AI in the Prisoner's Dilemma: Why Cooperation Is the Most Underrated Professional Skill in the AI Age
AI & Non-linear Thinking · Article I
Two junior consultants at a Frankfurt strategy firm are staffed on the same client engagement. Both know that only one of them will get the positive review that leads to early promotion. The rational move - the one game theory appears to predict - is to compete: hoard information, claim credit, undermine subtly. They both do exactly this. The client notices the dysfunction within weeks. Internal reviews flag communication breakdowns, duplicated work, and a defensive atmosphere that makes the engagement painful for everyone involved. The project is rated "below expectations." Neither consultant gets promoted.
A third consultant, who wasn't even on the engagement, gets the promotion slot. She hadn't outperformed them on any single deliverable. She'd spent the previous eight months doing something neither of them thought to do: sharing credit on cross-team projects, making herself useful to partners outside her immediate reporting line, and building a reputation as someone who made everyone around her better. She didn't win by competing harder. She won by playing a different game entirely.
This is the Prisoner's Dilemma - and most professionals misread it completely.
The Prisoner's Dilemma is the most famous problem in game theory, and it describes a situation most people encounter without recognising it. Two rational actors, each optimising for their own outcome, produce a result that is worse for both. The classic setup: two suspects are arrested. If both stay silent, they get a light sentence. If one betrays the other, the betrayer walks free and the other gets the maximum. If both betray, both get a heavy sentence. The rational move for each individual is to betray - regardless of what the other does. Both betray. Both lose.
Here's what most people miss: this logic only holds in a single-round game. In a one-shot interaction where you'll never see the other person again, defection is rational. But professional life is almost never a one-shot game. You work with the same people for years. You encounter the same clients, the same partners, the same industry network. Your reputation follows you between firms, between cities, between decades. Professional life is an iterated Prisoner's Dilemma — and the iterated version has fundamentally different mathematics.
Robert Axelrod proved this in his famous computer tournaments in the 1980s. He invited game theorists, mathematicians, and computer scientists to submit strategies for repeated Prisoner's Dilemma games. The strategies competed against each other across hundreds of rounds. The most sophisticated, aggressive, and cleverly deceptive strategies all lost. The winner was the simplest strategy submitted: tit-for-tat. Start by cooperating. Then do whatever the other player did last round. Cooperate with cooperators. Retaliate against defectors. Forgive quickly. The lesson was stark: in repeated interactions, consistent cooperation with clear boundaries beats every form of cleverness.
This principle plays out across professional environments in ways that are obvious in retrospect and invisible in the moment.
Consider a Geneva-based NGO coalition negotiating funding allocation across four organisations. Each NGO has an incentive to overstate its needs and understate its resources - the classic defection move. In the first year, two of the four organisations do exactly this. They secure larger initial allocations. By year two, the other organisations have noticed. Trust collapses. The donor reviews the coalition's internal dynamics, flags coordination problems, and reduces the total envelope. The two defectors gained marginally in year one and cost everyone - including themselves - significantly in year two. The coalition that survived was the one where the remaining two organisations maintained transparent reporting. They became the donor's preferred partners for the next funding cycle. Cooperation didn't just feel virtuous - it was the winning strategy in a repeated game.
Or consider two co-founders of a Berlin fintech startup. In the early months, both contribute equally. As the company grows, the division of credit becomes ambiguous. One co-founder begins positioning himself as the visionary in media interviews, subtly downplaying the other's technical contributions. The other notices and retaliates by building relationships with investors independently, creating parallel communication channels that undermine unified decision-making. Within eighteen months, the board intervenes. Not because the product failed - because the founders' defection spiral made the company ungovernable. A competing startup with less impressive technology but a cooperating founding team closes a Series A in the same quarter. The investors explicitly cite "founding team cohesion" as a deciding factor.
Now consider a different kind of consulting firm - one where cooperation is structurally embedded rather than left to individual goodwill. A Dutch firm redesigned its promotion criteria three years ago. Instead of evaluating consultants solely on individual client feedback, they added a metric: how often are you requested by colleagues for cross-team projects? The result was predictable to anyone who understands iterated games: the consultants who shared knowledge, supported junior staff, and built internal alliances rose fastest. Not because the firm was altruistic. Because the incentive structure finally aligned with the mathematics of repeated interaction. They stopped rewarding one-shot defection and started rewarding iterated cooperation.
This is where AI changes the equation - and not in the way most people expect.
When AI commoditises individual analytical output, the Prisoner's Dilemma shifts dramatically. Both Frankfurt consultants can now produce the same quality slide deck. Both can generate the same market analysis, the same financial model, the same strategic framework. The tools are identical. The outputs are converging. In a world where individual production is increasingly interchangeable, the differentiator is no longer what you produce. It's how you collaborate. Trust, reliability, the ability to make a team function better than the sum of its parts - these become the scarce resources precisely because AI has made everything else abundant.
The game theory doesn't change. The payoff matrix does. When individual output carried a high premium, the reward for defection - hoarding information, claiming sole credit - was substantial. You could genuinely differentiate yourself by being the smartest person in the room. AI collapses that premium. When everyone has access to the same analytical firepower, being the smartest person in the room is no longer a viable strategy. Being the most trusted person in the room is.
This isn't a soft argument about being nice. It's a hard argument about payoff structures. Cooperation in an iterated game with commoditised individual output has a higher expected return than defection. The math is unambiguous. The professionals who understand this early - who invest in trust, in networks, in making others better - are building a compounding advantage that accelerates as AI commoditises more of what individuals used to do alone.
So what does this look like in practice? Three things worth doing this week.
First, map your last five significant professional interactions. For each one, ask: was I cooperating or competing? And was this a one-shot game or an iterated one? Most people discover that they're competing in iterated games - which is mathematically irrational. If you'll work with this person again, or if your reputation in this interaction will follow you, cooperation has a higher expected return.
Second, identify your retaliatory threshold. Axelrod's tit-for-tat works because it cooperates by default but retaliates immediately against defection. The professionals who get exploited aren't the ones who cooperate - they're the ones who cooperate without boundaries. Know where your line is. Cooperate generously, but make the cost of defection clear and immediate. Forgive quickly once the defection stops.
Third, ask the uncomfortable question: who in your professional environment are you currently competing against, and is that competition actually serving you? In many cases, the person you think of as a competitor is someone you'd be better off collaborating with. The promotion, the client, the recognition - these feel like zero-sum games. Most of them aren't. Most of them are iterated games where the pie grows when both players cooperate.
The person you're competing with at work is probably not your real opponent. Your real opponent is the assumption that it's a competition at all.
Viktor Hristov