But in order to sensibly use the new discoveries you need scientists to interpret the discoveries. Sure, you put your system to work on analyzing some physical process, and it spits out some equations that can describe it. What do you do with those equations? How did you know what physical processes to point your system at?
Agreed that it changes the job of a scientist. It sort of moves science from the hunter-gatherer stage to the agricultural stage. Instead of going out and trying to understand things out in the world you’re looking at the computer analysis of stuff and trying to figure out what the computers are saying. That doesn’t mean less work for scientists, it could mean a lot more work for scientists or scientist-ish type professions.
Yeah, I read that. This way of learning games is not new. Samuel’s checker playing program became a world champion though learning - he played various versions against each other. Of course Go is a lot more complex than checkers - but the first version of the checkers program came out in 1959.
Cite.
I think it is interesting that the concept that computers can learn is still news to people. It shows that many still think that computers blindly follow preset programming, which hasn’t been true for almost 50 years at least.
What I found interesting in the article is that bit about a piece of software called “Viv” … successor to Siri, they said … that has the potential to automate all phone support and phone customer service jobs … destroying 250 million jobs worldwide. That create a bit of bother, I suspect.
I suppose there could be continuity in terms of there continuing to be people in white coats hanging around the computers and thinking of themselves as scientists. But I expect it won’t be long at all before they are about as much scientists, in the way we would think of that role now, as my three year old is an orchestra conductor because he has an iPad app that allows him to tap various sections of an animated orchestra and get them to start playing. IOW, the machines will be like “aww, it’s cute that you think you are doing the science”. Or maybe they won’t have that kind of personality and will just impassively do all the work without thinking too much about how superfluous their human attendants really are.
Sorry for responding to myself, but I just had time to read the whole thing in the dead tree edition, and it is about the dumbest thing I’ve ever seen come out of MIT, and I include the stupid porn parody we put in our dorm magazine.
It appears they seem to think Deep Blue works by indexing chess positions, and, gee whiz, there are a lot of them. Duh. They seem to think neural networks are something new and revolutionary, where all data mining tools I’ve seen, including open source ones, include neural networks.
One is a professor of management, and one is a principal research scientist, whatever that means, and at least one of them seems to have learned Basic at some time in the past and thinks he is an expert.
Well, I suspect a batch of people from the AI Lab will wander over to whatever building number houses the business school and dope slap these clowns into next week.
Whether neural networks are novel is not the point. It appears that they are saying that neural networks, or Deep Learning, have been able to become bigger, faster, etc. as computing power has increased. So the ability to handle the complexities of Go are possible now, unlike 50 years ago and checkers.
Same approach, but whole new level of capacity and performance. And as that complexity increases, we are learning how that approach represents a way to overcome that paradox in new areas of capability.
Again, I am not an AI guy - that is my read of that Op Ed. If incorrect, or if the AlphaGo innovations are different, I would appreciate hearing about it.
My argument is with the column, not with the researchers, who certainly know better. The column’s point is that “this changes everything - computers can learn!” as if this were new.
As we get more computing power and refine search and learning strategies we can play more complex games - tic-tac-toe in the '50s, checkers in the '60s, chess in the '80s, and go today. Learning the rules of go is cool, but my understanding is that go is a very complex game with very simple rules - so I’m not sure how relevant this is to the real world.
Yeah, the rules of Go are pretty simple. In fact I can teach you the rules in a few minutes which is one of the things that, imho, makes it beautiful. However, I am with you on the not sure how big of a deal this is going to be in other areas. The reason is that the A.I., while impressive, is probably not flexable without retraining.
What I mean by that is this. Take Go which has a rule for a position called Ko. In a Ko, each player can take then retake the same stones forever. To get around this a rule was put into place which is rather simple. If player A takes a Ko, then player B cannot make his next move to retake the Ko. This gives Player A the time to resolve the Ko without running into an infinite loop. If the Ko position is big enough, it can dominate the whole game where A takes the Ko then B makes some other big threat, A responds to the threat and B takes the Ko back. Then A starts it again by trying to make a big enough threat to B so that B will have to ignore the Ko allowing A to retake it. Lather, rinse repeat. This is called aKo fight.
