I have a news flash for the author of this article, Korea is not the only country where people have a concern about this!
Can somebody please explain why after 50 years of attempts by computer scientists, producing human quality translations from one language to another still seems at least 50 years off? A Spanish teacher neighbor told me that she tells her class, "Don't you dare use Google Translate to do your homework! I can tell immediately". I have been told that it doesn't even do that good a job between Spanish and Portuguese.
I was friends for years with a genuine genius polymath. His name was Jack Margolis, a radio talk show host (Arbogast & Margolis) on KLAC in L.A. He was a prolific author, Child's Garden of Grass his biggest hit, and he wrote for Laugh In, the innovative comedy TV show and was an insider at the Playboy Mansion. He was a big time Go player, frequently besting the best in L.A. All this was in the '60s. When a computer program can match this breadth of intelligence, I'll throw down my stone.
AlphaGo is a deep neural network (13 layers) but only has small numbers of neurons in each layer (around 400). The human brain has far more connections (around a quadrillion) and it seems unlikely that current or even foreseeable technology will allow nets of that size to be updated in real time. Whether the brain has additional tricks up its sleeve remains to be seen, but even without them the gap is formidable.
If intelligence is nonlinear, some great advances in mathematics are needed for here to be a problem. If finite difference approximations are used to the model the nonlinear mathematics, artificial intelligence will contain too many errors to be truly autonomous.
With the explosion in information and knowledge, my understanding at least, is that artificial intelligence will become essential, so we can spend more time in what is really human, becoming emotionally vulnerable to experience love towards each other, to be able to play for the heck of it, to be be able to 'be' instead of enslaving ourselves in the acquisition of things. Perhaps, once the 'machines' free ourselves from physical labor and from routine tasks, a way to start using our imagination, a creative way to assert ourselves, by reading (and writing?) poetry, and enjoying music and the arts, and travelling to embrace other cultures, to educate ourselves in inclusion, and of the evils of intolerance, of the ability to transcend, to acquire wisdom. This may be the century of a new 'renaissance', provided we don't allow indifference to the ills and violence in the world to prosper. By all means, let artificial intelligence, as deviced by enlightened human souls, contribute to our freedom and even to a more just society.
1
And John Henry could not outlast the steel driving machine, a horse cannot outlast a car, a diver can't outlast a submarine...
3
Everyone I know in South Korea has been consumed by Go fever since this challenge match started. When AlphaGo defeated Lee Sedol for a fourth and final time, all their hearts sank. One Korean English daily opined that the challenge match was unfair because Lee Sedol was an undeniably smart, expert
player while AlphaGo had been programmed with centuries of Go knowledge, enjoyed Google's Cloud computing power and was supported by a small army of very smart people. Four matches to one doesn't sound very good, but I think Lee Sedol did a very good job to find AlphaGo's weaknesses in one game and exploit them to win that game. Bravo, Mr. Lee.
player while AlphaGo had been programmed with centuries of Go knowledge, enjoyed Google's Cloud computing power and was supported by a small army of very smart people. Four matches to one doesn't sound very good, but I think Lee Sedol did a very good job to find AlphaGo's weaknesses in one game and exploit them to win that game. Bravo, Mr. Lee.
3
Can we be sure that a very odd/unexpected move in "Go" might
cause the Computer to stumble ?
cause the Computer to stumble ?
2
Come now it is a game with rules, not a life with few that are always the same. Nothing to worry about here.
Sorry, but I'm extremely skeptical what computer can do in translating languages. "The communication of our thoughts by means of language, whether spoken or written, constitutes a peculiar art, which cannot be acquired in any perfection but by long-continued practice." -- P. M. Roget.
1
Much the same could be said about playing Go. I have a feeling Dr. Roget would be stunned at the progress so far.
This is the specter that great minds such as Stephen Hawking, Bill Gates, Elon Musk and many others have been raising: what to do with AI morphs into super intelligence? How would we control it?
Pull the plug is the easy answer of course but it's not that easy...
Pull the plug is the easy answer of course but it's not that easy...
3
I'm kind of appalled by the "us vs. them" mentality that people seem to have taken on en masse, with "them" being "machines". I also find Lee Sedol's reaction unnecessarily negative/depressing. Lee had already proven himself to be world champion in prior matches; a lot of people seem to forget that being beaten by a computer doesn't diminish that at all. It's so sad that such a great achievement in AI is being so villified by the media and the people, as if humanity's honor were at stake and was defeated -- HELLO people, AlphaGo was created by humans FOR humans, using technology that can be used to make the world a better place.
4
Kinda missing the point about fears on AI.
AlphaGo is controllable. We don't fear AlphaGo.
We fear AI that we can't control. And that becomes more and more a possibility as they get smarter.
AlphaGo is controllable. We don't fear AlphaGo.
