Mauboussin: Algorithmic models fail when applied to non-stationary, dynamic systems
Michael Mauboussin · #28 Michael Mauboussin: A Decision Making Jedi · Jan 23, 2018 · at 32:17
Investor and author Michael Mauboussin discusses the limits and risks of algorithmic decision-making and artificial intelligence with Shane Parrish.
0:00 / 0:26exact quote · 26.5s
720p mp4 · rendered on demand · StarZero watermark
“You have to be very, very careful about systems that are changing, you know, the fancy term for its non-stationarity. So the statistical properties and underlying systems change over time. And that's another concern I might have, and we see it a little bit of this in finance. It's this concept of overfitting, right? So you have these data, you can match history beautifully but you fail to anticipate that the world itself is dynamic, and so the rules that worked in the past may not work in the future.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Michael Mauboussin
Mauboussin: NBA outcomes rely on skill; hockey and baseball involve luck
“The NBA is the sport that's furthest from randomness, furthest from pure luck configuration, and sports like actually ice hockey and baseball are actually quite much closer to being random.”
Mauboussin: Senior executive decisions depend far more on luck than entry-level work
“Usually when you're a lower level worker, your job actually tends to have a higher skill component, right? So you're doing something, you're on an accounts receivable clerk and you're collecting accounts receivables, right? We can measure what we're doing. It'…”
Mauboussin: Network effects create increasing returns that defy economic mean reversion
“So we're taught in economics, microeconomics, and by the way, it's still largely true that, Returns on incremental capital tend to migrate toward the cost of capital, and typically the reason for that is competition, right? If you're generating, Shane, massive…”
Mauboussin: Overcome algorithm aversion by allowing minor human adjustments
“And it turns out, they said, there's a way to overcome this which I think makes enormous amounts of sense. And they said, the way you overcome this is you say to people, you're going to, your first answer is going to be what the algorithm tells you, right? Jus…”
Mauboussin: Only pay for franchise value in stable, low-churn industries
“You don't want to pay for you know, huge amount of competitive advantage.
In an industry where there's a ton of entry and exit, where market shares are very volatile, and by contrast, you might be much more willing to pay for franchise value for an industry w…”