Whether an AI that plays StarCraft, DotA, or Overwatch succeeds or fails against top players, we'd have learned nothing from the outcome. Wins -- congrats, you've trained on enough data. Fails -- go back, train on 10x more games, add some bells & whistles to your setup, succeed.
-
Show this thread
-
Replying to @fchollet
Bad take, Francois. There is value in creating AI that can outperform humans on complex tasks even if we truly learn nothing from that engineering effort. At a minimum, those successes inspire and move the bar for AI ever higher.
1 reply 2 retweets 27 likes -
Replying to @randal_olson @fchollet
Your take also misses the fact that we still learn from those herculean engineering efforts. High-quality training data doesn’t magically appear; you have to learn how to create it en masse. Algorithms that can learn complex control tasks at scale don’t magically exist; we (1/2)
1 reply 2 retweets 16 likes -
Replying to @randal_olson @fchollet
must invent and optimize them. Hardware that can support such a task doesn’t appear out of nowhere; we must build it. Much is learned along the way to solving those problems. (2/2)
1 reply 0 retweets 8 likes -
Replying to @randal_olson
For sure, such an endeavor can be valuable in terms of PR, inspiration, and as an engineering exercise with positive 2nd-order effects. My point from the perspective of learning something about intelligence, specifically. The science of AI.
2 replies 1 retweet 12 likes -
Replying to @fchollet
That statement applies to most AI research nowadays IMO. Anything involving deep learning is highly unlikely to advance AGI research, and that’s OK. Core AGI research is akin to walking through a maze blind-folded because we’re trying to mimic something we don’t understand.
2 replies 1 retweet 18 likes
I don't think so. The purpose of science is to generate knowledge, not to generate PR. AI research is a science, and its purpose is to produce knowledge about generalization. Most deep learning research these days at least *attempts* to do so.
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.