Computer Scientists Invents Game-Developing Computer AI

MojoKid writes "Over the past few years, short game writing 'jams' have become a popular way to bring developers together in a conference with a single overarching theme. These competitions are typically 24-48 hours long and involve a great deal of caffeine, frantic coding, and creative design. The 28th Ludum Dare conference held from December 13 — 16 of this past year was one such game jam — but in this case, it had an unusual participant: Angelina. Angelina is a computer AI designed by Mike Cook of Goldsmiths, London University. His long-term goal is to discover whether an AI can complete tasks that are generally perceived as creative. The long-term goal is to create an AI that can 'design meaningful, intelligent and enjoyable games completely autonomously.' Angelina's entry into Ludum Dare, dubbed 'To That Sect'" is a simple 3D title that looks like it hails from the Wolfenstein era. Angelina's initial game is simple, but in reality Angelina is an AI that can understand the use of metaphor and build thematically appropriate content, which is pretty impressive. As future versions of the AI improve, the end result could be an artificial intelligence that 'understands' human storytelling in a way no species on Earth can match."

Comments

What's that smell?

[narcc]

Smells like bullshit to me. What do you think?

Re:What's that smell?

[GODISNOWHERE]

This story isn't bullshit, and might make for mildly interesting cocktail party chat, but it isn't really newsworthy.

As future versions of the AI improve, the end result could be an artificial intelligence that "understands" human storytelling in a way no species on Earth can match.

This probably does qualify as bullshit, and it was only was only added because the author thought the story itself isn't strong enough to stand without it. Tech writers have to fill quotas. The problem with this peroration isn't just that it's stupid and wrong—it is—the problem is that it gives people the wrong expectations for what AI can do. AI has already had significant payoffs. The Dynamic Analysis and Planning Tool (DART), an "intelligent agent" (a dirty word after the AI winter) written in Common Lisp and used by the U.S military was introduced in 1991 and by 1995 had saved enough money to pay for all of the money DARPA has spent on AI in the previous thirty years.

Re:What's that smell?

[Anonymous Coward]

Agreed. It's entirely possible to create a program that will create a game, complete with a story, boss stages and the lot. In fact, if the army of developers who shaped Angelina was big enough, she could be developed to create games of any genre (be it FPS, RTS, RPG, arcade, etc). However, that will not make Angelina any more intelligent than the default calculator provided by our respective OSes.

Now, show us Angelina making decisions to autonomously change the genre, story, bosses, etc, in a way that fulfills Dennett'esque or Sartre'esque imagination theories and we can start calling it AI. Till then these stories only serve as a means of impressing the uninitiated - which is definitely important if the field of AI is to get the attention it deserves.

Cloudberry Kingdom

[HalAtWork]

Cloudberry Kingdom, Spelunky, and many rogue-likes all do this on a smaller level, but are always constrained by parameters. While they seek to create an AI that will take on more of the tasks, it will still have to be fed parameters created by an author, so unless this AI can create itself, how can it be called truly creative? Rather it is just procedural generation. It may be worth doing but calling it creative is hyperbolic.