Actually, porn detection works quite well on Facebook. No detector can ever be perfect, but it does a pretty good job. The Tumblr story is simply another case of "you can always do it wrong".
AI riddle: why is playing chess a solved problem, yet detecting porn is a huge mess, and autonomous driving still in beta, human safety drivers required, after decades of research?https://www.wired.com/story/tumblr-porn-ai-adult-content/ …
-
-
- 4 more replies
New conversation -
-
-
I think that it is possible that the people that Verizon employs to make tumblr Apple store compliant within 2 months don't play in the same league as those that Google employs to solve Go within 2 years.
-
Sorry
@plinz, but that won’t cover the driverless car case. - 1 more reply
New conversation -
-
-
2/ Second, the bounded state space of games gives current AI techniques some edge, but humans actually do quite well in unbounded state spaces due to symbolic abstractions, which AI has not mastered yet.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Not a riddle unsolved, imho. Playing chess variables are a bounded set, finite space prob. Autonomous driving
#AI is unbounded w/ infinite space prob. I.e. N space calculus far, far more difficult than say 5 space or 6 space. Buck's Adv Calc is generally taught in grad school.
pic.twitter.com/BGNrngGjPi
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Data. In Chess, Go and many boardgames, you can play zillions of games (i.e. 'interactions with the environment') for basically zero cost and time. For real world problems, you need to be far more data efficient as you don't get to have so much free feedback from the environment.
-
That's also why I don't agree that Chess is solved. We can beat a human. But how many interactions do we need to have for that vs an expert player to become an expert? I'm guessing the latter is much lower, as expert generalization inference per sample is superior to AlphaZero's.
End of conversation
New conversation -
-
-
Chess is a very well-defined, constrained problem that can be easily formulated in terms of mathematics/combinatorics. The other problems require understanding of the physical reality and human behavior, on the level that is not easily captured mathematically.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Prof, If the agent is still losing few games against it's competitor, how can we call it a solved game?
Thanks. 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.