AI chip?
- Thread starter _xxx_
- Start date
In the past there was a processor for graphics on the graphics boards. There are disk I/O boards with on-chip processors. Likewise with network I/O.
Not sure we need a specific AI processor though. If programmers wrote less bloated program code, I'm sure today's very capable main processors could do quite a bit in the way of that. Especially as typically, a game doesn't track all that many entities at the same time...
Not sure we need a specific AI processor though. If programmers wrote less bloated program code, I'm sure today's very capable main processors could do quite a bit in the way of that. Especially as typically, a game doesn't track all that many entities at the same time...
The only point I have to say is "Why didn't anyone think of it any sooner?"
chavvdarrr
Veteran
Lol, another dead-born idea.
Last edited by a moderator:
_xxx_ said:Because they didn't have an AI chip to aid the thinking?
AI is something way too programmable to have on a special chip. It would act more like a general purpose CPU than a special "hardware AI chip".
I really can't see this taking off, especially with multicore processors being Intel and AMD's solution to them not seeing the gains that they used to.
I really can't see this taking off, especially with multicore processors being Intel and AMD's solution to them not seeing the gains that they used to.
I was reading a old issue of CGW last night and it had a whole article on AI in games. They had an Origin programmer explaining how the AI in Wing Commander III works. What he described was about as complex as I've seen in modern games honestly. So, I don't think the AI issue is CPU capability so much as it is complexity of the programming. Creating an intelligence isn't exactly a simple task I'd say
The most demanding games for AI seem to me to be games like Civilation, Master of Orion, RTS games, X2 The Threat, and massively single player RPGs like Morrowind. It's why turns in some of those games take minutes to complete. Rome Total War's turns can take a decent chunk of time. But I don't see people running out to buy an expensive addon board to cut 1 minute down to 10 seconds in that kind of game. If that's even possible. And Morrowind, while it was very CPU limited, that was more due to the graphics engine being stupid than anything else I think.
If you go back to the late '90s there's a lot of hoopla over "artificial life" and how it's going to revolutionize gaming. Well, it never happened. If any of you've played Creatures, that's what was covered in a big, exciting article in Next Generation around '97. AI just doesn't really seem to be going anywhere. I'd say it's a combination of the earlier mentioned difficulty with creating complex AI and partly due to the popular genres of today not needing complicated AI. As has been said by ERP and other developers, AI isn't so much about realism as it is about simply making the game fun with some tricks the NPCs do. If you make a complicated AI it gets too hard and most people will just bitch and moan. What can you really expect AI in FPS's to do? Duck better? lol. Scripted events seem to be the best we can expect. And that works best in the simple linear FPS gameplay I think.
I'm fairly excited to see what Bethesda pulls off with their Radiant AI in Oblivion though. That seems like the kind of game that could benefit from "artificial life". They just need to somehow get around the potential of overly "smart" NPCs doing dumb things. Individually they may work ok, but we're talking about inhabiting an entire village with these simulated people and that is going to get complicated....
The most demanding games for AI seem to me to be games like Civilation, Master of Orion, RTS games, X2 The Threat, and massively single player RPGs like Morrowind. It's why turns in some of those games take minutes to complete. Rome Total War's turns can take a decent chunk of time. But I don't see people running out to buy an expensive addon board to cut 1 minute down to 10 seconds in that kind of game. If that's even possible. And Morrowind, while it was very CPU limited, that was more due to the graphics engine being stupid than anything else I think.
If you go back to the late '90s there's a lot of hoopla over "artificial life" and how it's going to revolutionize gaming. Well, it never happened. If any of you've played Creatures, that's what was covered in a big, exciting article in Next Generation around '97. AI just doesn't really seem to be going anywhere. I'd say it's a combination of the earlier mentioned difficulty with creating complex AI and partly due to the popular genres of today not needing complicated AI. As has been said by ERP and other developers, AI isn't so much about realism as it is about simply making the game fun with some tricks the NPCs do. If you make a complicated AI it gets too hard and most people will just bitch and moan. What can you really expect AI in FPS's to do? Duck better? lol. Scripted events seem to be the best we can expect. And that works best in the simple linear FPS gameplay I think.
I'm fairly excited to see what Bethesda pulls off with their Radiant AI in Oblivion though. That seems like the kind of game that could benefit from "artificial life". They just need to somehow get around the potential of overly "smart" NPCs doing dumb things. Individually they may work ok, but we're talking about inhabiting an entire village with these simulated people and that is going to get complicated....
Last edited by a moderator:
What the article describes is not AI, it's scripting.
For AI you need more: something that learns from past actions and makes up it's own mind. Which has been tried and done, to various degrees of success. But game designers in general don't like it very much, as they would not be able to predict it, so they cannot offer a storyline. Or debug one!
Anyway, I'm hard pressed to see the demand for a chip like described. If it was something like a neural network that stored actor states and could run them all in parallel, I might change my mind.
For AI you need more: something that learns from past actions and makes up it's own mind. Which has been tried and done, to various degrees of success. But game designers in general don't like it very much, as they would not be able to predict it, so they cannot offer a storyline. Or debug one!
Anyway, I'm hard pressed to see the demand for a chip like described. If it was something like a neural network that stored actor states and could run them all in parallel, I might change my mind.
swaaye said:
Honeslty, the more I read the more I think Radiant AI is just a nice phrase for a list of scripted sequences written by the programmers that the designers can choose for each NPC.
A dedicated porn chip? They would sell like hotcakes.phenix said:
epic
Blazkowicz
Legend
DiGuru said:
there exist things called "ZISC cards", which implement a network of neurons; though I'd don't really think what it does is AI. Still, maybe this pattern recognition stuff can be used for processing of sensorial input for bots.
http://en.wikipedia.org/wiki/ZISC
Cartoon Corpse
Veteran
what would the CPU be doing with it's horsepower if all this is offloaded to specialty chips?
graphics, physics, AI
would you even need much of a CPU? or would that free it up to handle much more sophisticated abstractions of game objects?...like an administrator of the big picture?
graphics, physics, AI
would you even need much of a CPU? or would that free it up to handle much more sophisticated abstractions of game objects?...like an administrator of the big picture?
Cartoon Corpse
Veteran
whatever gets me to infinity in a virtual nutshell (or something reasonably beyond my ability to notice it's not) is what im all for.
then later, the infinity in a nutshell chip. INPU! then the cpu can just keep score or something.
then later, the infinity in a nutshell chip. INPU! then the cpu can just keep score or something.
Similar threads
