It’s official: The world champion of Go, Lee Sedol, will face off against Google Deepmind’s powerful artificial intelligence, called AlphaGo.
A week ago, Google DeepMind’s team claimed to have built the best AI for the game in a scientific paper, and issued a public challenge to Sedol. DeepMind hopes it can prove that a powerful artificial intelligence is capable of beating the best human player in one of the most mathematically complicated board games ever created.
The match between Sedol and DeepMind will be broadcast live on YouTube beginning March 9. The prize for the event is $1 million.
[...]
AlphaGo already successfully defeated three-time European Go champion Fan Hui in a series of five matches, but Sedol is considered to be a different challenge entirely. Sedol has dominated the game of Go throughout his professional career, earning him recognition as one of the great players of the modern era. His match against AlphaGo will be one of the most difficult challenges of his life.
~~~~
Match days: 9, 10, 12, 13, 15 March - will be livestreamed on YouTube. More details soon. We are very excited to be coming to South Korea!
~~~
COMMENT: Not that a forecast changed on March 9th will really have a major effect on one's score for this IFP, it'll certainly be telling to see how that first game falls (and if this ends up going to the "fifth" game before a winner can be determined -- it might help one's score to update one's forecast at the conclusion of that game).
How much did you think about this AlphaGo prediction? Would you be willing to wager money in a bet? If so, at what odds?
I am trying to figure out how much I can learn from your 8% prediction given your superforecaster status. You are clearly great at forecasting, but maybe in this particular case you didn't think as carefully about it as some of the other people estimating 60+% that AlphaGo will win (?), so I'm not sure how much I should believe your low estimate and lower mine from what it otherwise would be. Thanks.
Good on you for questioning the validity of the forecasts. I should tell you, though, that this wasn't my first forecast on this IFP. This was --> https://www.gjopen.com/comments/comments/118800. Because of the way the UI is for this site, it's not always easy to see the history of an individual's forecasts on a given IFP. Often times, I'll write (lowering/raising or no change) at the top of my forecast because I'm usually updating a previous forecast on a given IFP. Similarly, I often link to past forecasts on an IFP when updating. For this forecast, however, I didn't.
To answer your questions:
1) How much did I think about *this* forecast? Probably best to answer for (all) my forecasts on this IFP because I have two (so far). I'd say... hmm, I dunno.... maybe 45-75 minutes? I'm also including every time I read an article since I've made this forecasts (that are about this IFP). Google alerts is very helpful, in this regard. Also, just because I'm not actively thinking about an IFP doesn't mean some part of my brain isn't processing it, so number of minutes spent thinking about an IFP might not be the best way of gauging. Actually, to that end, someone who spends over 6 hours on an IFP might not be the best person to come up with an answer either (information overload and all that, but that's a conversation for another thread, I suppose).
2) Would I be willing to wager money? Hmm, no. I understand what you're trying to get at with that kind of a question, but GJ is different from gambling/odds. Is there even an oddsmaker out there right now who's taking bets on this? If there isn't, I'm sure there will be at some point. With gambling, you're trying to get your money in with the best odds. With forecasting, you're trying to predict the likelihood of an event. Sure that's very similar to gambling, but one quick example of how that's different: if someone gave me 100 to 1 odds that the Broncos would win by 14 points, I probably would have taken it (even though all of the "math" was pointing to a probable Panthers win).
It sounds like you're trying to ask me how confident I am of this forecast. Well, when I made it (https://www.gjopen.com/comments/comments/118800), I was relatively confident and I still am mostly confident in it. Of course with all the forecasts going in the other direction, it's natural for one to doubt one's self. The consensus isn't always right, though -- otherwise, there'd be no reason to discriminate between superforecasters and forecasters and dart throwing chimps. If you'd like a couple of examples...
Eye-balling the consensus on that IFP over the lifespan puts it at about 30%. It resolved yesterday at 100%. (I was one of those who was forecasting closer to 100%).
For most of the lifespan of this IFP, Trump was the candidate pegged to win (by plurality). He finished with about a 40% chance. In fact, Cruz's chances were the highest towards the beginning of this IFP and were at (his) lowest when the IFP closed. (I picked Cruz to win.)
The consensus on this one was over 85% for almost the first three months of the IFP's existence and only dropped to 75% with about 2 weeks before the IFP's closure. That was a BIG swing/miss by the consensus (myself included! as I was in for 100% over the lifespan of the IFP -- oops).
