@ylecun
Humans and animals are not built from scratch; both are born with varying degrees of instincts and mechanisms. Neural nets aim to be similar to a brain but without any of the foundational aspects.
@ylecun
Very interesting. Tho for humans we have 247 stimulus and a lot more sensors / information input for learning. Anda lot of experience takes years to build up. Wonder if in future possible to emulate animal experience on a computer with enough compute.
@ylecun
Recent advancements in ML and AI allow us to solve a large number of problems we couldn't solve before. However we need to be realistic of what is really possible and what is not and acknowledge that we still have a long way to go to achieve real intelligence.
@ylecun
If you want to speed up learning you need pain/pleasure valuing.
Pain and pleasure are simply approach and avoid characterizing inclinations. Pain is an avoid reaction. It has features that include information why a pattern in data is undesirable. Pleasure is the opposite.
@ylecun
Humans and animals' intelligence is a byproduct of their intrinsic symbiosis of all signal emitors and sensors. Today's AI is nowhere near any of that.
@ylecun
i will still be called "machine learning" ?
all this terms should be rebranded
ML : trained algorithm (TA)
AI: just call it algorithm
AGI: ungovernable by human algorithm, but observable
ASI: ungovernable and unobservable algo