This article is over-analyzing the over-analyzing of baseball.
(I thought, reading this, that it was going to lead to the limits of machine learning, and the baseball stuff was just blogger blithering. No, that's the whole article.)
That's a highly unfair characterization of a thoughtful, carefully written piece on how too much of a good(?) thing—technology, statistics—can ruin the perceived value of human activities and products.
Well, it's talking about the pointlessness of optimising arbitrary measures of success so you could say that it is, indeed, about machine learning - the field of research and its core practice.
It is not pointless. These optimizations win games, which brings in more money. Professional baseball is a business like any other, so it should come as no surprise that it will be optimized like any other business.
> We can now measure and analyze dimensions of personal and social life that we wouldn’t even have thought to measure a decade or two ago. It was always theoretically possible for us to count our steps, but altogether impractical. Now we can have it done for us passively.
Massive fan of Convivial Society, although I think some pieces necessarily won't give the full background of Illich's belief that technological progress is not necessarily a good thing, and should be examined.
I feel that archery is much the same way to me as baseball is discussed here - I enjoy shooting a recurve bow without sights or counterweights or additions. I could improve my shooting drastically in all common metrics if I wanted to swap to a modern bow, I'd be more accurate, using more force, at a greater distance.
But as none of those metrics measure how satisfied I feel after shooting, it doesn't matter to me and I won't swap.
Even in a context were sheer efficiency is key someone, for example a warrior, may prefer a 'classic' (rugged, always-ready (nothing to tune), easy to fix, compact, light...) bow to a 'better' contemporary (complicated) one.
The problem with baseball is not that teams are solving it, but that the game and it's rules and structures are not optimizing to adapt and make it more difficult. The most surprising thing to me is that the game wasn't "solved" a long time ago. We shouldn't lament when teams play a game very efficiently, we should design better games.
I think chess would be a more straightforward example to illustrate the article's point. If you look at old games, they used to be a lot more fun, with more variety and risky, moves; see, e.g., Anderssen's "heroic" style. As the game became better understood, these kind of off-the-wall strategies became less common since they rarely paid off against well-prepared adversaries. And of course, the trend came to a logical -- and more poignant -- conclusion than in baseball, with Deep Blue and other chess AI taking over. I believe a similar story could be told for Go.
I would counter that these off-the-wall approaches have become more common with more advanced chess AIs, i.e. AlphaGo Zero and Leela Chess Zero. They frequently play gambits and other aggressive lines that previous chess engines didn't. I think these limits may just be the growing pains of applying rationality to a game, and in the long term statistics and optimization can lead us to better games in general.
10 comments
[ 4.3 ms ] story [ 35.2 ms ] thread(I thought, reading this, that it was going to lead to the limits of machine learning, and the baseball stuff was just blogger blithering. No, that's the whole article.)
Pedometers are much older than "a decade or two": https://en.wikipedia.org/wiki/Pedometer#History
https://walkertracker.com/the-evolution-of-the-pedometer/
I feel that archery is much the same way to me as baseball is discussed here - I enjoy shooting a recurve bow without sights or counterweights or additions. I could improve my shooting drastically in all common metrics if I wanted to swap to a modern bow, I'd be more accurate, using more force, at a greater distance.
But as none of those metrics measure how satisfied I feel after shooting, it doesn't matter to me and I won't swap.