AlphaGo vs Lee Sedol, who will win?

Kirby · **#61**

hyperpape wrote:

Kirby wrote:

It's a small mistake, and has nothing to do with the development of their AI - but at least it gives me the idea that he might not understand the magnitude of difference between a pro and top pro if he doesn't even know how to pronounce the name of the person they are challenging.

It's an embarrassing and unprofessional mistake, but it may also be a product of reading more about a subject than you talk about it. I don't know how to pronounce Lee's name either. I wouldn't pronounce it 'See-dol', but I don't know how 'dol' is supposed to sound.

Related: in graduate school, I was good friends with a Serbian grad student. After knowing him for a year, he and his girlfriend got married, and she joined him in the States. I constantly mispronounced her name ('Yelen-a' instead of 'Ye-lena'). I could hear the difference, I could say it right when being careful, but in conversation, it came out wrong. I finally had to stop myself, and apologize for getting it wrong so much.

tl;dr: mouth-words are hard.

Yes, I agree - it can be difficult. My opinion, though, is that while both sides are confident, the truth lies in the area where deep learning and Go intersect - which side is truly underestimating the other? If the deep mind developers are underestimating the skill difference between Lee Sedol and a weaker pro, their confidence is more likely to be unfounded. OTOH, many pros probably don't understand the specifics of how AlphaGo works all that well - so I know their confidence is at least partially unfounded.

Most people that have an idea of what's going on in the pro world would have a better pronunciation of Sedol's name, in my opinion, especially if you are going to challenge him.

So this kind of minor error, to me, indicates that the guy doesn't know as much about Lee Sedol as I had expected. Maybe I'm reading too much into it, but after following various articles in Korean and in English, I would bet on Lee Sedol for this match - especially if the 'deepmind' user on Tygem really is AlphaGo.

I'd be happy and sad at the same time to be wrong.

xed_over · **#62**

Kirby wrote:

So this kind of minor error, to me, indicates that the guy doesn't know as much about Lee Sedol as I had expected. Maybe I'm reading too much into it,

I think you're reading too much into this. I think one's native language and accent have a lot to do with mispronouncing.

When I was learning Japanese, I was constantly mispronouncing the words, until I switched to reading kana, instead of romaji.

Because with romaji, my brain already expected certain combinations of letters I saw to be pronounced a certain way, and it was a very hard habit to break.

But I didn't have any preconceived notions of pronunciation with the kana "symbols", and my pronunciation ended up much better.

pookpooi · **#63**

I'm not sure what do you mean about underestimation. I strongly believe that DeepMind know Lee Sedol true strength, regardless of their pronunciation on his name. I don't know how good AlphaGo is right now but I do know that it'll play at its full strength. Demis Hassabis also give a very neutral opinion in each interview, press conference and lecture, maybe even too humble in my opinion. DeepMind play this publicity stunt game very safe. But don't get me wrong, I'm not saying that what Lee Sedol said is too aggressive either, as I said in the comment before, I think DeepMind is already predicting this, secretly relying on him to give controversial interview to make an even bigger impact to the match.

Mef · **#64**

Kirby wrote:

pookpooi wrote:

I wouldn't be so surprised if DeepMind say they're already expecting those words from Lee Sedol.

I think DeepMind is confident from their side, too - to be sure.

I lost a little bit of respect for them, though, when I saw the nature video clip where one of the lead programmers on the team mentioned their upcoming challenge against "Lee See-dol".

It's a small mistake, and has nothing to do with the development of their AI - but at least it gives me the idea that he might not understand the magnitude of difference between a pro and top pro if he doesn't even know how to pronounce the name of the person they are challenging.

That being said, a lot of pros don't understand deep learning that well, either, so anything could happen next week, in my opinion :-)

What I think would be particularly telling would be for AlphaGo to publish a couple "commentaries" on well studied pro games. Basically, have the engine identify places where it disagrees with the move a professional made, then supply an alternative line. That would give a bit more insight into how the engine is approaching a game and how its evaluation compares to that of top pros.

Kirby · **#65**

xed_over wrote:

I think you're reading too much into this.

Maybe. Anyway, again in my opinion, at least one side is underestimating the other. It'll be interesting to see what happens next week.

Kirby · **#66**

pookpooi wrote:

I'm not sure what do you mean about underestimation. I strongly believe that DeepMind know Lee Sedol true strength, regardless of their pronunciation on his name. I don't know how good AlphaGo is right now but I do know that it'll play at its full strength. Demis Hassabis also give a very neutral opinion in each interview, press conference and lecture, maybe even too humble in my opinion. DeepMind play this publicity stunt game very safe. But don't get me wrong, I'm not saying that what Lee Sedol said is too aggressive either, as I said in the comment before, I think DeepMind is already predicting this, secretly relying on him to give controversial interview to make an even bigger impact to the match.

Yes, Demis Hassabis seemed somewhat humble - I think he said something to the effect of the odds being 50/50. However, incidentally, the developer that mispronounced Lee Sedol's name in this case (David Silver), was in my opinion less humble. He's the guy that said that they'd be pretty disappointed if they lost, and that they were expecting to win, and that their reputation was riding on the line.

David Silver wrote:

I haven’t put any money on AlphaGo winning, but I do think we have a lot of reputation riding on this bet. So let’s just say we’ll be very disappointed if we lose the match in March. But you never know, anything is possible. Humans inevitably have a lot of tricks up their sleeve that we’re not able to train against.

