Doomscrolling and the Variable-Reward Loop

In 1957, the Harvard behaviourist B.F. Skinner and his collaborator Charles Ferster published a 750-page book called "Schedules of Reinforcement," which catalogued every way they'd found to make a pigeon press a key. They tested fixed schedules — a pellet for every fifth peck, or a pellet every thirty seconds — and they tested variable ones, where the requirement changed unpredictably from one reward to the next. The variable-ratio schedule, in which a pigeon was rewarded after an unpredictable number of pecks averaging some target value, produced response rates higher than any other arrangement they had tested. The pigeons pecked steadily, without pauses, for hours. When the rewards stopped entirely, birds trained on variable-ratio schedules kept pecking long after birds on fixed schedules had given up. Resistance to extinction was the technical term. The behaviour outlasted the contingency that had built it.

The same schedule had already been industrialised, accidentally, by another American invention. The slot machine, refined in the 1890s and 1900s by Charles Fey in San Francisco, paid out on a schedule mathematically equivalent to Skinner's variable ratio: an unpredictable number of pulls between rewards, a small average reward size, and an occasional large jackpot. The combination produces what gambling researchers later called the "zone" — the trance state observed in problem gamblers in front of electronic machines, with elevated heart rate, narrowed attention, and a striking insensitivity to time, hunger, and social cues. The MIT anthropologist Natasha Dow Schüll documented this in her 2012 book "Addiction by Design," based on years of fieldwork in Las Vegas. The machines, she argued, were not designed to win; they were designed to keep players in the zone for as long as possible. Time on device was the metric. Loss of money was secondary.

The neuroscience that makes the schedule so potent was worked out in the 1990s. Wolfram Schultz, a Cambridge neurophysiologist, recorded from dopamine neurons in the midbrain of monkeys and discovered that the cells did not fire when a predictable reward arrived. They fired when something better than expected happened. Roy Wise, at the U.S. National Institute on Drug Abuse, established the same prediction-error signature in rats. Dopamine, contrary to the popular shorthand, isn't the brain's pleasure chemical — it is the brain's surprise chemical. A predictable reward generates almost no dopamine response. An unpredictable one floods the system. Variable-ratio schedules are, in effect, a machine for generating prediction errors: every pull, every peck, every refresh might be the one. The expected value of the action is uncertain by design.

The architecture of every modern feed is a direct application. The infinite scroll, the specific interaction pattern in which content loads continuously as the user pulls down, was prototyped by the designer Aza Raskin in 2006 — initially for a startup search interface, before it diffused to Twitter, Facebook, Instagram, and TikTok. Raskin himself has spent the years since publicly apologising for it, telling the BBC in 2018 that the design "behaves like a slot machine in your pocket." The pull-to-refresh gesture, introduced in the iPhone Tweetie app in 2009 by the developer Loren Brichter, is even more direct: the gesture is identical to pulling a slot-machine arm, and the variable reward — a new tweet, no new tweet, several new tweets — arrives on a schedule statistically indistinguishable from a Las Vegas multi-line slot.

The numbers on actual usage are higher than most people guess. RescueTime's longitudinal tracking data, drawn from millions of installations, has put average daily smartphone interactions at between 58 and 96 sessions per day for adults. Apple's own Screen Time data has put the median daily pickup count for U.S. iPhone users above 80. The session lengths are short — most are under a minute — but the count, multiplied by the latency cost of switching attention, is what consumes the day. A 2014 study by Sophia Mark and colleagues at UC Irvine measured a recovery time of about 23 minutes after each interruption to return to deep focus on the original task. At eighty pickups, the math collapses any expectation of sustained attention regardless of intent.

The behaviour persists not because users misunderstand it but because the schedule is doing exactly what it was designed to do. Tristan Harris, a former Google design ethicist who founded the Center for Humane Technology in 2018, has described the asymmetry plainly: a single user has roughly the cognitive bandwidth of one human; the system on the other side of the screen has billions of dollars of optimisation budget aimed at maximising engagement, and a feedback loop measured in milliseconds. Skinner himself, in his last book in 1990, argued that the same operant principles he had documented in pigeons would, in the hands of any sufficiently large operator with sufficient data, become the dominant force shaping mass human behaviour. The variable-ratio schedule he and Ferster had described in 1957 — the most powerful contingency they could find for sustaining a behaviour without a reliable reward — is now embedded in roughly four billion pockets, refreshed on average every fifteen minutes of waking life, twenty-four hours a day, in every time zone, indefinitely.