~~
"2) Would I be willing to wager money? Hmm, no. I understand what you're trying to get at with that kind of a question, but GJ is different from gambling/odds. Is there even an oddsmaker out there right now who's taking bets on this? If there isn't, I'm sure there will be at some point. With gambling, you're trying to get your money in with the best odds. With forecasting, you're trying to predict the likelihood of an event. Sure that's very similar to gambling, but one quick example of how that's different: if someone gave me 100 to 1 odds that the Broncos would win by 14 points, I probably would have taken it (even though all of the "math" was pointing to a probable Panthers win)."
I think I see what you're saying: whether it makes sense to bet depends on what odds you get.
But if you are estimating ~8%, then wouldn't betting on Lee Sedol with, say, 3:2 odds clearly be favorable (e.g. If Sedol wins, you get $2, if AlphaGo wins, you pay $3)? Your expected earnings from this bet would be positive if the probability of AlphaGo winning was <40%. And your estimate is ~8%, which seems definitely less than 40%.
So I'm confused: Would you be willing to wager (a small amount of money, such that if you lost, it wouldn't affect you, so no reason to be risk averse) at these odds?
(Note: I'm not actually proposing a bet with you. I'm just wondering why you wouldn't be willing to accept such a bet if you really believe the probability is ~8%, and very clearly less than 40%.)
From an odds standpoint, absolutely -- with the way I've set my forecast, 3:2 odds would be a no-brainer bet. To answer your question, would I be willing, yes, I would be willing (but I won't).
Another point on "8%" -- GJ is *technically* a game and so part of my forecast of 8% is playing the game.
Also, this comment (https://www.gjopen.com/comments/comments/128202) from @nicholas might be more grist for the (negative) on this IFP. It looks like the analysis is pretty solid, but that the consensus is dragging the forecast up.
No change.
TIL --> This IFP could have resolved on a technicality (i.e. in the negative, as Sedol could have declined the invitation?)
Feb 5: "Google AI Will Compete Against Go World Champion Lee Sedol On YouTube" --> http://gizmodo.com/google-ai-will-compete-against-go-world-champion-lee-se-1757289813
It’s official: The world champion of Go, Lee Sedol, will face off against Google Deepmind’s powerful artificial intelligence, called AlphaGo.
A week ago, Google DeepMind’s team claimed to have built the best AI for the game in a scientific paper, and issued a public challenge to Sedol. DeepMind hopes it can prove that a powerful artificial intelligence is capable of beating the best human player in one of the most mathematically complicated board games ever created.
The match between Sedol and DeepMind will be broadcast live on YouTube beginning March 9. The prize for the event is $1 million.
[...]
AlphaGo already successfully defeated three-time European Go champion Fan Hui in a series of five matches, but Sedol is considered to be a different challenge entirely. Sedol has dominated the game of Go throughout his professional career, earning him recognition as one of the great players of the modern era. His match against AlphaGo will be one of the most difficult challenges of his life.
~~~~
And this tweet is important --> https://twitter.com/demishassabis/status/695379217870008321
Match days: 9, 10, 12, 13, 15 March - will be livestreamed on YouTube. More details soon. We are very excited to be coming to South Korea!
~~~
COMMENT: Not that a forecast changed on March 9th will really have a major effect on one's score for this IFP, it'll certainly be telling to see how that first game falls (and if this ends up going to the "fifth" game before a winner can be determined -- it might help one's score to update one's forecast at the conclusion of that game).
How much did you think about this AlphaGo prediction? Would you be willing to wager money in a bet? If so, at what odds?
I am trying to figure out how much I can learn from your 8% prediction given your superforecaster status. You are clearly great at forecasting, but maybe in this particular case you didn't think as carefully about it as some of the other people estimating 60+% that AlphaGo will win (?), so I'm not sure how much I should believe your low estimate and lower mine from what it otherwise would be. Thanks.
Hi @WJK,
Thanks for your message.
Good on you for questioning the validity of the forecasts. I should tell you, though, that this wasn't my first forecast on this IFP. This was --> https://www.gjopen.com/comments/comments/118800. Because of the way the UI is for this site, it's not always easy to see the history of an individual's forecasts on a given IFP. Often times, I'll write (lowering/raising or no change) at the top of my forecast because I'm usually updating a previous forecast on a given IFP. Similarly, I often link to past forecasts on an IFP when updating. For this forecast, however, I didn't.
To answer your questions:
1) How much did I think about *this* forecast? Probably best to answer for (all) my forecasts on this IFP because I have two (so far). I'd say... hmm, I dunno.... maybe 45-75 minutes? I'm also including every time I read an article since I've made this forecasts (that are about this IFP). Google alerts is very helpful, in this regard. Also, just because I'm not actively thinking about an IFP doesn't mean some part of my brain isn't processing it, so number of minutes spent thinking about an IFP might not be the best way of gauging. Actually, to that end, someone who spends over 6 hours on an IFP might not be the best person to come up with an answer either (information overload and all that, but that's a conversation for another thread, I suppose).
