Print Story Misc. Misc. indeed.
Misc.
By codemonkey uk (Thu Apr 08, 2004 at 01:10:40 AM EST) (all tags)
Inside: Script Kiddie Prayer, & GDC AI Roundtable.

Also, MLP:



Script Kiddie Prayer
Author unknown:
Our Father, who 0wnz heaven, j00 r0ck!
May all 0ur base someday be belong to you!
May j00 0wn earth just like j00 0wn heaven.
Give us this day our warez, mp3z, and pr0n through a phat pipe.
And cut us some slack when we act like n00b lamerz, just as we teach n00bz when they act lame on us.
Please don't give us root access on some poor d00d'z box when we're too pissed off to think about what's right and wrong, and if you could keep the fbi off our backs, we'd appreciate it.
For j00 0wn r00t on all our b0x3s 4ever and ever, 4m3n.
GDC AI Rountable
This roundtable was not as popular as the "By The Book" roundtable, and was billed as a "general interest" discussion. The first topic participants discussed the "state of the industry". One question raised early in the discussion was: is the development of AI in games was stagnating? The consensus being that no, it was not, but it may appear to be because the focus of AI programmers was now more on providing a service to the game, than creating AI technology for technologies sake. It was agreed that "fun" came first, and their was no point creating ultra smart AI that dished out one humiliating, crushing defeat after another.

It was observed that the illusion of intelligence was often more important than intelligence itself, and that other elements of the game, such as audio and animation could often contribute more obviously to creating this illusion than the AI itself.

The conversation moved on to more technical issues later half of the discussion, though nothing particularly groundbreaking was mentioned until one participant dropped the bombshell that his company was currently developing an AI hardware solution aimed at the consumer market, and would have a demo "soon".

I spoke to him in private after the talk. He told me that they would be manufacturing a chip and expected the cost to be ~$20 / unit. His company aims to sell this chip to graphics card and motherboard manufactures as an added-value component, rather than selling a PCI (or whatever) expansion card direct to the consumer.

Although I was impressed from a technical standpoint, I have always doubted the commercial viability of an ai-acceleration hardware device, and his business plan did nothing to dissuade me of this opinion.

< Projection | BBC White season: 'Rivers of Blood' >
Misc. Misc. indeed. | 13 comments (13 topical, 0 hidden) | Trackback
That AI stuff by Dr Thrustgood (5.00 / 1) #1 Thu Apr 08, 2004 at 01:15:08 AM EST
Isn't it more about attacking things from the other angle, that is, making bots stupid so that they're fun as opposed to, as you say, the perfect killer?

Still, I want my all-seeing fifteen-foot white rabbit!





Idly wonder by djotto (5.00 / 1) #2 Thu Apr 08, 2004 at 01:52:27 AM EST
what the point is, when you're talking about machines with 2Ghz+ processors.

Is he encoding specific algorithms in the hardware? Then why not sell them as a library?




massively parallel by codemonkey uk (3.00 / 0) #3 Thu Apr 08, 2004 at 02:03:55 AM EST
The chip does "navigation" and "terrain analysis". Apparently 1000x faster than the same task can be done on a general purpose processor. He claims the algorithms are better suited to implementation in hardware than in software, so presumably they involve some kind of massive parallelization.

--- Thad ---
Growing a mustache for charity.
[ Parent ]

A* on a chip? by Gully Foyle (5.00 / 1) #5 Thu Apr 08, 2004 at 03:03:45 AM EST
Somesuch thing has been suggested on and off for years, but nothing's ever come to market.

[ Parent ]

Not A* by codemonkey uk (3.00 / 0) #6 Thu Apr 08, 2004 at 03:23:41 AM EST
One of the points that this guy raised was that optimal algorithmically is not the same as optimal in hardware. The system implemented on the chip is not A*.

--- Thad ---
Growing a mustache for charity.
[ Parent ]

Okay, so maybe not A* strictly by Gully Foyle (5.00 / 1) #7 Thu Apr 08, 2004 at 03:32:01 AM EST
If you're going to search a graph for routing, then you're pretty likely to use some form of A, even if you don't use an admissible heuristic to make it A*.

Actually, doesn't the ability to select your heuristic mean that A generalises to all search strategies?

[ Parent ]

Not all, but quite a few by codemonkey uk (3.00 / 0) #8 Thu Apr 08, 2004 at 03:39:57 AM EST
For example, a constant value 'heuristic', and a constant value link 'cost' turns A* into a breadth first search...

--- Thad ---
Growing a mustache for charity.
[ Parent ]

Why not all? by Gully Foyle (6.00 / 1) #12 Sun Apr 11, 2004 at 06:58:08 AM EST
Since the heuristic can, in theory, be arbitrarily complex, surely you can tailor it to take any path through the search tree you like?

Maybe I'm just misremembering. It's been a while since I was an undergrad...

[ Parent ]

Well if you put it like that.. by codemonkey uk (3.00 / 0) #13 Sun Apr 11, 2004 at 07:29:41 AM EST
Then yes.

--- Thad ---
Growing a mustache for charity.
[ Parent ]

In my experience... by Breaker (5.00 / 1) #4 Thu Apr 08, 2004 at 02:07:26 AM EST
AI in games is not actually like that.

They cheat like mad.  Playing Civ, a computer controlled city that's on the verge of collapse due to your assault will miraculously rebalance and spring a surprise tank or whatever next turn.  In Unreal Tournament, some of the players have preternatural abilities of either finding you in a maze, and dodging your shots way before any player could.

Not that I mind, however, there's seems to be a controlling process that is allowing use of the "cheats" internally to provide a better gaming experience.




Cheating by ucblockhead (5.00 / 1) #9 Thu Apr 08, 2004 at 05:11:36 AM EST
Too often designers use AI cheating because it's easier. The Civ games are some of the worst offenders. (Though Civ III is markedly better than Civ II was.)

The only 4X game I've found where the designers seem to have actually worked at making a real AI is "Galactic Civilizations".

Bots for FPS are almost an entirely different subject. For a 4X game, the problem is generally that the AIs are deficient in strategy. In FPS games, making the bots seem like real people is the problem. It is trivial to make a bot that hits every time. It is harder to make one that shoots like a real person.

One annoyance I've had with Unreal is that the bot seems to know exactly when the "double damage" will appear. I've multiple times I've witnessed it moving over the double damage spot within a few seconds of it appearing.

(Once I figured this out, it was easy to kill the fucker by camping near that spot.)
---
[ucblockhead is] useless and subhuman
[ Parent ]

X-COM / Magic & Mayhem AI Never Cheats by codemonkey uk (3.00 / 0) #10 Thu Apr 08, 2004 at 06:40:09 AM EST
I can say that for sure, because (a) I worked on it, and (b) it was a point of pride/principal in the Mythos office.

Cheating in general is frowned on by game AI developers, but mostly because people spot it, and when they do it spoils the game-experience.

--- Thad ---
Growing a mustache for charity.
[ Parent ]

XCOM by ucblockhead (5.00 / 1) #11 Thu Apr 08, 2004 at 07:12:20 AM EST
Yeah, now that I think about it, I never caught XCOM cheating. Though in a game like that, it's easier to balance because the computer opponent, by definition, doesn't have the same capabilities as the human opponent. (Someday I'll have to check out Magic and Mayhem.)

I can't think of a 4X game where the AI didn't either cheat or suck except for Galactic Civilizations.
---
[ucblockhead is] useless and subhuman
[ Parent ]

Misc. Misc. indeed. | 13 comments (13 topical, 0 hidden) | Trackback