Given his confidence there, it seemed to me that they were pretty sure that they would win. Hence, my amusement when the same guy couldn't even pronounce the name of the opponent they were facing.

Don't get me wrong - maybe DeepMind has this in the bag. But I'm pretty sure both sides have confidence that they will win.

Kirby · **#67**

Mef wrote:

What I think would be particularly telling would be for AlphaGo to publish a couple "commentaries" on well studied pro games. Basically, have the engine identify places where it disagrees with the move a professional made, then supply an alternative line. That would give a bit more insight into how the engine is approaching a game and how its evaluation compares to that of top pros.

That would certainly be pretty cool. As a side note, I think there's potential for the technology behind AlphaGo to be used as an instructional aid for us amateurs, too - you can maybe have pro-level teaching without hiring a pro.

tj86430 · **#68**

Kirby wrote:

he might not understand the magnitude of difference between a pro and top pro if he doesn't even know how to pronounce the name of the person they are challenging.

I just don't get this argument. What has being able to pronounce a name correctly to do with appreciating the magnitude of the difference between a pro and a top pro?

I would never expect any foreigner to be able to correctly pronounce any Finnish names correctly (spelling is a different matter), but still I believe they may appreciate other things about Finns or Finland.

Kirby · **#69**

tj86430 wrote:

Kirby wrote:

he might not understand the magnitude of difference between a pro and top pro if he doesn't even know how to pronounce the name of the person they are challenging.

I just don't get this argument. What has being able to pronounce a name correctly to do with appreciating the magnitude of the difference between a pro and a top pro?

I would never expect any foreigner to be able to correctly pronounce any Finnish names correctly (spelling is a different matter), but still I believe they may appreciate other things about Finns or Finland.

First of all, it is not an argument - I am not trying to convince anyone here, and it is only my opinion. Maybe it was a mistake to share publicly, but I share just about everything publicly, so I guess that's why.

Maybe it doesn't make sense to other people, but I would bet that everyone on this forum can have a better idea of how to pronounce Lee Sedol's name. Can't pronounce Kim Jiseok? Sure. Can't pronounce Won Seongjin? Understandable. Hyperpape said he can't pronounce "Dol", but I'm sure he'd do a decent job.

It's not because pronunciation is related to understanding Go. It's because Lee Sedol is so ****ing famous that anyone that follows the pro Go scene knows his name.

To me, it'd be like the leader of a smaller country say that they will overtake the United States by defeating "President Abama".

Anyway, I am not trying to get anyone to share my viewpoint, and I probably shouldn't have brought it up. And it's not a big reason that I think deepmind underestimates top pros, either.

Mostly, I just think the situation is amusing.

Uberdude · **#70**

Under 2 days until the first game! I didn't set up the poll to close then, but I'll post a screenshot of the votes the day before the match. So vote now if you haven't already!

Elom · **#71**

Instead of predicting the result out of nothing, where do we think Alphago would rank on the world rating list. Top 100? top 50? Even a top 30 player would have their problems against Lee-Si^dol in a 5-Game match.

Kirby · **#72**

Elom wrote:

Lee-Si^dol

Haha. Funny... :roll:

Quote:

Instead of predicting the result out of nothing, where do we think Alphago would rank on the world rating list. Top 100? top 50? Even a top 30 player would have their problems against Lee-Si^dol in a 5-Game match.

Your question is a harder question than deciding whether AlphaGo would win against Lee Sedol. Whether or not he would win against Lee Sedol is a question with a 'yes' or 'no' answer, but you are asking for a more detailed idea of where AlphaGo ranks.

Since I don't know whether or not AlphaGo will beat Lee Sedol, it's even harder to say where it would rank among professional players.

But maybe we can estimate based on what professionals seem to think. It's been said that AlphaGo will be a "scary" opponent to Lee Sedol in another 2 years, based on looking at the game records with Fan Hui.

Looking at the ratings at goratings.org, Fan Hui has a rating of 3053, whereas Lee Sedol has a rating of 3533 - a difference of 480 points. It's been 5 months since AlphaGo played Fan Hui, and if we assume that AlphaGo can grow 480 points in 2 years, AlphaGo is 5/24th of the way there - about 0.2083. Assuming AlphaGo grows at a linear rate, AlphaGo would have improved by about 99.999 rating points since October.

That analysis includes a lot of assumptions :-)

But if it's true, AlphaGo would be around the top 200 players on Goratings.org with a rating around 3150~3200. Factor in Lee Sedol's overconfidence, and maybe we can say AlphaGo is in the top 100.

That being said, like I said earlier, I don't even know if Lee Sedol's confidence is founded, and AlphaGo may even already be stronger. Maybe AlphaGo is already number 1 in the world.

Who knows?

wineandgolover · **#73**

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.

pookpooi · **#74**

I just check bitbet and someone just bet 10 bitcoin ($4000) on AlphaGo! Meanwhile I'm broke...

Kirby · **#75**

wineandgolover wrote:

And if somebody wanted to give me good odds, I'd bet a lot on him winning 5-0.

Not quite sure what I'd bet at this point. In one interview, Lee Sedol said he thinks he has about 80% chance of winning any given game, and in this more recent one, he says he'd be disappointed to lose even one.

If it's really 80% chance of winning, though, that'd put the chances of winning all 5 at slightly over 30%.

CutFirstThinkLater · **#76**

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.

Kirby · **#77**

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.

pookpooi · **#78**

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!

tj86430 · **#79**

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?

Kirby · **#80**

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.

AlphaGo vs Lee Sedol, who will win?

Who is online