The Explore/Exploit Edition

Noah here. Way back in the first week of WITI (2019), I wrote about the explore/exploit tradeoff. The tradeoff is about understanding the give and take between exploration—going wide, trying new things—and exploitation, or going deep into one thing. Explore/exploit is a big area of computer science, particularly artificial intelligence, and machine learning, where you need to program a machine to survey the landscape of solutions and then choose the right one. The question that needs to be answered is how long to explore before you dive deep into exploit?

This extends far past just CS, though. As Brian Christian points out in his book Algorithms to Live By: The Computer Science of Human Decisions, there are actually tons of interesting stuff that lives in this tension. Do you go to the restaurant you like or try the new one that just opened? Should you load up an old favorite on Spotify or see what they’ve chosen for you this week? The answer, as we have all figured out and computer science has proven, is it depends. To figure out the best approach you’ve also got to know the time limit. In simplified terms, if you have lots of time left, exploration makes sense, if you’re approaching a deadline, exploitation is optimal.

Subscribe now

Why is this interesting?

In the last few weeks, two very interesting articles about explore/exploit have come out. The first, from The Atlantic’s Derek Thompson, looked at where “career hot streaks” come from—essentially trying to understand what can lead some people to enter particularly productive work periods.

In [Northwestern University economist Dashun] Wang’s most recent analysis, he found that artists and scientists tend to experiment with diverse styles or topics before their hot streak begins. This period of exploration is followed by a period of creatively productive focus. “Our data shows that people ought to explore a bunch of things at work, deliberate about the best fit for their skills, and then exploit what they’ve learned,” Wang said. This precise sequence—exploration, followed by exploitation—was the single best predictor of the onset of a hot streak.

Interestingly, this goes well beyond careers to the fundamentals of learning. In a recent study by Alison Gopnik and Emily Liquin, they looked at the way children and adults balance their exploration versus exploitation. What they found is that the children did more exploring than the adults: a fact that doesn’t mean their approach was better, but rather might point at some of what make kids such unbelievable learners. From a WSJ piece by Gopnik:

We used what’s called a “learning trap.” When we grown-ups try something new, from oysters to opera, and get a bad result, we usually won’t try it again. That might seem like the most basic kind of intelligence—even rats stay away from a path that leads to a shock. But it has an important downside. If we quickly conclude that all oysters and operas are indigestible, and reject them ever after, we will never learn that the world is more complicated than that. A stale clam or lame Aida may keep us from ever discovering the delights of a sparkling Belon oyster or a scintillating Magic Flute.

As Christian relays in his book, there’s even an ideal split on explore/exploit: 37 percent. If you know the exact time dimensions you’re working within (say you have one month to find a new apartment, for instance), you’d be best off spending 37 percent of it exploring before you go deep with the best option. But, as the childhood learning study points out, it probably all depends on how you’re defining optimal. Here’s Gopnik again concluding her WSJ piece: “We grown-ups are often so anxious to exploit that we don’t explore, so afraid of losing stars that we miss the chance to learn something new. Children, in contrast, are natural explorers, willing to sacrifice stars for the sake of information. You need both types of thinking to thrive, but we grown-ups might learn something from those insatiably curious kids.” (NRB)

Quick Links:

• A two thousand dollar Japanese longbow (CJN)

• Streaming geospatial to the Metaverse (CJN)

• Graydon recommends this back catalogue (CJN)

• [Sponsored: Atlas Bars are GMO-free and keto-friendly with 15 grams of protein and just a gram of sugar. We’ve tried Atlas and they are very good. Use this link for WITI readers to get 20 percent off. Get started with the sample pack.]

--

WITI x McKinsey:

An ongoing partnership where we highlight interesting McKinsey research, writing, and data.

The art of apprenticeship. Postpandemic skill gaps need filling, and formal learning alone won’t do the trick. Scaling the lost art of one-on-one learning can make the difference. Here's how.

--

Thanks for reading,

Noah (NRB) & Colin (CJN)

—

Why is this interesting? is a daily email from Noah Brier & Colin Nagy (and friends!) about interesting things. If you’ve enjoyed this edition, please consider forwarding it to a friend. If you’re reading it for the first time, consider subscribing (it’s free!).