A voice of reason at the BigNeuro panel: "we are very very far from human-level AI... maybe decades or centuries" - Yoshua Bengio #NIPS2017
-
But we can only make predictions when we have enough data and sufficiently reliable models. When it comes to "solving intelligence", we have basically no data about what it entails, and no way to model progress towards a goal we can't even define. We don't know what we don't know
We'll know it's getting closer when we start developing proper definitions of the goal and reasonable ways to model progress. Since we don't have that today, we know we're far, but we don't know anything more
-
-
In particular we don't know "how far". Like walking towards a place that's beyond the horizon line. We'll know we're getting close (and how close) only when it's in sight.
-
I take it to mean even the experts have absolutely no idea how and when we "solve" human intelligence. We've seen this story so many times before.
End of conversation
New conversation -
-
-
For predicting specifics, yes, agree. For getting ready though, broadly speaking, a few decades is very little time. Though I’m not sure what “solving intelligence” will even mean, I’m pretty sure tremendous effects of this process are already here, and accelerating.
-
Hard problem because intelligence evolved without the goal of becoming intelligence.
End of conversation
New conversation -
-
-
Sure good definitions would help, but will you keep saying “we know same as 2017” as 2027 AI automates a number of tasks not conceived possible now, because you haven’t developed better definitions?
-
There is an ocean of difference between automating vertical tasks (what you can specify, you can automate), and "general" intelligence or "human-level" intelligence. Progress on the former does not entail the latter
- Show replies
New conversation -
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.