Anyway, I suspect that with AlphaGo if you changed the rules to where, for example, if a Ko arises the person who took the Ko has to resolve it on their next move by rule, then AlphaGo would have to go retrain to learn how to play with the new rule. A highly ranked player would be able to adjust to the new rule pretty much instantly and weaker players probably wouldn’t have many problems with it either.
Of course, I could be wrong about AlphaGo and how it would handle the situation.
Although players would have no issue with learning the rules of Go-no-ko (let’s call the modified version that ;)), they would not play their best game of that version immediately.
Players who are masters of ko fights would have to jettison such strategies, and instead adapt to a reality where losing players can sometimes force a draw by repetition.
Meanwhile AlphaGo’s level improved at a spectacular rate in the last few months between beating the euro champion and lee seedol (which, as others have pointed out, is the reason for mr seedol’s over-confidence: based on its past games it was not that strong a player).
It will learn go-no-ko strategies very quickly: at this stage, I’d be willing to bet faster than humans will.
As a final point, in the example you’ve picked the game has an additional way of terminating: draw by repetition (or if go-no-ko doesn’t allow draw by repetition, infinite repitition). This makes the game somewhat easier for AI, because it will prune some of the search branches.
I thought it was supplied a few posts above mine, but now I don’t see it. It was from another thread, someone edited it out of their post, or I hallucinated it! Sorry. :smack:
In this case, I wouldn’t say that it’s a sign of the future, and -oh myGod- the robots are replacing people.
I would look at it like any other cased layoffs at a factory. They were losing money, so they just laid off the night shift. (or , in this case, the shift of workers who were not strictly needed and who did not produce any directly measurable income for the company.)
Yes, they try to spin it as a postive thing, by advertising "look!!! you can use our website!!!
(But the website is probably just a standard, off-the-shelf financial calculator progam that you could google by yourself anyway.)
The bank just decided to cut costs by not offering perks to the customers who don’t have enough money to be of interest to upper management. Note that they did not lay off the advisors who deal with the rich customers.
I heard it recently prognosticated that it’s possible that by 2030 (that’s 2030), 90% of the global workforce could be supplanted by machines. That’s not just ATM’s and junk food drive-throughs – that’s practically everything, save for a few sectors that inherently require human interaction (…at least until “A.I.” is born).
If this seemingly extreme forecast is even somewhat accurate, we’re all going to be experiencing mass redundancies and in a turnover time frame that’s far too short for societies to adjust to. This will result in societal instability and malaise. That is, unless governments institute pre-emptive legislation to protecting the jobs under threat. For we know corporations will always find the path of most “productivity” (read: cost-cutting) - due to the universal human foible dubbed “greed”, cosseted by capitalism - and so cannot be relid upon to transition any such potential seismic shift smoothly. Just look at Detroit for how well it has “transitioned” from ‘The Motor City’ of the 70’s to something that today more resembles the Dystopic depictions in films like The Road…!
Much like global warming modelling, science does not have the precedent to accurately gauge these future events; which means that events which may seem like a hundred years away today, could well be only a couple of decades in in the rear view mirror. Especially when it comes to computing – it’s progress can and generally is exponential, almost always outstripping conservative estimations of what is achievable (see: governments playing catch-up with hackers and cyber warfare). If, for example, quantum computing emerged, it would be a whole new ball game – what seems impossible measured against today’s tech and relative progress, can be completely blown out of the water with computing power / speed the likes of which said advancement will yield.
I highly doubt robots will takeover peacefully. Not so much that we’ll be meeting them at our respective ‘Conord Bridges’ (…though, that may happen centuries down the track), but rather that we’ll likely be fighting each other. For we all know what people without anything to do (i.e., purpose) are like and susceptible to… and I’m quite certain the Donald Drumpfs of the world won’t be rushing to offer $100K handouts to the proletariat to keep them happy, contented and in decent living standards.
