What neural nets do is approximate a mathematical functions: a mapping from a set of bits to another. I can see the artistic styles used in the experiment being expressed as a mathematical transform: apply some filters, turn a few knobs on your image editor.
-
-
Replying to @norswap @slatestarcodex
AI has not progressed constantly but rather by fits and starts. And each time the view of what was possible has been over-optimistic, on the "natural language solved by next summer" scale.
1 reply 0 retweets 2 likes -
Replying to @norswap @slatestarcodex
What we see currently, I think: the low-hanging fruits of a new technique (actually an old technique finally made practical) and the higher-hanging fruits of painstakingly combining it with other well established AI building blocks (Alpha Go).
1 reply 0 retweets 2 likes -
Replying to @norswap @slatestarcodex
Computing is, as a rule, not that surprising. Solving Go is certainly not outside the realm of the imaginable. Neural nets were conceived decades ago. But the vaguely-defined superintelligence, that is something else, more akin to making you 20 mph cars reach Mach 1.
1 reply 0 retweets 2 likes -
Replying to @norswap @slatestarcodex
Many things that the FAQ describes it as capable off, I think can be achieved - even with today's ideas given enough work. But to think of techniques that would enable all of them, and go beyond, I haven't an inkling how that could be done. No one has.
1 reply 0 retweets 0 likes -
Replying to @norswap
Everything in AI seems boring once it's been accomplished. People used to say that beating humans at chess would mean an AI had true creative thought; now even making beautiful art doesn't. When AIs are superintelligent, that'll look like "just math" too.
1 reply 2 retweets 5 likes -
Replying to @slatestarcodex @norswap
Although there has definitely been a lot of hype about AI, a lot of the "natural language solved by next summer" style stories are urban legends. See https://www.openphilanthropy.org/focus/global-catastrophic-risks/potential-risks-advanced-artificial-intelligence/what-should-we-learn-past-ai-forecasts … .
1 reply 0 retweets 1 like -
Replying to @slatestarcodex @norswap
My impression from talking to AI researchers is people generally underestimated pace of AI progress over past 5-10 years. EG just before AlphaGo, people predicted would be decades before AIs could beat people at Go. Some aspects of vision/speech recognition equally impressive.
1 reply 0 retweets 0 likes -
Replying to @slatestarcodex @norswap
I agree can't make superintelligence today, requires new insights. But to stick to your metaphor, only 40 years between 20 mph Model T and first Mach 1 flight. Good forecaster in 1910 wouldn't have said "There's no way to reach Mach 1 now, so probably we never will".
1 reply 0 retweets 1 like -
Replying to @slatestarcodex @norswap
I'd be astounded if we never got AGI; idea that the brain can't be duplicated by technology sounds too supernatural. Only ? is whether in 40 years or 400. Most experts predict closer to 40.
1 reply 0 retweets 2 likes
Not saying they're definitely right, but seems crazy to stake everything on assumption that they're definitely wrong without investigating.
-
-
Replying to @slatestarcodex
I still don't agree, but I don't think there is enough evidence to warrant much more discussion. Clearly, there are worse things to spend your money on. I'm a bit curious on how efficient countermeasures can be devised without pinning down what that elusive superintelligence is.
0 replies 0 retweets 0 likesThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.