Casey Has A Heart

26 Jul, 2026

One of the more remarkable predictions about AI was told more than sixty years ago on The Twilight Zone.

The Mighty Casey

On June 17, 1960, CBS aired "The Mighty Casey," the thirty-fifth episode of The Twilight Zone's first season, written by Rod Serling. The last-place Hoboken Zephyrs receive a gift from an inventor: a pitcher named Casey who throws with flawless, inhuman precision. The Zephyrs start winning. Then Casey is hit by a pitch, examined at a hospital — and found to have no heartbeat. He is a robot, and the league rules him ineligible. A player, says the rulebook, must be human.

The inventor's fix is elegant and catastrophic: an artificial heart, so Casey can qualify. It works. And the heart brings empathy with it. Casey can no longer bring himself to strike anyone out. He stands on the mound feeling sorry for the batters. Compassion replaces perfection. He quits baseball and becomes a social worker, and Serling's closing narration hints, "unsubstantiated, of course," that the manager later took the blueprints west and built a pitching staff that pitched like nothing human.

The Twilight Zone (1959–1964) television series logo, derived from the original title screen. Rod Serling's landmark anthology introduced "The Mighty Casey" (1960), an episode imagining an android baseball pitcher decades before modern AI and sports robotics. Copyright status: Public domain (PD-textlogo) via Wikimedia Commons because the logo consists only of simple geometric shapes and text and does not meet the threshold of originality for copyright protection. Original logo design attributed to Cayuga Productions, Inc. and CBS Productions. The series name and logo may remain protected as trademarks in some jurisdictions.

There is a production story behind the episode almost too fitting to believe: Paul Douglas, originally cast as the manager, died of a heart attack within days of filming wrapping, and Serling reportedly spent some $27,000 — much of it his own money — to reshoot with Jack Warden. An episode about a heart, remade because of one. Even the name carries freight, reaching back to Ernest Thayer's "Casey at the Bat" of 1888, with a lovely inversion between the two: Thayer's Casey is a slugger who fails from overconfidence; Serling's is a pitcher who fails from compassion.

Viewed from 2026, the episode asks strikingly modern questions. Can a machine outperform a human? Should it? What matters beyond raw performance? But looking back across sixty-six years, the striking thing is this:

Science fiction predicted robotic athletes. Reality produced robotic coaches.

Serling put the robot in the right place

The first clue that Serling was onto something is Casey's position. He didn't imagine a robot shortstop or a robot center fielder. He imagined a robot pitcher — and pitching is, structurally, the one job in baseball a machine could plausibly do.

The pitcher's plate sits exactly 60 feet 6 inches from home plate, on a mound whose slope is mandated by the rulebook. Every pitch begins from the same place, aimed at the same seventeen-inch target. The task is discrete — play starts when the pitcher chooses. It is closed-loop — fixed distance, fixed target, no contact with defenders. It is repeatable — the same mechanics every time, with a limited decision tree of pitch type and location. Compare the shortstop improvising a throw off an unpredictable hop, the center fielder judging a fly at a sprint, the quarterback doing all of it while being chased.

Robotics has a name for this divide. Hans Moravec observed in 1988 that it is easy to make computers perform at adult level on intelligence tests and nearly impossible to give them the perception and mobility of a one-year-old. Steven Pinker compressed Moravec's paradox further: the hard problems are easy and the easy problems are hard. Sensorimotor skill carries a billion years of evolutionary refinement; abstract reasoning is a recent trick.

History confirms the pattern with almost comic precision. Unimate, the first industrial robot, went to work on a General Motors line in 1961 — one year after the episode aired — doing what a pitcher does in the abstract: repeatable, precise delivery from a fixed base. Machines have literally been pitching since 1897, when Princeton mathematician Charles Hinton built a gunpowder-powered pitching gun that threw curveballs. No one, in 130 years since, has built a fielding machine. And in 2009, a University of Tokyo lab built a robot pitcher that threw 90 percent strikes — at 25 mph, with a foam ball — plus a robot batter that hit nearly every strike. IEEE Spectrum later noted the lab had demonstrated throwing, batting, tracking, running, and catching as separate systems but never integrated them into one baseball-playing robot. Each narrow skill is tractable. The whole game is not. That is Moravec's paradox wearing a uniform.

The robotic pitcher exists. It throws batting practice.

Here is where the story turns, because the robot Serling imagined has been built. It just isn't on anyone's roster.

The Trajekt Arc, built by a Toronto-area startup, replicates specific major-league pitchers — release point, arm slot, spin, velocity, movement — while displaying synced video of that pitcher's delivery at the release point. A hitter stepping in sees tonight's opposing starter wind up and sees the ball emerge from the right spot in the video, carrying that pitcher's exact slider. ESPN reported leases around $15,000 a month, and described a pinch-hitter studying a reliever on the Trajekt mid-game before singling off the real thing. By April 2025, 26 of 30 MLB clubs had lease agreements. Pitchers have complained, understandably, that this is not entirely fair.

The Trajekt is the most vivid machine in a much larger stack. TrackMan's Doppler radar measures spin and break from big-league parks to college showcases. Rapsodo's portable units carry the same analytics down to youth fields — and feed the Trajekt's replications. HitTrax turns indoor cages into simulators with virtual ballparks. WIN Reality puts recreated major-league deliveries into VR headsets. Blast Motion has been the official bat sensor of MLB since 2016. Above it all sits Hawk-Eye, Statcast's optical backbone since 2020: twelve cameras per park tracking not just the ball but eighteen skeletal points per player, thirty times a second — every game now generates roughly 25 million data points.