We fear AI that we can't control. And that becomes more and more a possibility as they get smarter.
3
When computer programs are used to make the world a better place, someone must make a judgment, both before and after the feat, as to whether the act will/did make the world better. Computer software can be used to help us do things we cannot otherwise do; with the help of the proper program, I can defeat chess players much stronger than I; I simply have to make the moves the computer instructs me to make. What do we do when we grant to the software the decision as to whether we should even play the game? What do we do when we disagree amongst ourselves as to what a "better world" looks like, but the programmers have already determined a path?
Whilst AlphaGo is a good feat of software engineering, and the algorithms it has may be transferrable to other situations, it is, at this point at least, an idiot savant. It knows how to do one task exceptionally well, but it would be helpless in the face of other tasks that humans routinely perform. It is our overall adaptability that gives us the edge. This is the real hurdle for AI to clear.
10
I had lunch with a Korean colleague that was moving to South Carolina. At the table were two other coworkers from China. The conversation was all Go - I had no idea that this story was so big. It certainly was nowhere on my radar screen. For them, it was truly the new March Madness.
1
One day, "just pull out its electric plug" may become tantamount to murder or even revolution. What looks lame and harmless today looks quite different with the superior vision of hindsight.
2
AlphaGo and similar systems (DeepBlue for Chess, Watson for Jeopardy, etc.) definitely display "knowledge" but they do not have intelligence. None of these systems can generalize from their base of knowledge to wider spheres of understanding, or can even discuss or think about what they are doing when they are using their knowledge. That is not to say that truly intelligent computers are not possible (they are) but we just don't have them yet.
It is too bad that this sort of distinction is not made more often in articles about these systems, because media coverage of this sort of event usually is limited to "oh my god, are the machines are smarter than we are" or "this isn't creativity, so we are safe".
Unless you believe in magic, it is obvious that a truly intelligent computer is possible -- one that can generalize, learn, display creativity, formulate explanations, think about itself thinking, and do whatever else we might call intelligence. Obvious because what is the brain except a very complicated computer? (unless you believe it has some kind of magic in it). If the brain is not magic then it must follow the rules of physics, and it must be possible to simulate it using a sufficiently powerful computer.
It is too bad that this sort of distinction is not made more often in articles about these systems, because media coverage of this sort of event usually is limited to "oh my god, are the machines are smarter than we are" or "this isn't creativity, so we are safe".
Unless you believe in magic, it is obvious that a truly intelligent computer is possible -- one that can generalize, learn, display creativity, formulate explanations, think about itself thinking, and do whatever else we might call intelligence. Obvious because what is the brain except a very complicated computer? (unless you believe it has some kind of magic in it). If the brain is not magic then it must follow the rules of physics, and it must be possible to simulate it using a sufficiently powerful computer.
5
The remarkable thing about AlphaGo is that it does not have an extensive knowledge of the game explicitly programmed into it. Instead, it has acquired its go knowledge -- knowledge of what makes a position favorable, and what moves should be considered in a position -- by example, from training with a very large number of games by top-level human players. This makes it far from clear that AlphaGo is not in its own way generalizing from its base of knowledge to wider spheres of understanding.
1
Baloney. Of course neural networks like AlphaGo generalize, that's the whole point! The main difference from human intelligence is that our numbers of neurons and connections still greatly exceed those in AlphaGo and other similar artificial neural networks. Whether that difference will soon be transcended, and if not why not, is an interesting scientific question.
I'm 59 years old and first learned Go back in 1969 as a seventh grader. As it consumed at least one or two hours for a round of Go, I couldn't spend that much time to the game until 2001, 32 years later. Another problem was that it was hard to find an opponent of same level at that time (before the coming of internet). I was an amateur 2-kyu ("geup" in Korean) in 2001, but now play as a 3-dan (five advances in a span of 14 years). To play good at Go, you have to read Go books and play as many games as possible over a board of 19 lines by 19 lines (a round of Go generally takes 30 minutes to an hour). Now, you can enjoy the game over the internet 7/24. It's amazing that AlphaGo, developed by Google, beat Sedol Lee, the best Go player of this decade, winning the first three matches in the best-of-the-five series. It was a really fun to watch them for a week of March 9-15.
6
First, GO or Baduk is a beautiful elegant game under appreciated in the West though over the years we've been told by Kissinger and others that an understanding of GO will help us to understand how the East thinks and sees the world. The thought process is different relying, some say, on the intuition that comes from experience and the ability to recognize patterns more so than positions.
It is indeed remarkable that Hassabis and the team at DeepMind have developed an artificial thought process. I don't think we see an intelligence at work here other than that of the team at DeepMind. But the automated thought process that can run and produce answers is WOW. The application to all kinds of human problems seem endless and exciting. Of course like all powerful human inventions this can be used for all kinds of terrible things. Decisions to be made about regulation and control seem inevitable.
