Hi everyone! I'm in the process of starting up a company offering services and software for real-time AI, with particular regard to videogames. My team includes experts in theoretical neuroscience, AI, and machine learning, as well as developers, testers, and marketing professionals. Even after having talked to several game developers, it is not yet clear to me how much game AI is worth. Which percentage of the bugdet for game development can go into high-quality AI? How much time/resources do you devote to AI? Can anyone come up with some figures? Thank you! RFI are of course welcome! cheers AUV_ZH

most teams I've worked on have 1-4 AI programmers. So the "budget" for AI is:

1-4 * yearly salary of engineer * num years for project

probably throw 1/4 of a manager at that too...
and at least one full-time designer...

That said it's entirely project-dependent. Black & White is all about the pet AI so their "budget" is high. A tetris game likely has a zero or very low AI budget.

As for how much it's "worth".... check out other 3rd party AI developers. There are at least 2-3 professional groups out there selling to the big publishers. Don't remember the names offhand. One of them is something like Kinegon, but i think i have the spelling totally wrong.

-me

Hi !
I think it's a nice idea, as more and more game developement will involve outsourcing of all sorts, including AI. Instead of every team reinventing the wheel over and over, services of small buisness such as yours will probably help improve the overall game quality (at least, tahts what I hope)
So I didn't come with any advice but rather with a question: do you plan to build on-demand solution for every studio requiring you service, or will you rather build a "general purpose" AI engine and sell user licences for it ? (or maybe both ?)

All the best luck !
Janta.

I spoke with several game developers: most of them are not completely satisfied with the general-purpose AI libraries available on the market now, because they must be adapted to fit their specific applications. We intend to provide consulting services regarding the choice of the best suited computational algorithms (e.g., Neural networks, hidden Markov models, bayesian networks, genetic algorithms, etc), development of tailor-made libraries, and assistance in the integration and testing phases.

Quote:Original post by auv_zh
I spoke with several game developers: most of them are not completely satisfied with the general-purpose AI libraries available on the market now, because they must be adapted to fit their specific applications. We intend to provide consulting services regarding the choice of the best suited computational algorithms (e.g., Neural networks, hidden Markov models, bayesian networks, genetic algorithms, etc), development of tailor-made libraries, and assistance in the integration and testing phases.

Do you think Neural Networks, hidden Markov models, bayesian networks and genetic algorithms are particulary adapted to gaming applications?

Quote:Original post by Steadtler
Do you think Neural Networks, hidden Markov models, bayesian networks and genetic algorithms are particulary adapted to gaming applications?

I am making no allegation regarding the suitability of particular computational paradigms to videogame AI. I just wanted to explain what I meant with "computational paradigm". Neural Networks, hidden Markov models, bayesian networks and genetic algorithms can be successfully employed to tackle certain classes of problems, but, for any given problem, several other paradigms could be successfully used. Success in this case mostly depends on the familiarity with the research done in the last 20 years and its real-world applications.

Cheers
auv_zh

I haven't spent a significant amount of time looking at packages, so take this with a grain of salt.

The most reuse I have seen inside the industry has been along two axes:

1) AI frameworks. These are typically focused on domain specific problems as as planning systems specialized for shooters.
2) Tool supported AI systems. These generally fall into two categories - general authoring tools (a bit equivalent to a scene graph editors for a world editor) or domain specific tools (such as an editor for building sports game formations).

In my experience, the runtime/logic isn't the difficult part to manage in most AI systems. The complexity comes from the supporting all of the little specialized bits needed over time, which typically is either datadriven.

The sales pitch for these packages is tough - either you pitch a system that totally replaces AI programers (which must be an extremely complete, tool based solution), or you sell tools to the AI programmers themselves to increase their productivity.

In general, it is probably save to assume that most games have 2 dedicated AI programmers. If your package enables a company to drop down to only 1 AI programmer, you might be able to go as high as 50k. You might want to look at packages from Havok or RAD, as devs frequently pay for their packages.

Quote:Original post by fup
I guess they are the two ends of the spectrum. On average I would estimate that most games developed (across all platforms) have 1 or less dedicated AI programmer.

Is that because of lack of resources, poor AI for the buck, no need for AI at all or what?

thanks!
auv_zh

Heh, well... Condemened and Fear both had a single AI programmer. From the other companies I've talked with recently though, it sounds many have more than one. Granted, you are very right about smaller games having 1 dedicated AI programmer, if that. I figured the middleware was targeted at larger companies.

Unfortunately, my team currently has 0 AI programmers. I moved from AI programmer to lead a few months ago - we are still looking for a replacement. If you know of anyone with game dev experience who may be interested, point them at me. :)