?

Log in

No account? Create an account
GDC 2008 - mini recap - Blather, Rinse, Repeat
February 25th, 2008
06:59 am

[Link]

Previous Entry Share Next Entry
GDC 2008 - mini recap
I spent last week at the 2008 Game Developers' Conference (btw, I don't think they use an apostrophe, for fear of putting it in the wrong place). I bumped into several old friends/acquaintences/coworkers, and actually had lunch with Steve "Planetfall" Meretzky (he thinks the Nintendo DS is an interesting graphic adventure platform).

Now that I've dropped my one name, some of the themes in sessions I went to included:

  • procedural content - Will Wright showed a graph a few years ago about the increasing size of games, measured in content. Early games (e.g. "Pong") didn't require a lot of fanciness, but these days, you've got an army of artists filling up your DVD-ROM. (Or, I guess, Blu-Ray. Congratulations, Sony.) These things tend to be exponential, rather than cyclic, which would indicate that future games will require more and more art. If one naively extrapolates out, somewhere in 2014, every human on the planet will be enlisted to work on Grand Theft Madden: The Ineviting. I joke about the title, and I might be wrong about the date, but ballooning content requirements are real.

    Procedural content is a way to leverage these computer things to help generate some of that content. Will Wright, famously, is procedurally generating a galaxy of planets for Spore. Far Cry 2 procedurally generated skies and terrains for some sort of African first person shooter. Not much news, but good to see who's working on what.

  • player-generated content - another way to provide new and interesting stuff for your game is to abdicate and force allow the players to create stuff on their own. Spore, of course, will be providing creature editors to the players, which will give them the ability to swap content with their friends, or the world at large. There are several interesting challenges here, from the difficulty of making super-simple tools for in-game authoring, to providing editorial review (filtering out the inevitable objectionable content - if necessary).

  • Startups - I went to a couple of talks about the experiences of getting small companies off the ground. One of them had a bunch to say about venture capital, the most useful of which was "If you can't turn $1 million into a revenue stream of tens or hundreds of millions, venture capitalists won't be interested". Another panel discussion brought forward the experiences of three guys. The first guy's project crashed and burned, and took the friendships of the team down with it. The second guy's project is still ongoing, which you can think of as the baby bear's porridge being a compromise between too hot or too cold, or you can think of it as teetering on the edge between success and failure. The third guy has had some modest success. A lot of familiar refrains ("you'll work hard and you might work long hours, but you'll love it", "do it right out of school if you can, before you get a wife and a mortgage, and kids"). There were some interesting data points, including the successful guy living in a cheap apartment on $20,000 a year.

  • Everything's linear if you apply enough logarithms - Ray Kurzweil gave one of the keynotes. He didn't talk about games, really, but he did talk about how technology (and, in general, information sciences) make advances exponentially, which has profound and possibly alarming implications. If you think back to the parable about the inventor of the game of Chess, who got paid in rice by the emperor of China - one grain of rice on the first square of the board, two grains of rice on the second square, 4, 8, 16, and so on. After the board was half-full, the guy had about as much rice as one medium-sized field. By the end of the board, he would have had more rice than you could fit on the planet. It's a parable.

    But that's exponential growth for you. Computers are getting smaller and faster, yielding an exponential payoff of computing power per dollar or per cubic inch, or just about any other metric. Kasparov dismissed the possibility of a chess playing computer back in 1989, but didn't believe (or appreciate) that a doubling of chess-playing power every year would lead to over 250x the performance in 8 years. By 1997, Deep Blue played Kasparov to a 3.5/2.5 defeat. Presumably, 11 years later, we should have another 2000x performance increase.

    If you buy his conjecture that advances are coming increasingly quickly, the implications seem inevitable, and perhaps a bit scary. Computers are getting more capable on an exponential curve (which includes, but is not limited to, Moore's Law, which says that Intel doubles their transistor count every 18 months). Humans are evolving on a much slower (possibly linear) curve. At some point, probably in our lifetimes, the curves will meet, and we'll see computers that are as smart as humans. And, much in the same way that I can play an emulation of Pac Man on modern hardware, computers will be capable of faithfully emulating the workings of a human brain. This may not be an efficient way to achieve human-level intelligence, but it's a sufficient one. Once we have a machine that's smart enough to consider its own design and create offspring, selecting for increased capability and/or efficiency, we'll see the exponential growth of computing devices leave our control. You might be thinking of the Sorcerer's Apprentice at this point, with hundreds of brooms-turned-automatons carrying buckets of water. I don't suspect it'll be so bad as that. Hopefully.

    Another piece of exponential technological growth to consider is the application of information science to medicine. Most of the drugs that we currently have were discovered, rather than designed, which is a much slower process. As we apply exponential growth of technology to medicine, we will see gene therapies that switch the insulin receptors in humans, which will have significant impacts on obesity and diabetes. (Laugh line from the talk: "Current diet pills mostly address appetite, which is like a birth control pill that works by removing sexual drive".) Also, human life expectancy is increasing. What's surprising is that it may be increasing exponentially. It seems to be increasing something like three months every year right now, and may increase to one year per year in the meaningfully short future. And it's not just future generations that will benefit - much of this will pay off for us chumps that got born already: "I'm not talking about genetic benefits for babies - I'm talking about benefits for baby boomers". I think I mangled part of that quote, but the essence is pretty much as he intended, I think.

