Everyone knows I love to play games. I think the process of solving a game – with all of the reasoning, analytics, and creativity that can go into it – is exhilarating and a microcosm for how we live life, which is more or less one big game. The way I approach games is also related to why I don’t play as many games as I used to…because no one wants to play with me.
One really cool game a lot of people don’t know about is called ‘Go.’ It originated in China thousands of years ago and is considered one of the most strategic games in the world – considerably more difficult to solve than chess. It requires a ton of specifically human elements of thought – like intuition and creative thinking – such that it has always been difficult for artificial intelligence to solve.
Until recently, anyway, when a computer – AlphaGo – beat Korean professional Go player Lee Sedol in four of five games. The match has been compared to the historic Deep Blue vs. Garry Kasparov chess match in 1997.
Up until last year, the best Go programs weren’t even close to competing with the top humans. That changed with improvements in neural networking.
AlphaGo is most significantly different from previous AI efforts in that it applies neural networks, in which evaluation heuristics are not hard-coded by human beings, but instead to a large extent learned by the program itself, through tens of millions of past go matches as well as its own matches with itself. Not even AlphaGo’s developer team are able to point out how AlphaGo evaluates the game position and picks its next move.
This is a massive step for AI, as AlphaGo in many ways relies on sound judgement and reasoning.
Odd Moves
What’s interesting about both the Kasparov and Sedol matches is that Deep Blue and AlphaGo both made extremely odd moves that many professionals have said a human player wouldn’t make.
From Game 1 of the Go match:
One of the creators of AlphaGo, Demis Hassabis, said that the system was confident in victory from the midway point of the game, even though the professional commentators could not tell which player was ahead.
Michael Redmond noted that AlphaGo’s 19th stone (move 37) was “creative” and “unique.” Lee Sedol took an unusually long time to respond to the move. An Younggil (8p) called AlphaGo’s move 37 “a rare and intriguing shoulder hit.” He stated that control passed between the players several times before the endgame, and especially praised AlphaGo’s moves 151, 157, and 159, calling them “brilliant.”
There were multiple other moves throughout the games that Sedol repeated a human would not make, including one sequence in which AlphaGo completely abandoned one part of the board it had been working on – a highly uncharacteristic move by human standards.
Application to DFS
I find all this so interesting because I think it has direct implications to modeling in daily fantasy sports, in a few ways.
First, the means by which we build our models – the stats we choose to analyze and how we weight them – should be approached from a completely unbiased point of view.
Our new Director of PGA – Colin Davy – stirred up some Twitter controversy (nothing cooler than nerds arguing on Twitter about advanced golf analytics) by arguing that Strokes Gained is an overrated statistic for daily fantasy golf.
He laid out his argument in that article (and expanded upon it in last night’s podcast) – so I won’t dive into it too much – but Colin approached PGA modeling from a fresh perspective, with as little bias as possible, and decided there were perhaps better ways to build a model than using Strokes Gained (the most convincing argument of which is that we can reproduce the predictive ability of the metric using variations of more basic stats, all of which are available on all tours, which we have).
It reminded me of something I heard from a good friend of mine – and one of the best tournament DFS players in the world – Jay Raynor (BeepImAJeep), which is that many of the world’s top Scrabble players don’t even speak English, likely because they don’t suffer from the same biases English-speaking players do when it comes to favoring certain words or word types.
By the way, Jay did this last night:
Just accidentally shaved off 1/3 of my eyebrow. See you tomorrow SD @DraftKingspic.twitter.com/8Q1MvZAh4c
— Jay Raynor (@BeepImaJeep) March 16, 2016
I’m not sure how that happens, but I couldn’t stop laughing for like 10 minutes last night.
So we can learn something from AlphaGo and AI in general when it comes to the foundation of building a model, but I also think we can do it when interpreting results.
Currently, there’s no model in DFS that can outproduce the top human players. Many of those guys use models, but model + human is superior to model alone.
However, I think the strange moves from Deep Blue and AlphaGo can serve as a reminder to us to not immediately dismiss the “weird” players who might move to the top of the models you create. Sometimes that could be the result of a model’s blind spots, but sometimes it might be that our own blind spots or biases are the reason for thinking such a high rating is “weird” in the first place.
Don’t be so quick to dismiss “oddities” in your model; those might just end up being the most profitable plays you can find. I can recall seeing many “what the hell?” lineups from BeepImAJeep – a player who admits to not caring much about sports – that ended up being incredibly profitable. Because Jay approaches the game in such a unique and outside-the-box sort of way, he doesn’t suffer from the same biases that a lot of us do – the “oh man I have to play this guy” or “there’s no way I can stick that guy into my lineup” that seems to be a daily DFS occurrence.
Before leaving, I wanted to leave you with one more quote explaining AlphaGo’s methodology that I think you’ll agree mirrors how we should be approaching DFS:
AlphaGo showed anomalies and moves from a broader perspective which professional Go players described as looking like mistakes at the first sight but an intentional strategy in hindsight. As one of the creators of the system explained, AlphaGo does not attempt to maximize its points or its margin of victory, but tries to maximize its probability of winning. If AlphaGo must choose between a scenario where it will win by 20 points with 80 percent probability and another where it will win by 1 and a half points with 99 percent probability, it will choose the latter, even if it must give up points to achieve it. In particular, move 167 by AlphaGo seemed to give Lee a fighting chance and was declared to look like an obvious mistake by commentators. An Younggil stated “So when AlphaGo plays a slack looking move, we may regard it as a mistake, but perhaps it should more accurately be viewed as a declaration of victory?”
