An algorithm problem.
Interviewee, Isaac Faber: https://www.linkedin.com/in/isaacfaber/
Isaac Faber outlines "There's overlap, but ML is a set of techniques that allow you to feed experience into a machine; and, AI sort of performs tasks traditionally the purview of a human."
Feeding experience into a machine? Folks, this one's on reinforcement learning.
In psychology, 'intelligence' is mostly defined as the ability to react to 'new' situations.
Humans do this well in part due to neural pathways connecting distant brain regions, as though multiple minds 'assess' a situation and simultaneously compare notes.
In computer science terms, it seems like chaos wired in, 'deep learning' where outputs connect to inputs at all other layers.
"Is there anything like that in c.science?"
"Yes, this is the territory of reinforcement learning. Supervised algorithms are pretty simple - almost college maths... but with reinforcement learning it falls off a cliff!"
"Alpha Go only needed to 'remember' the last turn of the game. But it still required incredible computational resource."
We talk energy comparisons in human vs machine, the implications - the human brain has a much greater 'computation' to energy ratio.
"I don't think this is a computation problem, it's an algorithm problem. For human level intelligence, at some point the underlying mechanics have to transcend."
Comments