This doesn't really describe how development is going to get faster. It just lumps the next big leap into unsupervised learning - which isn't really an understood concept as the author describes.
The "unsupervised learning as black box" argument always bothers me. It looks too much like the consciousness debate. The theory that UL is an emergent property from massively scaled NAIs is at least plausible IMO, but that wasn't really discussed.
(One challenge: though neural networks generalize very well, we still lack a decent theory to describe them, so much of the field proceeds by intuition. This is both cool and extremely bad. “It’s amazing to me that these very vague, intuitive arguments turned out to correspond to what is actually happening,” says Ilya Sutskever, research director at OpenAI., of the move to create ever-deeper neural network architectures. Work needs to be done here. “Theory often follows experiment in machine learning,” says Yoshua Bengio, one of the founders of the field. Modern AI researchers are like people trying to invent flying machines without the formulas of aerodynamics, says Yann Lecun, Facebook’s head of AI.)
One of the most interesting aspects of the field - we don't have robust ways of predicting what will work without trying it.
There's no real guarantee that current success with neural networks will continue to ramp up all the way to hard AI. They're impressive, and getting more interesting all the time, but the downside of the whole "though neural networks generalize very well, we still lack a decent theory to describe them, so much of the field proceeds by intuition" is that we don't know when they'll suddenly stop being useful for some problem.
I agree that the road to hard AI is still foggy, but the applications of the current technology by itself has only scratched the surface.
For example, since Deep Learning has already shown great advance in the audio-visual tasks, the entertainment business which heavily utilize those senses will get a boost all around.
5 comments
[ 110 ms ] story [ 483 ms ] threadThe "unsupervised learning as black box" argument always bothers me. It looks too much like the consciousness debate. The theory that UL is an emergent property from massively scaled NAIs is at least plausible IMO, but that wasn't really discussed.
One of the most interesting aspects of the field - we don't have robust ways of predicting what will work without trying it.
An apt analogy, given that flying machines were designed by individuals without a robust understanding of fluid mechanics.
For example, since Deep Learning has already shown great advance in the audio-visual tasks, the entertainment business which heavily utilize those senses will get a boost all around.