    I picked up two of his books and am finding them engaging. So far, I'm alarmed that the future's coming faster than I figured, and it doesn't look quite the same as I had expected.


I often hope for an idea or two to kick me out of my ruts when I go to these conferences, and I may have got some small nudges here and there from the various programming talks. The Kurzweil keynote is giving me much larger ideas to grapple with than I was prepared for. I'll have to chew on them for a while.

(3 comments | Leave a comment)

Comments
 
[User Picture]
From:ginsu
Date:February 25th, 2008 05:08 pm (UTC)
(Link)
"If you can't turn $1 million into a revenue stream of tens or hundreds of millions, venture capitalists won't be interested"

I think a more accurate phrasing would be "If you can't convince venture capitalists that $1 million will turn into..." The persuasion is the important thing, not the ideas per se, since the actual fiscal outcome cannot objectively be established (cf. Rocket Science Games).

If you buy his conjecture that advances are coming increasingly quickly, the implications seem inevitable, and perhaps a bit scary... At some point, probably in our lifetimes, the curves will meet, and we'll see computers that are as smart as humans

Well, in terms of brute horsepower, a pocket calculator is smarter than most of us. And brute horsepower is what beat Kasparov, after all -- chess is a tiny, closed system, and as such is very susceptible to brute force. Simply put the same game on a 16x16 board and I predict Kasparov will defeat any computer.

We live in a gigantic, open system many orders of magnitude more complex than that. So what makes us "smart" is not the horsepower, but a comprehension of human context -- based on a lifetime of sensory information, its organization in the brain, and a sense of possibility that comes with it.

People involved in things like the Cyc project have tried to recreate that context artificially, but I am dubious of the outcome. I think truly "smart" computers will require something more like Asimov's robot concept -- something off-the-scale-sophisticated that has senses as we have them, experiences over time similar to ours, and which can learn from them, as we learn. And I certainly don't expect that to roll along in my lifetime.
[User Picture]
From:tsmaster
Date:February 25th, 2008 07:03 pm (UTC)
(Link)
I think a more accurate phrasing would be "If you can't convince venture capitalists that $1 million will turn into..." The persuasion is the important thing, not the ideas per se, since the actual fiscal outcome cannot objectively be established (cf. Rocket Science Games).

Sure they know you can't predict an individual project's outcome - but my point is that if I'm working on a project that might give a 3x return over all time, rather than a 10 or 100x return on an ongoing basis, that's not the story that the VCs want to hear. Which is fine, there are other ways to get funding for smaller projects.


We live in a gigantic, open system many orders of magnitude more complex than that.

Agreed, and each order of magnitude requires between three and four doublings. If there's a challenge right now that's 100x or 1000x harder to play than chess, we should be able to build a machine right now that is human class at it.

Simply put the same game on a 16x16 board and I predict Kasparov will defeat any computer.

I'd expect it to be exactly the other way around - because so much of Kasparov's knowledge of chess is built up through examples on 8x8 boards, his experience will be largely useless, but the brute force approach will work as well as ever.

I think truly "smart" computers will require something more like Asimov's robot concept -- something off-the-scale-sophisticated that has senses as we have them, experiences over time similar to ours, and which can learn from them, as we learn.

Right, that's what I was saying about being able to faithfully emulate a human brain. Projections of that seem like it should be possible somewhere around 2016 to 2020.
[User Picture]
From:ginsu
Date:February 25th, 2008 07:36 pm (UTC)
(Link)
my point is that if I'm working on a project that might give a 3x return over all time, rather than a 10 or 100x return on an ongoing basis, that's not the story that the VCs want to hear.

Sure, that story won't sell. But it's also true that those numbers don't mean anything and the predictions they make aren't accurate. This is nearly always the case for creative work, which, in the end, is an attempt to appeal to the emotions of mass audiences. Not a very tidy problem of prognostication and certainly not one VCs know anything about.

If there's a challenge right now that's 100x or 1000x harder to play than chess, we should be able to build a machine right now that is human class at it.

True, but I don't think those challenges are very useful in terms of making a computer smart in the way a person is smart. They will more closely resemble other board games, such as Go, than questions like "How can we significantly diminish the threat of mass starvation as a result of overpopulation?" (Which Norman Borlaug managed to address fairly skillfully.)

because so much of Kasparov's knowledge of chess is built up through examples on 8x8 boards, his experience will be largely useless, but the brute force approach will work as well as ever.

Whereas I think Kasparov's pattern-recog skills will be almost as useful as ever at focusing on the combinations that matter, whereas the staggering increase in total possible combinations will almost completely cripple the computer. This is why, for a very long time, computers were laughably bad at endgames even on a standard board.

that's what I was saying about being able to faithfully emulate a human brain. Projections of that seem like it should be possible somewhere around 2016 to 2020

Unless someone's building a computer with senses, a computer that goes through what a person goes through in life, collecting data in the same way and drawing conclusions with the same level of accuracy (or inaccuracy), I'll be surprised to see this happen. I guess we'll know in a dozen years. =)
My Website Powered by LiveJournal.com