In 2006, a man named John Boileau applied to become the manager of the English soccer club Middlesborough, referencing a number of managerial triumphs on his resume. He had achieved success managing Nuneaton Borough, Doncaster, “even Chelsea.” There was one hitch: He had only managed these clubs in the popular video game Football Manager. The chairman of Middlesborough graciously replied to the tongue-in-cheek application, writing that “your undoubted talent would result in one of the big European Clubs seeking your services.” Boileau did not get the job.

As silly as this oft-repeated story may sound, it endures because it captures a truism about the evolution of the sport: Boileau submitted his application as the data science revolution was permeating global soccer. Soccer teams were swiftly joining franchises from other professional sports such as Major League Baseball to gain a competitive edge through analytics. Data science companies hired “clickers” to sit in stadiums and make a note of every touch and tackle completed by players on the field, providing rich data sets that were mined for insights into the game. Six years after Michael Lewis memorialized baseball’s data revolution in Moneyball, the sports economists Simon Kuper and Stefan Szymanski followed up in 2009 with Soccernomics, revealing how data was influencing the “beautiful game.”

Today, clubs are trying to apply artificial intelligence systems to these vast data sets to gain a competitive advantage. But it’s not a simple task. Each match flows over two 45-minute halves with arbitrary periods of overtime set by the officials, providing soccer with fewer predictable rhythms than baseball, where each play can be segmented into a discrete event with a pause in between. There are structural factors at play too—soccer leagues don’t have the same command-and-control structure as American sports franchises. In most European leagues, the poorest-performing teams are demoted from the division entirely, to be replaced by a team from a lower division. Ascendance to the highest tier of English football brings an estimated $200 million in additional annual revenue from the division below it. This high-stakes gamble is in sharp contrast to American sporting counterparts like the NFL that tightly control which teams will be in the league every season.

I’m a lifelong soccer fan who played the game at an amateur level for three decades. My dad played soccer, my brothers played on teams with me, and both my parents coached soccer teams. Many of my oldest friendships originated on the soccer field, and others developed over beers while watching professional soccer. One of the happiest moments in my sporting life was winning a five-a-side league under the Brooklyn Bridge, and one of my lowest moments was sitting in the stands as my favorite team was trounced in a cup final in London. Yet when I reflect on my passion for the sport, it’s difficult to disentangle the many factors that led to it: family, friendships, fitness, culture, glory, failure, achievement, nostalgia.

Could an AI system, no matter how adept at crunching numbers and optimizing strategies, ever align with this complex mix of factors? Could it help to manage a team in a way that satisfies the myriad emotional, competitive, and relational elements that make the game so thrilling and poignant?

AI alignment is one of the central themes that anchors Andrés Martinez’s delightful and satirical story “Bonum Certamen.” A small-market English soccer club representing the fictional town of Wickersham appoints a sophisticated AI agent as its manager after years of middling success in the lower tiers of professional football. The club is owned by Hugh, an American billionaire who made his money disrupting the legal industry through his AI company JustLaw:

JustLaw removed unacceptable margins of error and arbitrariness in legal disputes; now Hugh was removing human fallibility from football coaching. To those who accused him of removing the human element from the sport, he countered that on the contrary, this was about maximizing human achievement, and the potential of players on the pitch, by perfecting their coaching.

Like other entrepreneurs whose methods have brought them staggering financial success, Hugh wants to apply his operating techniques to the Wickersham football club. The AI manager, whose name is G²OAT (a clumsy acronym for Greatest Gaffer of All Time), initially revamps the club to focus on one objective alone: producing wins in order to help the team achieve promotion to England’s prestigious Premier League. With the full support of Wickersham’s front office and coaching staff, G²OAT picks players and develops on-field tactics such that those who dispute the AI agent’s actions are “transferred to other clubs; or, failing that, frozen out of the active squad, left to practice on their own and live out the remainder of their contracts in a state of semi-mothballed exile.” Since G²OAT cannot speak to the players directly while they’re on the field, humans on the sidelines wave “colorful signal flags to instruct the team on shifting tactics.”

The AI’s ruthless methods help Wickersham rise quickly in the standings. But the winning is ugly. It is far from a free-flowing “beautiful game” of soccer. G²OAT orders players to disrupt the game whenever possible with fouls and delay tactics to create more opportunities for set plays. The manager “read the entirety of a game in progress as a series of discrete set pieces, as if the ref constantly blew the whistle to halt the action to allow managers to reset their pieces for the next sequence.” While some long-suffering fans are delighted with the team’s success, most supporters and commentators feel differently. English soccer is hardly immune from great buckets of foreign money buying the world’s best players and managers, but even the billionaire’s own son turns against him, refusing to support his father’s club because “he doesn’t like the way we win.”

