, but don't expect that to be fast at all. When we were fooling with neural nets they were slow as dogs, but ours had a fwe dozen hidden units, probably MUCH smaller than you'd need for human simulation. Multithreading this could be done without a huge amount of work, so multicores would really help this scheme out at least.
 
You must log in or register to reply here.
Back
Top