Life In 19x19

wineandgolover wrote:Has anybody found a place where we can bet real dollars, pounds, or euros on the outcome? I have as much interest in Bitcoin as I do Facebook, that is to say, none.

But if somebody wanted to bet even money, I'd bet a bit on Lee Sedol to win. And if somebody wanted to give me good odds, I'd bet a lot on him winning 5-0.

My gut feeling is that probability wise, the highest chance is Lee winning by 4-1 followed closely by AlphaGo winning 5-0. If I were to bet money, I'd actually bet on AlphaGo winning 5-0 because it's the "vastly underrated choice with a pretty good chance" (and I'm saying this as a long time fan of Lee.).

I think a close-contested match will be unlikely, since 1. AlphaGo seems to improve pretty quickly (compared to humans) so the chances that its current level is very close to Lee is relatively low (even if you take into account diminishing returns), and 2. in Go, the stronger player's chance to win is a lot bigger than e.g. chess. The main reason many top pros are similar in strength is the same as why the world's best sprinters are similar in speed - top humans tend to cluster around the biological limit set by nature. Since machines are not bound to this biological limit, the chances that the machine's strength and the human's strength is similar becomes lower (I've discussed this matter in post #44) - and just a minor difference in strength (i.e. something a bit bigger than the difference between top pros) will be enough to produce a lopsided result. So if Alphago wins, it'll probably be 5-0;, if Lee wins, 4-1 since humans are prone to mistakes.

There are 3 reasons I think the pros are underestimating AlphaGo. First, from what I've seen, their speculations are mostly based on what they've seen from the Fan Hui games and their past knowledge of Go programs and don't show a lot of understanding of how AlphaGo works. Second, the Deepmind team, who appears to be go players themselves and also have the best knowledge of AlphaGo's status, sounds pretty confident in their interviews (OC though, what else can they do? ). Third, and the biggest reason, is that AlphaGo playing Lee is likely to be quite stronger than the tygem "deepmind" account - because of better hardware and bluffing (Deepmind team seems well aware that this account is known in Korea, so they've probably tried to hide its full strength).

That said, I do think AlphaGo's improvement speed would have slowed down, so I'll go with the majority with Lee winning this time. However as I said in the beginning, If I were to bet money, I'll bet on the A.I. winning by a landslide.

I keep going back and forth - maybe I am swayed too much by the media. Currently, I am leaning toward saying AlphaGo will win.

CutFirstThinkLater, it surprises me that the result that you think has the highest probability differs from what you would bet money on.

I just heard that Baidu + AMD will arrange a go match between (yet to be announced) AI and Ke Jie with even more prize money. In Japan they also have DeepZenGo project (which will probably match with Iyama Yuta) so AI is now taking multi human territoies at once!

Kirby wrote:
CutFirstThinkLater, it surprises me that the result that you think has the highest probability differs from what you would bet money on.

Surely it depends on odds offered?

pookpooi wrote:I just heard that Baidu + AMD will arrange a go match between (yet to be announced) AI and Ke Jie with even more prize money. In Japan they also have DeepZenGo project (which will probably match with Iyama Yuta) so AI is now taking multi human territoies at once!

I saw the same news about Ke Jie, and apparently he said something to the effect of, "You should have your program beat a professional first, before thinking about challenging me."

I'm recalling this from memory, so the quote may be off.

tj86430 wrote:

Kirby wrote:
CutFirstThinkLater, it surprises me that the result that you think has the highest probability differs from what you would bet money on.

Surely it depends on odds offered?

Oh, do you mean betting sites are offering more if you bet on AlphaGo? I didn't assume anything about odds from his post.

Kirby wrote:

tj86430 wrote:

Kirby wrote:
CutFirstThinkLater, it surprises me that the result that you think has the highest probability differs from what you would bet money on.

Surely it depends on odds offered?

Oh, do you mean betting sites are offering more if you bet on AlphaGo? I didn't assume anything about odds from his post.

I don't know which odds (if any) are offered. I just assumed that if someone is betting on other than favourite (in his mind), it must be because of the odds.

It looks like that bitbet site is offering evens odds. Given most people think 4/1 or 5/0 to Lee is the likely outcome this is a very good value bet.

longshanks wrote:It looks like that bitbet site is offering evens odds. Given most people think 4/1 or 5/0 to Lee is the likely outcome this is a very good value bet.

My only issue with this bet is that there's a secondary bet involved, which is that you actually get your payout should you win. The reviews on bitbet are pretty shady, and it looks like currently they are suspending payouts because of a Chinese attack of some sort.

L19 users could always come to a personal arrangement for betting amongst themselves. I will offer a £10 bet that Lee will win the series at evens (so if you want to take me up on this bet and you think AlphaGo will win then the loser pays the winner £10 via Paypal or inter-bank transfer (I'm in the UK so would prefer not to give paypal their cut)). I reserve the right to accept or decline offers based on your L19 reputation.

Kirby wrote:CutFirstThinkLater, it surprises me that the result that you think has the highest probability differs from what you would bet money on.

Because I feel both sides have basically similar chances of winning with Lee only slightly ahead, and despite this, the overwhelming majority of the Go community (and the general public at least in Korea, where I live) seems to think Lee will win easily. So if a bet goes up, and the majority bets for Lee, I thought betting against Lee would promise a bigger expected value.

However, it seems at least in the west, I underestimated the weight "Google" carries in the layperson's mind. I visited BitBet.com and it looks like a lot of people there who know nothing about Go (which probably is the majority) are showing their faith in the IT giant. If that's the case, surely I should re-think my betting strategy.

GoRatings got five new games for Fan Hui, and they're all losses. Hope everyone has time to put the new ELO figures into their calculations before the first game starts.

hyperpape wrote:GoRatings got five new games for Fan Hui, and they're all losses. Hope everyone has time to put the new ELO figures into their calculations before the first game starts.

Evil!

Here are the final poll results.

AlphaGo wins the first game by resignation! 28 people are already wrong, more than the 17 who voted for any margin of AlphaGo victory!