TOSOS Game AI Project Beyond Artificial

For the past several years now, John and I have been tinkering with an online game creation/distribution system and we've finally put together an alpha version. We call it the Geewhiz Foundation. There are three main driving factors behind what we're doing. The first was to take the mindless work out of making the game and make it very quick and easy to gen up a game. The second was to make it easy to mod existing games online so that we could trade the assets back and forth (We've had terrible luck with artists.

A while ago I read, or rather listened, to "Zen and the Art of Motorcycle Maintenance." Its more fun than any other modern philosophy book I've tried to get into--its got great narrative structure. But the thing that struck me while listening to it was the description of quality. I'm going to paraphrase: It is something which everyone can recognize almost instantly as being good, but nobody can really describe in detail why. This sounds pretty much exactly like the class of NP problems.

I posted some movies I made for the demonstration at UM2007. They are at http://www.cs.unm.edu/~ndfabian/dtree_cloning.html. They should run very similarly in the code we released at the end of last year, but without the ugly red/blue filters :)

I'll make a real entry at some point soon.

A movie is interesting because the protagonist screws up at some point.
A game is interesting because the protagonist (you) must never screw up.

People that know me like to know I talk endlessly. Occasionally I like to make an effort to know what I'm talking about. I prefer to be a doer rather than a talker, but since I talk so much it means I have to do that much more. I've spent the last few months talking in various ways about my thesis work. I thought it would be a good time to actually show what I'm talking about. Thus John and I have decided to release the game I've been using to do the work as GPL. Read on for more information and some of the history.

[Of the title "Giant Brains or Machines that Think."] The simplest way of showing how preposterous that title is is by pointing at its two companion volumnes--still to be written--"Giant Hearts or Machines that Fall in Love" and "Giant Souls or Machines that Believe in God", the most fascinating feature of the latter, of course, being that they can believe in God much faster than you.

http://www.cs.utexas.edu/users/EWD/transcriptions/EWD09xx/EWD936.html

I've had this fascination with the limits of knowledge since I first learned about the Incompleteness theorem. You grow up thinking that its quite possible to mathematically describe anything and that the world runs like clockwork. And then at some point, I also learned about Quantum physics and the randomness at the bottom. It seems wild to think it, we're so used to the idea that something causes something all the way down that if we get to the bottom we say "Oh well, its just entirely random." Its just about as close as saying its just entirely magic.

The typical stove has four burners arranged in a square, and a dial to control each. How do you arrange the dials? The simplest answer is to put them in a row. But this is a simple answer to the wrong question. http://paulgraham.com/taste.html

The other day a friend of mine and I were talking about games and bemoaning the fact that at some point after you've played for a while you start optimizing the game. And as a result you hardly playthe game anymore. Its hard to explain, but Gabe at penny arcade has a really good anecdote. Picture this: It has less to do with commanding an army of aliens and more to do with typing numbers into a spreadsheet. Maybe if you were Ender then typing the numbers in is a good idea, but ... you're not.

Positive-only learning is sort of one half of reinforcement learning. http://www.cs.ualberta.ca/~sutton/book/the-book.html Its the part where you don't see the down-side. The things that went wrong. In most cases we only see patterns of behavior, but can't really determine whether individual decisions are good or bad or which are mistakes, etc. We can usually only understand things in aggregate, if at all..