Not exactly about books (except for scifi), but, for PG, an interesting topic.
From The Wall Street Journal:
Over the centuries, humans have created countless technologies to save ourselves from backbreaking physical labor and mindless routine. AI continues that progress by taking over many rote cognitive tasks that don’t require human judgment, strategic planning or creativity. Whether it’s browsing millions of legal documents or scrutinizing CT scans, machines can now do much expert work faster and more precisely than their human creators. New forms of artificial intelligence will surpass us in new and surprising ways, thanks to machine-learning techniques that generate their own knowledge—and even their own code. Humans, meanwhile, will continue up the ladder to management.
We’re not being replaced by AI. We’re being promoted.
. . . .
Our outdated vocabulary stokes our fears. “Artificial intelligence” sounds like an unnatural rival of our own, as an artificial sweetener is to sugar. We should instead think of AI as “augmented intelligence.” Our increasingly intelligent machines are making us smarter, just as our past technology—from pulleys to hydraulics, and sailing ships to rocket ships—made us stronger and faster. For the first time, machines aren’t just giving us answers more quickly and accurately. They are generating new knowledge that helps us better understand the world.
. . . .
In December, Demis Hassabis at DeepMind, an artificial-intelligence unit of Google parent Alphabet Inc., unveiled his latest chess program, the first of its kind. The project, called AlphaZero, is a generic machine-learning algorithm with no chess knowledge beyond the rules. After playing against itself for several hours, AlphaZero crushed one of the world’s strongest traditional programs, which, like every other successful chess program in history, had been programmed with existing human knowledge of how best to play the game.
AlphaZero’s domination was produced with no opening library of moves, no human input about the relative value of the pieces—nothing at all. This unique creation generated its own knowledge to become the strongest chess-playing entity in history; human experience would have held it back. For decades, better programming and processors led to incremental improvements in AI. AlphaZero was a sudden leap—the sort we should expect to see more of as machine-learning models move into disciplines like cancer screenings, asset management, law enforcement and education, to name a few. With only minor exaggeration, in four hours AlphaZero taught itself all the chess knowledge—and more—that computer science had long assumed would come from human masters.
Link to the rest at The Wall Street Journal
Sadly they’re finding that they are programming the programmer/user’s faults/flaws into the AI as well …
And that will be what will then teach ‘you’ …
That’s a happy thought. May it be so!
I’m no chess expert, but I suspect that prowess in chess is in direct proportion to the ability to think ‘ahead’. As humans, we can only figure out a very limited number of possibilities into the future. A machine, however, should be able to map every possible move.
I don’t have the math to work it out, but the equation for number of possible moves in chess should be something like:
16 [pieces] x 64 [squares] x by direction [the piece can move] x special moves [jumps etc]
Hmm…okay, my head is spinning just thinking through the logic, but a machine intelligence would be able to do the math. The fact that AlphaZero took a number of hours to become the best chess ‘player’ in the world is testament to how many permutations it had to process before if was ready to play.
I suspect the only game an AI would not be able to win would be one based in part, or entirely, on random chance.
They often set those systems to only think so many moves ahead, otherwise a human would have no chance at all.
lol – that makes sense. So essentially they just took the brakes off. 🙂