4 ms·

which "new lead" in computation do you think will allow it? I don't know if hardware is holding back progress in AI, especially with distributed computing; it

by byoung2 17y ago

which "new lead" in computation do you think will allow it? I don't know if hardware is holding back progress in AI, especially with distributed computing; it is more likely a question of software and programming. I think in order for a computer to exhibit a reasonable facsimile of human intelligence, it will have to do more than just run programs written by humans. It would have to have the ability to write and rewrite its own code.

reynolds 17y ago
Are humans at a place where we can write and rewrite our own DNA?
- byoung2 17y ago
  The computer equivalent of human DNA would have more to do with hardware than code. When I talk about computers rewriting code, I'm talking more about rewriting patterns of thought processing. While humans can't rewrite DNA on the fly, we can definitely teach ourselves new ways of approaching problems.
  - dkersten 17y ago
    An FPGA of sorts?
coffeeaddicted 17y ago
Hardware is holding back progress in AI. Well, I can only speak for game AI, but given that game AI even has to work with simpler worlds and can fake many things I suppose it's not much different in other AI areas. And game AI programming is a lot about programming against the limited available resources. You mention for example the ability to rewrite own code, but what you think about might be more about unsupervised learning in general. I occasionally did experiments in that area - and so far always stopped them after a while due to lack of processing power. I don't say that there isn't also a need to improve the software side, but missing processing power was at least so far the thing that was always blocking me more. Just to give you one example. Think of bots in shooters. Humans have no problem browsing the whole screen in realtime and evaluating the data. But try even just doing the simplest image processing in realtime with a typical resolution and you will already find out how hard that is. So bots can't do that. Instead they cheat and shoot a few single visibility lines and even those are already targeted because bots can know where opponents are and so they only do for example linechecks to certain areas like head, body, arm, legs and try to find out if that line is blocked. And even that might already be too slow with many opponents so you have reduce it further and shoot only a certain number of lines each frame. And there is no processing yet - this is only about getting the simple information of - is there something or not. This is how slow computers still are. You could certainly do a little more with giving a single bot the whole processing power of your computer, but with the increasing complexity of game worlds the needed processing power grows rather fast (current addition of simple solid-body physics in games makes AI programming already way harder). For AI tasks computers are still very slow.

by byoung2 17y ago

reynolds 17y ago

Are humans at a place where we can write and rewrite our own DNA?

byoung2 17y ago
The computer equivalent of human DNA would have more to do with hardware than code. When I talk about computers rewriting code, I'm talking more about rewriting patterns of thought processing. While humans can't rewrite DNA on the fly, we can definitely teach ourselves new ways of approaching problems.
- dkersten 17y ago
  An FPGA of sorts?

coffeeaddicted 17y ago

Hardware is holding back progress in AI. Well, I can only speak for game AI, but given that game AI even has to work with simpler worlds and can fake many things I suppose it's not much different in other AI areas. And game AI programming is a lot about programming against the limited available resources. You mention for example the ability to rewrite own code, but what you think about might be more about unsupervised learning in general. I occasionally did experiments in that area - and so far always stopped them after a while due to lack of processing power. I don't say that there isn't also a need to improve the software side, but missing processing power was at least so far the thing that was always blocking me more. Just to give you one example. Think of bots in shooters. Humans have no problem browsing the whole screen in realtime and evaluating the data. But try even just doing the simplest image processing in realtime with a typical resolution and you will already find out how hard that is. So bots can't do that. Instead they cheat and shoot a few single visibility lines and even those are already targeted because bots can know where opponents are and so they only do for example linechecks to certain areas like head, body, arm, legs and try to find out if that line is blocked. And even that might already be too slow with many opponents so you have reduce it further and shoot only a certain number of lines each frame. And there is no processing yet - this is only about getting the simple information of - is there something or not. This is how slow computers still are. You could certainly do a little more with giving a single bot the whole processing power of your computer, but with the increasing complexity of game worlds the needed processing power grows rather fast (current addition of simple solid-body physics in games makes AI programming already way harder). For AI tasks computers are still very slow.