What this covers
Christian Rudder, co-founder of OKCupid, talks with host Patrick about what a dating site's records show and what building one taught him. The first stretch is the data. Men of every age, from 30 to 45, rate and contact the youngest women available, whatever age range they list on their profiles. That is a case of Stated vs Revealed Preference, which Rudder attributes partly to Social Desirability Bias. On race, he reports that black men, Asian men, and black women receive about 75 percent of the responses others get, and that this bias comes from every other group. Latino users, in this data, hold the same biases as white users and face none themselves. Rudder notes that person-to-person interaction data on race is otherwise almost nonexistent.
The middle of the conversation is about running the company. OKCupid measured success by the number of deep, separate conversations rather than time on site or raw messages, an instance of Working Backward from User Goals. Rudder also makes a contested claim that polarizing appeal draws more attention than uniform middling ratings, and Patrick compares it to the one-star and five-star reviews on best-selling books. They cover founder luck, the four co-founders' separate roles (including Max Krohn on the back end), the hard call on a stable but unproven startup, and SparkNotes' growth funnel through The Spark.
The last stretch turns to larger questions. Rudder distinguishes aggregated corporate ad targeting from NSA surveillance of named individuals. He forecasts human-level machine intelligence within roughly 20 years, argues for a Manhattan Project-style open effort on Superintelligence, and proposes taxing AI use to compensate displaced workers. Patrick adds Power law concentration in profit margins.
Rudder argues that OKCupid's behavioral data exposes gaps between what people say they want and what they do, that choosing the right metric and questions matters more than the math, and that corporate and government data collection should be judged separately, with superintelligence and automation posing the largest long-term risks.
- Stated preferences (age, race) diverge sharply from observed dating behavior, and person-to-person data reveals biases that surveys miss
- OKCupid rejected engagement and raw-message metrics in favor of deep conversations because the wrong metric would have failed the business
- Corporate ad targeting is aggregated while NSA surveillance targets individuals, so the privacy harms differ in kind
Returns and attention concentrate on a few distinctive, polarizing winners.
- Profit margins of the highest-margin companies have steadily risen, reflecting a magnified power-law pattern in which more and more returns accrue to a small number of winners, and easier company formation makes this hard to guard against.
“what's the margin of that group? That has been steadily rising. Meaning, more and more is accruing to the winners”
- A new and different offering that some people hate and others love tends to attract more interest than an iterative generic version of what everyone else does, which mirrors why best-selling books have many one-star and five-star reviews.
“if you're just offering like an iterative generic version of what everyone else is doing, probably not going to incite a lot of interest.”
Superintelligent AI could take over creative work, leaving open what humans will do.
- If creativity is pattern recognition, then a superintelligent AI could also take over the creative component of work, which leaves the question of what humans will do.
“if creativity in some way is a reflection of our level of intelligence and I think of creativity is just recognizing patterns, if an AI is smarter than us, it could potentially take that too”