WorldCon, part 4 -- Friday panels, etc.

[alexxkay]

Got to the hotel fairly early, and checked baggage.

11:00 Locus Awards
Presented by Connie Willis, who had a very silly "PowerPoint presentation" (actually cardboard signs) of Locus 'history'. I particularly liked the one on the Ark, "Flood likely to boost demand for disaster SF". She mispronounced Cory Doctorow's name, though, to her great embarrassment. Even *worse* embarrassment when he mentioned that, at an earlier awards ceremony that she'd hosted this year (which he wasn't physically at, but he called a fan's cellphone from somewhere on the other side of the planet), she had mispronounced the name of his winning story. Neil Gaiman cleaned up, winning awards for Best Novella, Short Story, and Non-Fiction Book. He said that discoverng that "Sandman: Endless Nights" is actually non-fiction will haunt his nights forever.

EDIT: Forgot to mention Terry Pratchett's acceptance speech for "Best YA Novel" for The Wee Free Men. First, I had no idea he had such a silly-sounding voice. High-pitched, and with a bit of a lisp. After explaining how he'd written a book full of, basically, violent Glaswegian soccer hooligans, and which moreover featured witches as positive, respected characters, he was greatly disappointed to be winning awards and getting library recommendations. "Won't anybody have the common dethenthy to *burn* it?!"

12:00 The Future of Forensic Evidence
Mildly interesting. Was more about current-day cutting-edge than future speculation, so I already knew much of what was discussed.

1:00 Interactive Fiction: The Nexus of Storytelling, Simulation and AI in Video Games
I had hoped to crash this panel, but I showed up a bit toolate to gracefully attemt that. This being a topic near and dear to my heart, I didn't actually hear much new. It was distressing that the panelists, despite saying that they wanted to see new and different paradigms of storytelling, mostly weren't interested in discussing anything that wasn't, at root, a fantasy RPG. They were totally unreceptive to looking at The Sims as an example of successful interactive storytelling, for example. Chatted a bit with learnedax afterwards about automated story-generation, then dragged him along to...

2:00 The MIT Media Lab: A Visit From the Future
Naturally, kestrell and I *had* to go to this one. Marvin Minsky apparently thinks he's just a few years away from building a "learning machine". Favorite observation (paraphrased): "Emotions are easy, they just involve selectively shutting down parts of the mind. It's thinking in the first place that's hard."

3:00- ~5:30 was dealing with hotel SNAFU, but I did catch the last bit of "Drunk on Technology". Sadly, my brain was so fried by the hotel problems that none of it went into permanent memory. Oh wait, there was one thing that Cory Doctorow said he wanted (I forget if this was a prototype device or just a thought experiment): a toaster with an internet connection, that would toast your bread with a picture of today's weather on it.

After that was dinner. After that, I hung out in the ConCourse with various Carolingians for a while. Was pretty brain-dead, though, so went to bed early

Comments

[metageek.livejournal.com]

Marvin Minsky apparently thinks he's just a few years away from building a "learning machine".

Did he give any explanation why he thinks so?
(<sarcasm>And how long has he thought this?</sarcasm>)
I mean, one traditional flaw of AI research is that the Great Breakthrough is always claimed to be a few years away.

On the other hand, part of the reason for this is that, by the time a Revolutionary Breakthrough happens, it's just an Incremental Improvement, and the target gets moved again. Sort of the opposite of "if you can't win, change the game". Did Minsky explain just what he expects the learning machine to be able to do?

[alexx-kay.livejournal.com]

He didn't go into any great detail. It was clear that he wasn't expecting human-equivalent right off the bat. I think the actual claim is that he'll have a machine that models his Society of Mind (http://www.amazon.com/exec/obidos/tg/detail/-/0671657135) concept at a sufficiently high level of detail that it can be reasonably said to be 'thinking'. Getting from that to human-equivalent (or further) requires, among other things, building an enormously huge knowledge base. He's already got a fairly large one (and is working with the CYC people to license theirs), but it's still much smaller than an actual human has.

I know that many people have made premature claims about the Great Breakthrough. I tend to give Minsky more benefit of the doubt, since I think he fundamentally groks how minds work much better than most people.

[jducoeur]

Hmm. Maybe, but I've always been a bit skeptical about Minsky's approaches -- they're interesting, but I don't really believe in them. Anything that pre-supposes a logical rulebase doesn't strike me as having anything much to do with the way humans actually think. (Pattern-matching. It's all pattern-matching, all the way down.)

It'll be interesting to see what he comes up with, though...

[alexx-kay.livejournal.com]

I don't think Minsky actually disagrees with you. He's talking about a *knowledge* base, and what he means by that is something considerably fuzzier than a rulebase.

[jducoeur]

Perhaps. I'm still dubious that anything stated even semi-formally is actually going to make good pattern-match material in the right sense -- it appears to me that humans learn almost entirely by example, and stating those examples in a way that works in a human-like way seems hard unless you're creating essentially pseudo-senses.

I *firmly* believe that learning has an great deal to do with the senses, and with neural feedback from the higher cognitive layers to the lower sense-specific ones -- hence the crucial role that visualization often plays in learning, for example. I believe that the first serious learning machines will be highly sense-oriented, even if those senses don't match our own. I'll be curious to see if Minsky is taking that into account...