Delicious LiveJournal Links for 3-6-2009

[andrewducker]

• http://www.guardian.co.uk/users/warrenellis
  Nice disclaimer or two :->
  (tags: funny WarrenEllis)
• Be sure to mind when you change your mind
  When people have their minds changed they remember having _always_ had the new position...
  (tags: psychology opinion memory mind)
• Stephen Wolfram announces that he's solved AI.
  Well, he pretty much does. Let's just say that I'm deeply sceptical of what he is claiming.
  (tags: computing technology mathematics)

Comments

[nmg.livejournal.com]

re: Wolfram, I'm getting a bit of a Nouveau AI-flavoured CYC feel about this.

[the-locster.livejournal.com]

uh huh. Of course knowledge by itself isn't strong AI, it's part of what is needed for strong AI. The ability to learn and construct knowledge for oneself is key but perhaps the big thing that is oft overlooked is the executive control, choosing what to learn and do and think about. Motivation and incentive, what form do these take in a machine intelligence with no need to reproduce?

Something like this could provide a pretty neat search engine which in turn accelerates all kinds of research, much like google and the internet have done in recent years. It's perhaps another step towards strong AI. Ultimately of course we still need more powerful computers, you just can't run something as smart as a human on today's (standard) computer hardware, even if it does operate in a fundamentally different way. Information processing isn't intelligence but a crapton is required for intelligence.

[andrewducker]

Converting data into knowledge is the difficult bit. "Meaning" requires AI, from my understanding. The ability to parse plain English generally requires likewise. To claim that he can capture general data, parse it into something meaningful that's also open to computer manipulation, and then use natural language to get something useful out of it? Either his claims are vastly overblown or he's solved AI...

[the-locster.livejournal.com]

"Meaning" requires AI, from my understanding. The ability to parse plain English generally requires likewise.

I agree. You need intelligence to determine context and unravel the true meaning. It's not a simple parsing but an iterative process (between high level thought and low level parsing) that homes in on (gradient descent, minimizing error, etc) the correct or most likely meaning.

I largely suspect Occam's razor applies here. Given the competing hypotheses that Wolfram has cracked it Vs. Wolfram is a nutball, my money is on option B :)

[andrewducker]

I'm with you there.

It'd be lovely to be proved wrong, of course :->

[andrewducker]

Absolutely.

[jccw.livejournal.com]

Yeah, this does sound a lot like Cyc and/or the "Semantic" Web reheated, neither of which are anywhere near to working. The problem with Wolfram is that he seems even more susceptible to "not invented here" syndrome than most smart people.

I was just talking to an AI prof here who said something interesting about how the Cyc people started out being totally ad hoc and hacking away and eventually *needed* to adopt better principles for modular and higher-order reasoning - all those things that us crazy PL and logic researchers have been going on about.

[andrewducker]

Yup, absolutely. Ontology is another part of the hard problem - so far only solvable by people.