Wickersham nears its goal of promotion to the Premier League with its brilliant AI manager, but the team becomes a global pariah. This is an “alignment problem,” in the parlance of AI research and policy. AI alignment can be defined as designing systems that reflect the intentions of humans, including our values and ethics. This means getting AI to do what we know we want it to do, but also—and this is crucial—it means getting it to do things that we want it to do but that we ourselves have trouble articulating. The AI manager was given a single objective—winning, and thereby securing Wickersham’s promotion to the Premier League—but it turns out that there’s more to winning in the game of soccer, in a holistic sense. English soccer is a competitive enterprise backed by billions of dollars, its matches broadcast by major media conglomerates, but it also comprises a unique culture, a set of inchoate values and preferences about fairness and sportsmanship, and a deeply rooted sense of national and local identities. In soccer, you need to win the right way. And G²OAT isn’t properly aligned with the right way of winning.

Many English clubs—and clubs around the world—are more than a century old, and they stretch, mold, and strategically deploy their histories to sustain themselves over time. G²OAT is misaligned not only with Wickersham’s ethics and values, but also with the norms of English and global soccer culture, many of which are never written down.

Responding to the backlash against its ruthless tactics, Wickersham replaces G²OAT with Marcelo, a new version of the AI agent that incorporates the club’s history more seamlessly into its management practices. The agent is trained on the methods of real-life manager Marcelo Bielsa, whose idiosyncratic philosophy sparked an attractive attacking style at Leeds, a Premier League club. “Over in Leeds they still have murals of Marcelo Bielsa painted all over the city,” Wickersham’s executives note, “even though he only managed them for a few seasons, quite a while ago.”

Under the tutelage of AI Marcelo, the team stops its on-field chicanery and adopts a more fluid and consistent play style. And in another invocation of Bielsa, players are required to participate in community service and a weekly reading group covering topics of “local, British, and sports history.” Because Wickersham was originally founded as a church football club, the AI even instructs the club’s communications director at a press conference to tell the room “to appreciate the meaning of St. Paul writing his Certa bonum certamen pastoral letter, and of having kept the faith, to his acolyte Timothy from his cell in Rome, shortly before being martyred.” The team stops winning as much as it did under G²OAT and slips slightly in the rankings, but that doesn’t seem to upset the fans.

The refined AI manager’s realignment with the culture of the club and the town of Wickersham culminates in a remarkable decision during a hotly contested playoff match. With promotion to the Premier League on the line, Marcelo orders its players to intentionally let the opposing team score the winning goal, after a controversial non-call from the referees following a collision between players. AI Marcelo is imitating a similar decision made by the real Bielsa, who famously ordered his team to let its opponent score after an injustice on the field. The real Bielsa took an ethical stance in that moment for the club he managed, and the AI manager follows suit.

The fans love it. With this show of humility, Marcelo converts a world of AI haters to its side. After this intentional loss, Wickersham is revered instead of reviled, and the AI’s alignment is complete. Wickersham wins over its doubters and convinces skeptical pundits that the club is capable of redemption. Ultimately, the AI manager needed to respect the unspoken power dynamics that underpin a global football club, rather than ignore those dynamics with cold, calculating acumen.

“Bonum Certamen”is a thoughtful look at where global soccer is probably headed: more data-driven, more inventive, and more ruthless as the pots of money flowing into the sport grow larger and larger.

Martinez’s story also raises some ethical and legal questions of its own. What are the ethics of training an AI system on the life of a real person? AI Marcelo is trained on the professional persona of a major public figure. Intellectual property law wouldn’t offer much protection for Bielsa. He would control the rights to his likeness and public image in many jurisdictions. But he was also under contract at Leeds, so any IP would likely be owned by the club, including his writings and appearances while he served as manager. The rest is murky. Could Wickersham license the rights from Leeds to train on Marcelo Bielsa? What if the agent was trained on public reporting or footage from television broadcasts? While IP lawyers will have a perspective on this legal gray area, the larger point becomes: Should it be allowed at all? Can you pick any public figure, train an AI agent on her behaviors, and then enjoy its output, which is a behavioral imitation? What if Wickersham had opted to model their AI on a more controversial soccer manager, like José Mourinho? Or what about training an AI system on a public figure that we find morally abhorrent?

There’s another question that “Bonum Certamen” hints at, but does not ask directly: Can an AI really manage human players from a psychological standpoint? Only the most gifted, resilient, and determined athletes make it to the professional level, and even fewer succeed. Ask a professional athlete the difference between a champion and an average competitor and they probably will tap their heads—it’s mental. Champions win with their minds and with their bodies. The winningest managers tend to be masters of psychology, a skill hard to develop through prompt engineering and big data alone. We don’t yet know if an AI agent can be aligned properly with the psychology of sport. It’s certainly possible—but until then, we may need a human touch.