It is indeed remarkable that Hassabis and the team at DeepMind have developed an artificial thought process. I don't think we see an intelligence at work here other than that of the team at DeepMind. But the automated thought process that can run and produce answers is WOW. The application to all kinds of human problems seem endless and exciting. Of course like all powerful human inventions this can be used for all kinds of terrible things. Decisions to be made about regulation and control seem inevitable.
1
We shouldn't confuse the 'A.I.' of AlphaGo with the intelligence of humans!
To my understanding, AlphaGo seems more like a lower level AI, equipped with an extensive database of historical Go play-by-play analyses.
If huge amounts of information equates to intelligence, then, well yes, AlphaGo is the more intelligent Go player here, but seriously, is it 'more creative' or 'smarter,' or is it even really 'playing' Go in this match?
Some people in Korea are calling this match an insult to the game of Go, and although this match is a great spectacle, I cant disagree with those people.
Kudos to Mr.Lee on beating out the database of the entire history of Go matches, at least one time out of five!
To my understanding, AlphaGo seems more like a lower level AI, equipped with an extensive database of historical Go play-by-play analyses.
If huge amounts of information equates to intelligence, then, well yes, AlphaGo is the more intelligent Go player here, but seriously, is it 'more creative' or 'smarter,' or is it even really 'playing' Go in this match?
Some people in Korea are calling this match an insult to the game of Go, and although this match is a great spectacle, I cant disagree with those people.
Kudos to Mr.Lee on beating out the database of the entire history of Go matches, at least one time out of five!
1
As far as I understand it really is a "True" A.I. (what ever that means). It really has a "mental model" of the game that may be similar to the model that humans have inside their heads (though implemented using very different "hardware"). There could never be a database that included all the possible moves and countermoves of all the possible games in existence (as that number would be bigger than the number of atoms in the universe). So each game consists of a sequence of moves that has never been played and never will be played in the history of the human race (or the universe). How the model was exactly built I don't understand (and the designers themselves don't understand or know what it contains). People can only access the model by playing against it - which is probably true of other people's models as well, although humans are generally able to describe their moves and strategies using technical terminology as well.
One question is if these machine learning models remain a "blackbox" how can they be validated for use where they may be responsible for accidents? Of course the mental models used by humans face the same problem, and validation is considered to be something like passing a driving test. But machines should be able to tell more about what they are thinking before they can be really trusted. But does this mean they need to develop a "consciousness" to be able to do this? Hmm, maybe that would not put all fears to rest after all...
One question is if these machine learning models remain a "blackbox" how can they be validated for use where they may be responsible for accidents? Of course the mental models used by humans face the same problem, and validation is considered to be something like passing a driving test. But machines should be able to tell more about what they are thinking before they can be really trusted. But does this mean they need to develop a "consciousness" to be able to do this? Hmm, maybe that would not put all fears to rest after all...
1
Are you saying that Lee Sedol can play Go without being "equipped with an extensive database of historical Go" matches?
"Huge amounts of information" is what all players need to be amoung the best.
"Huge amounts of information" is what all players need to be amoung the best.
3
Are you saying that Lee Sedol can play Go without being "equipped with an extensive database of historical Go" matches?
"Huge amounts of information" is what all players need to be amoung the best.
"Huge amounts of information" is what all players need to be amoung the best.
3
Considering how interested institutions like Korea Advanced Institute of Science and Technology (KAIST) are in developing robots (they won the recent DARPA-robotics challenge) I wonder why they didn't develop a robotic Go champ first but then I suppose there is an aesthetic part of the game that would be missing if they created a robo-Go champ and that is, maybe, not very Korean.
3
Artificial intelligence is limited to basic computational tasks and mechanized motor capacities. However, AI can never approach real intelligence.
Take this simple challenge between human and artificial and tell me which is capable of being intelligent:
Get me a whole wheat croissant and a cup of peppermint tea! Also, organic butter on that, please.
Take this simple challenge between human and artificial and tell me which is capable of being intelligent:
Get me a whole wheat croissant and a cup of peppermint tea! Also, organic butter on that, please.
"AI can never approach real intelligence."
There's no reasoning in your comment for this thesis. You only said that it's not there at the moment.
Maybe we arent as smart as we think?
There's no reasoning in your comment for this thesis. You only said that it's not there at the moment.
Maybe we arent as smart as we think?
3
I have been in the IT business for 40 years. Lots of reasoning here! We are actually much smarter than we think. But we don't practice being smart.
No smarts needed to build a machine. Nature has no need for them.
No smarts needed to build a machine. Nature has no need for them.
3
That's actually exactly the type of intelligence people are working on. Actual natural language processing that works is considered the next holy grail of AI. You people just keep moving the goalposts and redefining what machines can't do. First it was chess, then it was speech recognition, then it was image interpretation, then Go, and now this.
3