This season those same cameras started calling pitches: the Automated Ball-Strike challenge system went live across MLB in 2026 — two challenges per team, invoked with a tap of the helmet. Note the design: players preferred the challenge format over full automation. The human call stays primary; the machine is the appeal. The first robot to make the big leagues is an umpire's backstop.

Step back and look at what this industry is. None of these systems plays baseball. None fields a bunt, covers first base, or joins a clubhouse. Every dollar of it is spent making human players better. The market has voted, and it voted for augmentation.

The revolution between: Moneyball

Between Serling's question and today's answer sits the analytics revolution — and it wasn't really about Billy Beane either.

The pedigree is long. Henry Chadwick's box scores gave baseball its batting average in the nineteenth century. Branch Rickey hired Allan Roth in 1947 as the first full-time team statistician; Roth charted virtually every Dodgers pitch for seventeen years, tracking on-base percentage decades before anyone said "analytics." Bill James self-published the first Baseball Abstract in 1977 and coined "sabermetrics" — "the search for objective knowledge about baseball" — and was mocked or ignored by the establishment for two decades.

The core insight is beautifully simple. A team gets twenty-seven outs; outs are the scarce resource. Batting average ignores walks; on-base percentage counts every way of not making an out, and so predicts wins better. Yet the player market priced batting average. That gap was free money for whoever noticed. Michael Lewis's Moneyball (2003) told the story of the noticing: the 2002 Oakland A's, having lost their stars to free agency, fielded a roughly $40 million opening-day payroll against the Yankees' $126 million and won 103 games, including a 20-game streak that ended on a walk-off homer by Scott Hatteberg — the on-base-skills signing personified. Economists later confirmed the thesis: OBP genuinely was mispriced, and the mispricing vanished once the book named it. Boston hired 28-year-old Theo Epstein and Bill James in November 2002 and broke an 86-year curse two seasons later. By the late 2010s every club had an R&D department.

Notice what the algorithms did not do: they did not swing bats or throw pitches. They improved decisions. Computers changed roster construction before AI became fashionable — the first algorithmic era, run on spreadsheets.

Three questions, sixty years apart

Seen together, the milestones form a progression — scouts, then statistics, then machine learning, then the first AI systems — and each asked a different question.

The Mighty Casey asked: can a robot replace the player? Moneyball asked: can data help us choose better players? Modern AI asks something different again: can intelligent systems help human players, and human teams, get better?

Each stage kept the human on the field and moved the machine closer to the decisions around the human. Statcast turned every pitch into sixty data points. Training labs like Driveline brought pitch design and biomechanics into player development. A 2025 peer-reviewed study trained transformer models on five and a half million tracked pitches to predict Tommy John surgery risk up to a hundred days out. And the clearest evidence the future is arriving: MLB reportedly had to ban generative AI from dugout iPads this June, after one club — per former reliever Adam Ottavino's account — built an AI system that helped pick pitches in-game. Leagues do not ban technology nobody is using.

Meanwhile, the robots that were supposed to take the field are in warehouses. Every genuine humanoid deployment on Earth — BMW's plants, logistics floors, Hyundai's factories — is repetitive manipulation in controlled space: Unimate's descendants, not Casey's. At RoboCup 2026, the first full eleven-on-eleven humanoid soccer match on real hardware ended 4–0 between small, wobbly players; the field's own goal of beating human World Cup champions is set for 2050, and it looks ambitious. Beijing's humanoid half-marathon winner needed two hours forty minutes for a distance the winning human ran in 1:02. Body engineering, it turns out, is harder than reasoning. Moravec had the numbers.

The real surprise

Serling imagined a machine on the mound. What 2026 actually has is machines around the mound — replicating pitchers for batting practice, judging the strike zone on appeal, tracking every skeleton in the ballpark, predicting elbow injuries months out — and zero machines on any roster, anywhere.

The explanation is not only technical. It is about demand. Nobody really wants to watch robots play baseball; the robot boxing matches and robot marathons draw curiosity, not season tickets. Everybody wants humans who play better baseball. The paying customer for a perfect robotic pitcher turned out to be the hitters.

And here the episode pays off one more time. Serling's league ruled that a player must be human, and the rulebook was right about the audience: the humanity requirement wasn't sentimentality, it was the product definition. Casey, given a heart, became useless as a competitor. Today's Casey — the Trajekt — is valuable precisely because it has no heart to break: it will throw two hundred sliders to a rehabbing hitter without fatigue, fear, or an ulnar collateral ligament. The machine found its role not as the hero of the story but as the sparring partner who makes the hero better. Even Casey's ending fits: he left competition for social work. He went into a helping profession. So did the technology.

That shift — from competition to augmentation — is not a baseball story. It is the same arc as aviation, where autopilot changed what pilots do rather than replacing them; as medicine, where the durable wins are decision support rather than robot doctors; and as the enterprise AI debates I usually write about, where the conversation fixates on replacement while the real returns arrive as better preparation, better decisions, and faster learning by human teams. Sports simply ran the experiment first, in public, with a box score.

Science fiction imagined robotic players. Analytics changed how teams were built. AI is changing how teams learn. Three revolutions — and at the end of them, the field belongs more securely than ever to the players with the heartbeats.

But once AI becomes part of player development rather than player replacement, another question naturally follows: who governs the technology? The cameras capturing every skeleton, the sensors on every bat, the models predicting whose elbow fails next — someone has to decide who owns that data and what it may be used for. It turns out baseball answered that question, too, years before most industries knew to ask it. That story — the players' union that negotiated an AI governance framework before your board did — is Part II.

My blog