I.e. AutoML systems would come up with increasingly better AutoML systems, genetic programming would discover increasingly refined GP algos
The thing with recursive self-improvement in AI, is that if it were going to happen, it would already be happening
-
-
-
In an exponential loop, naturally, since that's what we were promised. Fact is, that's not how things work. End of the story.
- Show replies
New conversation -
-
-
Much like AutoML is the first thing ML people start working on after reaching proficiency, I assume the answer is yes
-
AutoML not meant to recursively self improve I thought. How about learning to learn by gradient by gradient descent? https://arxiv.org/abs/1606.04474
- Show replies
New conversation -
-
-
In your memory, have you seen a field advance as quickly as ML has the last few years? Yes of course it's driven by people, I say that's ok
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
It's happening if you include the developers are part of the system; it's less "wowwee zowee" than the fiction shows, but it IS fast!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I disagree. Current system complexity is much higher than current system understanding ability. Suspect there's a criticality threshold.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
It is happening, but so far only at one explicitly defined step. E.g. Genetic algorithms and DQNs that enhance model architectures.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This is crucial. These conversations almost always unfold the same way. "It's terrible how little policymakers understand about ML."
-
No one ever seems to consider the opposite point of view, that ML experts should understand more about how policy gets made.
- 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.