2) Would I be willing to wager money? Hmm, no. I understand what you're trying to get at with that kind of a question, but GJ is different from gambling/odds. Is there even an oddsmaker out there right now who's taking bets on this? If there isn't, I'm sure there will be at some point. With gambling, you're trying to get your money in with the best odds. With forecasting, you're trying to predict the likelihood of an event. Sure that's very similar to gambling, but one quick example of how that's different: if someone gave me 100 to 1 odds that the Broncos would win by 14 points, I probably would have taken it (even though all of the "math" was pointing to a probable Panthers win).
It sounds like you're trying to ask me how confident I am of this forecast. Well, when I made it (https://www.gjopen.com/comments/comments/118800), I was relatively confident and I still am mostly confident in it. Of course with all the forecasts going in the other direction, it's natural for one to doubt one's self. The consensus isn't always right, though -- otherwise, there'd be no reason to discriminate between superforecasters and forecasters and dart throwing chimps. If you'd like a couple of examples...
1) Will North Korea launch a land based missile with the capacity to reach Alaska, Hawaii, or the continental United States before 1 January 2017? --> https://www.gjopen.com/questions/8-will-north-korea-launch-a-land-based-missile-with-the-capacity-to-reach-alaska-hawaii-or-the-continental-united-states-before-1-january-2017
Eye-balling the consensus on that IFP over the lifespan puts it at about 30%. It resolved yesterday at 100%. (I was one of those who was forecasting closer to 100%).
2) Which Republican presidential candidate will win the Iowa caucuses on 1 February? --> https://www.gjopen.com/questions/119-which-republican-presidential-candidate-will-win-the-iowa-caucuses-on-1-february
For most of the lifespan of this IFP, Trump was the candidate pegged to win (by plurality). He finished with about a 40% chance. In fact, Cruz's chances were the highest towards the beginning of this IFP and were at (his) lowest when the IFP closed. (I picked Cruz to win.)
3) Will Iran release Jason Rezaian before 31 October 2016? --> https://www.gjopen.com/questions/7-will-iran-release-jason-rezaian-before-31-october-2015
The consensus looks to be about 50/50 over the lifespan. (I predicted his release.)
~~
Those are just a couple of examples where the consensus doesn't always land on the right answer. Here's one last example:
Will Apple sell more than 75 million iPhones in its December quarter? --> https://www.gjopen.com/questions/47-will-apple-sell-more-than-75-million-iphones-in-its-december-quarter
The consensus on this one was over 85% for almost the first three months of the IFP's existence and only dropped to 75% with about 2 weeks before the IFP's closure. That was a BIG swing/miss by the consensus (myself included! as I was in for 100% over the lifespan of the IFP -- oops).
~~
I hope some of what I said was helpful for you.
Thank you, that was helpful.
"2) Would I be willing to wager money? Hmm, no. I understand what you're trying to get at with that kind of a question, but GJ is different from gambling/odds. Is there even an oddsmaker out there right now who's taking bets on this? If there isn't, I'm sure there will be at some point. With gambling, you're trying to get your money in with the best odds. With forecasting, you're trying to predict the likelihood of an event. Sure that's very similar to gambling, but one quick example of how that's different: if someone gave me 100 to 1 odds that the Broncos would win by 14 points, I probably would have taken it (even though all of the "math" was pointing to a probable Panthers win)."
I think I see what you're saying: whether it makes sense to bet depends on what odds you get.
But if you are estimating ~8%, then wouldn't betting on Lee Sedol with, say, 3:2 odds clearly be favorable (e.g. If Sedol wins, you get $2, if AlphaGo wins, you pay $3)? Your expected earnings from this bet would be positive if the probability of AlphaGo winning was <40%. And your estimate is ~8%, which seems definitely less than 40%.
So I'm confused: Would you be willing to wager (a small amount of money, such that if you lost, it wouldn't affect you, so no reason to be risk averse) at these odds?
(Note: I'm not actually proposing a bet with you. I'm just wondering why you wouldn't be willing to accept such a bet if you really believe the probability is ~8%, and very clearly less than 40%.)
From an odds standpoint, absolutely -- with the way I've set my forecast, 3:2 odds would be a no-brainer bet. To answer your question, would I be willing, yes, I would be willing (but I won't).
Another point on "8%" -- GJ is *technically* a game and so part of my forecast of 8% is playing the game.
Also, this comment (https://www.gjopen.com/comments/comments/128202) from @nicholas might be more grist for the (negative) on this IFP. It looks like the analysis is pretty solid, but that the consensus is dragging the forecast up.