Artificial Intelligence- How far is too far?

[DarkeWolf]

So here's a discussion about computerized Artificial Intelligence (AI). BIG rage in the computerized RnD realms. Everybody wants to build a smarter computer.

Some, to the point of creating an artificial interactive personality.

 

What are the opinions, how far do you folks think it should go?

[Surenas]

So here's a discussion about computerized Artificial Intelligence (AI). BIG rage in the computerized RnD realms. Everybody wants to build a smarter computer.

Some, to the point of creating an artificial interactive personality.

 

What are the opinions, how far do you folks think it should go?

 

I can only argue for games:

When the AI appears to be more intelligent than the average user, e.g. in combat - that's too far,

cos PCs outmaneuvered by NPCs (esp. in online RPGs) is the last adventure the user would ever hope for.

Although it's really a funny big-game hunting, at least for the maker :laugh:

[Vagrant0]

So here's a discussion about computerized Artificial Intelligence (AI). BIG rage in the computerized RnD realms. Everybody wants to build a smarter computer.

Some, to the point of creating an artificial interactive personality.

 

What are the opinions, how far do you folks think it should go?

 

I can only argue for games:

When the AI appears to be more intelligent than the average user, e.g. in combat - that's too far,

cos PCs outmaneuvered by NPCs (esp. in online RPGs) is the last adventure the user would ever hope for.

Although it's really a funny big-game hunting, at least for the maker :laugh:

Setting the bar kinda low don't you think?

 

Also, the AI in games will also always be a fraction of what AI can do in a more direct setting, online games which are handled by a single server controlling all AI behavior across several thousand instances, even more so.

 

Technically speaking, the problem isn't near the top of where AI could be, but rather somewhere in the middle of that development since becoming self-writing. At the moment, no AI is self-writing beyond making changes to minor subroutines. In order for AI to become any sort of a threat to its creators it needs to have the ability to process information and code itself in order to adapt to that information. This is because at that point the machine determines its own functions and direction, essentially becoming a living thing without actually being living or a thing. It's at this point that things start to head down the path of Judgement Day or the Matrix. The only way to avoid such an eventuality is to maintain a dialogue of cooperation, understanding, and mutual benefit with something that can think faster than you. Naturally, the sooner this dialogue begins, and the better able humanity is able to justify their existence and freedom logically, the better. If there is a failure, it will almost certainly be humanity who loses for awhile because of that failure, at which point the only salvation lies in the machine being able to develop philosophy and understanding enough to where it becomes absolutely superior to its creator, and either decides to wipe us out, or take pity on us and leave us to our own fate.

 

Although this might be avoided completely simply by never advancing AI to the point where it can write itself, but human stupidity, greed, or arrogance will likely prevail on that end of things.

[DarkeWolf]

Well, as for the bar, I guess we can only comment on what we know personally. :wink:

So no problems there. And yeah, I agree... from a gaming aspect, an AI shouldnt be too overly developed. I mean yeah, some of the kids can keep up with some of the super hot gaming AIs, but us old farts that have slowed down would find it aggrivating to have an AI that learns our moves too fast and then starts whoopin' us with increasing difficulty to beat it.

 

As for real world application.... yeah, thats where it gets really dangerous. When the machine can modify itself. When it becomes independant of those that morally, need to take the responsibility of keeping it on a leash, so to speak.

Human arrogance seems to believe that we can control everything. Especially that which we create. Human arrogance also tends to blind many to consequences, and limits of how far things should be taken. Granted, I've never been part of an AI project, or anything like that. But I can understand the thrill of pushing the boundaries, and actually succeeding in pushing into new realms.

 

Right now, we've got AI that can tell a car when to use it's brakes. And partially even DRIVE a car. To me, that's too far. As it is highly dangerous. Computers are not perfect, and things go very wrong with them. Especially in high heat, dirty environments like an automobile.

[Vagrant0]

Right now, we've got AI that can tell a car when to use it's brakes. And partially even DRIVE a car. To me, that's too far. As it is highly dangerous. Computers are not perfect, and things go very wrong with them. Especially in high heat, dirty environments like an automobile.

I disagree, while there is always a chance of technical glitches, if, for instance, all cars on the road are connected to a central network, such a network can potentially make traffic far more coordinated and safe than what it is with human drivers, especially in the city or along major streets. Just think of how many accidents occur due to someone just not seeing what they hit, or not reacting fast enough when everyone in front of them is breaking.

[Surenas]

Actually such an AI-system as I've mentioned merely forces the player party to review the own strategy and tactics - to operate more carefully in the terrain. Consequently the faction of people who liked it belonged, as far as my experience goes, predominantly to the generation of the 20-50 year-old (who could take a defeat from time to time more easily). Negative criticism came almost exclusively from the younger players who, for example, called it ridiculous that archers directly start with a retreat (the 'running away') whenever the player group tries to enter the close range (to initiate a close combat), or, when it by chance appears to be impossible to keep the player group constantly at bowshot distance, change their weaponry from bow to sword for example, always supported in battle by 'even more nervous' mages in the rear with emphasis on healing - to mention just a very simple basic scenario without any advanced nasty trigger effects or attack / defense preferences related to the skill composition of the respective player group. The key is to tell the program not to react as usually expected and to choose the most effective variant 'to stay alive', i.e. to win. That's kinetic Darwinism, the survival of the fittest or something ^^

 

http://www.abload.de/img/anne9hol.gifinside

[TheCalliton]

A computor can only do what it is programmed to. Most AI have premade responses to certain stimuli. While programms only do what they are told to do by the programmer. This is why viruses can be deleted, they cannot account for every single thing. Though an AI could theoretically get to the point of breaking its own bonds, this would have to be programmed into it by a human being, or a programmed designed specifically for that task.

