Trusted Worlds

Too priceless to pass up - malware vendors have started including EULAs with their code. Next, they need to add an anti-reverse engineering clause ; )

In Mick West’s article “Intelligent Mistakes” in the April 2008 issue of Game Developer, he explores a twist on a fundamental game AI issue. It is not uncommon for players to see an AI make an essentially random move that works out well for the AI and impart that either a) the AI is really well done or b) the AI is “cheating”, i.e. the programmer lets the AI use inside knowledge to perform better than a player in the same position. That is, alot is in the eye of the beholder. This leads to the point of the article, a discussion of techniques for making a more realistic and well-matched opponent for the human player by designing the AI to make the same sorts of mistakes an otherwise intelligent human player of the same caliber would make. The AI gets better by getting dumber in a smart way.In terms of practical, modern game AI development, this is all straightforward and generally good. Game AI is (mostly) not about traditional AI development, not about neural networks and genetic algorithms and artificial life, but rather about good old smoke and mirrors. Shifting the discussion to trust issues…is making the AI fake dumb just as untrustworthy as the occasional game that designs the AI to cheat by looking thru walls to know your position?  How many players will be offended versus appreciative?As an aside, why is it ok for the game developer’s AI to do this, but not a player-created bot? Well, “because I’m the developer/I’m creating this to implement a certain level of challenge for the player” is usually a strong enough argument, unless your developers are also players and there’s financial gain involved, i.e. the AI version of the occasional developer cheating scandal in MMOs…

Should I feel cheated, or happy that the developers created something that passed a pseudo-Turing test? I’ll throw out 3 options for now:

1) The Practical Game Dev Answer - its smoke and mirrors, pseudo-Turing test and thus all of this is fair game in making the game work

2) The Deep Blue Answer - For certains types of games, part of the design/customer value proposition is to address the trust issues, i.e. to be clear is that the AI will play by the exact same rules as the player, so may the best player (man, woman, goat or software) win. Part of the goal is to see whether a less analytical, more intuitive human can beat a more or even massively analytical bot. Deep Blue isn’t allowed to cheat against Kasparov, although perhaps isn’t really the best example since in chess, the AI can’t do “look thru the walls” type cheating.

3) The Strong AI Futuristic Plotline - Ultimately, playing “fair games” against our AI overlords will be all at the good will of said overlord anyways. Ultimately, game play will be stratified into human leagues with dumbed down computer opponents to make us feel good and AI-only leagues which humans can never win at, at least until an indomitable hero named Jonathan E comes along to defeat the AI at their own level and restore liberty to humanity. At which point he’ll discover that that was itself a rigged game setup by the real AI overlords, and the game ends with trillions of AI twitterbotting “Jon-a-than! Jon-a-than!”

Alright, where was I?

Most game AI is about #1, but developers need to be clear about when they’re in #2 territory, i.e. recognize and respect the nuances of the needs of your target audience, and #3 is something I’d pay $60 to play on my XBox.

See here

An interesting question posed over at Play No Evil:

“Instead of banned blacklists, can we shift to “graylists” - players silently segregated based on their dubious playing habits?”

There’s a line of thinking in massively spectated offline real time skill gaming, aka football, etc,  of “Why not just let everyone cheat? They’re all doing it anyways, so its a level playing field.” I vehemently disagree with this, as there are quite a lot of people who want to do things without foolishly putting their body or their intellectual integrity at risk where its only necessitated by the frauds of other participants. So Taylor Phinney, the prodigy son of US cycling legend Davis Phinney, is following in his dad’s footsteps, and the whole family has to worry about whether their son could be forced to make the horrifying choice of a) consume dangerous drugs and make your life a fraud; b) give up your life’s ambition. That’s a physical game version of it, but there are plenty of online gaming analogs. So I’m all in favor of running the cheats out of town, or else they tend to end up owning the town.

But to the extent you could programmatically put the cheats into 1 bucket and let them have at each other…well, that could be fun for everybody. You’re effectively changing your policy to say, “It’s only illegal to cheat on non-cheat servers.” So you’ll have your PvP servers, your PvE servers, and CvC servers. You’ll probably even have some people intentionally cheating, just to get their entry card into the cheat server, maybe so they can run some sort of scam that they think will only work on cheaters (e.g. some social engineering grift that requires the mark to be exceptionally greedy).

Not saying its a great idea just yet, but its novel and interesting.

The March issue of Game Developer has a nice technical article on secure coding for games, focusing not on the usual suspects of a) piracy b) cheating c) DOS but rather on how flaws in game code can be used to own the players machine. Don’t trust the network and don’t trust half the art. Got it. The rest is just details