Algorithms distort the world.
They can trap us in local maxima, and restrict the kind of random serendipity that makes our personalities liquid.
On Tuesday, I had lunch with a friend who invests in artificial intelligence startups. To my surprise, he doesn’t use any algorithms in his personal life. No Spotify Discover Weekly. No Netflix suggestions. No Amazon recommendations.
I’m not ready to make the same commitment, but there’s a deep insight behind his rule for consumption: if you want to find emerging and under-rated ideas, stop using algorithms.
These ideas are hidden and obscure. They’re the opposite of viral. Instead of relying on algorithms to find them, identify your favorite people and follow the people who influenced them.
One good recommendation from somebody who knows you will help you more than hundreds of searches using the world’s best general-purpose algorithm. The more nuanced the subject, the more you should avoid algorithmic recommendations.
Amazon systematically under-recommends books with a high-variance in reviews. It tends to recommend books everybody likes, instead of ones that some people love and other people hate. That’s why I trust footnotes and personal recommendations more than Amazon reviews. For example, the genesis for my most popular essay ever came from Martin Gurri, the author of Revolt of the Public. While flipping through the hyperlinks, I found Andrey Miroshnichenko who wrote a fantastic little book called Human as Media, which I would have never found otherwise.
Likewise, my friend Nick finally ate at J.G. Melon last weekend. It’s my very favorite restaurant in New York. I’m convinced eating at a restaurant is more enjoyable when a friend recommends the place. I have no data to confirm this, but I suspect there’s a psychological reason for it. Maybe there’s more room for surprise or something like that.
My dad gave me lots of good advice, but among the best was his constant reminder to look at what everybody else is doing and do the opposite. His words have aged well, especially in the Internet age. By reinforcing our existing preferences, algorithms can narrow the variance in our consumption patterns.
Consider two areas: housing and online metrics.
Housing: Don’t tell a realtor what features you want in your next home. Tell them what other people value, but you don’t care about. For example, when I was looking for an apartment, I didn’t value a garage or an outdoor space, and I was willing to live with roommates. Many apartments in my price range came with features I didn’t need, and since I didn’t value them, I found a brand new apartment in Williamsburg that feels underpriced to me. Beyond that, people always talk about easy-to-measure specs such as the number of bedrooms and bathrooms. Look for things that are hard to measure instead, such as natural light, storage efficiency, and a digitally connected door for Amazon packages. All things being equal, the market will undervalue these features. And finally, since most people in cities want their apartment to be close to a subway, look for a place that’s close to a convenient bus route. Doing so could save you hundreds of dollars per month and shorten your commute to work.
Online Metrics: Internet metrics over-value quantity and under-value quality. It’s a measurement problem. For example, platforms can easily calculate how many people read my articles, but they can’t evaluate the quality of those people. This insight didn’t feel real until I attended Capital Camp, a conference hosted by two of the best people I know: Brent Beshore and Patrick O’Shaughnessy. I’d never seen anything like it. Nearly every person there was exceptional. There was more intellectual horsepower at the conference than any room I’ve ever been in. Since the conference, I’ve dropped the strategy of building the biggest possible online audience and tried to attract the most interesting and intelligent people I can instead. Sure, it’s hard to measure. But that’s exactly why this strategy repels competition. With that said, if you want to improve something, you have to measure it. In the words of Peter Drucker: “What gets measured, gets managed.” Inspired by Drucker’s wisdom, I measure the success of my essays by the number of interesting emails I receive after publishing them. If the emails aren’t interesting, I’m not attracting the right kinds of people. And if there aren’t enough emails, I’m not reaching enough people. Dropping the obsession with page views and building your own metric is a helpful way to differentiate yourself, find mispriced opportunities, and make algorithmic blind spots work in your favor.
Look for patterns in algorithmic recommendations, and avoid the standard recommendations. If you need to make up an algorithm, make up your own.
Doing so will save you money, keep you away from crowds, and improve the quality of what you consume.
This is an excerpt from a weekly newsletter I write called Monday Musings. You can subscribe here.