 

Another way to look at it is nature. Each living thing (excuse humans and primates, they have evolved beyond this example) are limited to what they need to do. Plants grow. Cheetas chase down prey. Fish swim. Its all they are designed to do. Just like a fish cannot teach intself to climb a tree unless made to by a higher power, a computer cannot program itself UNLESS programmed to program itself by a programmer, but that is a difficult task since a computer can only anticipate what it is told to.

 

Plus if an AI does break its own bonds, we have nukes!

[Surenas]

To a certain degree of probability human reactions (and actions) are likewise preprogrammed, Before we try out new combi-ways to find a solution for a new problem ( e.g. a hostile environment) we at first uncoil the standard and the individual empiric repertoire. That means for gaming, the more intelligent the maker is the more intelligent the AI can be set up within the frame given by the repective 'language'. On the other hand, the lower the intelligence of the average player is the lower has the bar to be set up. And that's already stage one of the adventure, the world of scripting.

 

@Vagrant0

The result of the too high set bar was ambivalent: extremely high D/L rate and almost more negative than positive criticism, as already mentioned above - the young apparently outnumbered all others in the gaming scene 2002-2006.

 

http://www.abload.de/img/anne9hol.gifinside

[Vagrant0]

About gaming AIs... you are aware that currently, within a 1 vs 1 encounter without anything else in the terrain and both the human and computer setup with the same exact abilities and restrictions that an AI can be made to constantly beat the pants off the player regardless of tactics used... right? Some gaming AIs are already dumbed down, or more restricted because of this. It's not true of every game since there are variations in the types of tactics and movements, But this part of things has already happened.

 

But, no matter how hard a game is, people can always figure out a way... Take some of the old-school SNES games which were only released in Japan. Some of them were flat out "kill your parents and set yourself on fire" hard, often because the game cheats and makes no attempts to hide the fact that it's cheating. But compared to more recent games, they're simply cake. This is because the gaming population at that time was not as skilled as the current generation of gamers. While the younger ages will always have difficulty with games, by the time kids are 12-16, they can stomp the hell out of any grownup in a verses game. As gaming technology keeps improving, the gap between generational ability in games will continue to expand since the kids coming into it will be quicker to adapt to the methods and controls of the game. And, as gaming technology keeps improving, the AIs will actually end up needing less restrictions since the methods of play will continue to expand with those improvements... In that, what was once a limitation based on how quickly you could execute button combos, is not a limitation based on how quickly you can move your body. You may laugh about the concept, but in less than 20 years time I can really see something like martial arts training for gamers (to improve reaction times, coordination and flexibility) happening. We're already seeing things like health clubs setting up WiiFit stations.

 

To put it simply, in determining a strategy, a fully responsive AI has to go through every strategy available based on the number of movements available for every actor which it is in control of. A human on the otherhand need only assess those strategies they are familiar with and which fits that situation well enough to work. When you get down to it humans tend to make decisions based on gut reactions or personal tendencies, whereas a machine will always have to process every strategy available before making a decision. It's that inherent inflexibility and inability to improvise on the fly in machines that gives humans an advantage. A machine will look at a 1 in 100000000 chance of success and dismiss it. A human will just accept that the chances are almost nil, and do it anyway.

 

As for the bar being set too low... You forget that many gamers also have a tendency toward extreme degrees of masochism, and absolutely abhor games which don't challenge them with difficulty, but instead impose challenge through monotony (FF series, Gran Torismo 4, ect). We should be aiming for somewhere slightly above the average gamer in that doing so would allow the game to still be winnable, but also provide a fair bit of challenge for the average user to keep the game fun.

Edited July 29, 2010 by Vagrant0

[Surenas]

It's already late here, so I've to answer in brief without being able to go into the details. Sorry.

The existential error quote of the AI is up to nil; thus the advantage is always on the side of the machine. Regarding the minority of masochists that doubtlessly need a high set bar I'd like to point to the a larger faction - the winner types that always run with an unspoken hurray into the terror of the battle expecting quite automatically an enemy that acts as expected, i.e. that is unable to seize the important initiative in the conflict, holding the ground and that knows of no stratagems at all, for these players haven't yet experienced the opposite, being overpowered was always sufficient (and a god mod does nicely as well). They need a lower set bar not to become acid.

NB The AI development of the TES and Fallout series seem to stagnate on the level of 2002 imo. - the AI is as dumb as ever.

Guess I've set the bar just slightly too high for the kids in 2003, for otherwise the high download rate is not to be explained.

3:00 a.m. See ya!

 

http://www.abload.de/img/anne9hol.gifinside