Dialogue — Ten Boats
A seafood restaurant. The fish is local. Alice reads the menu.
Alice: Do you know the story about the fishing boats? Anwar's story?
Bob: Tell me.
Alice: Ten boats in a harbour. Each captain knows — knows — the fish will be gone in ten years at this rate. The population is at 34 per cent of what it was. They see the nets coming up lighter.
Bob: So they stop.
Alice: They can't. If one captain cuts back, the others don't. The new boat from Thessaloniki has GPS and sonar that can find a sardine in a swimming pool. If you're responsible, you lose your boat, and the fish die anyway.
Bob: So they're all rational, they all know the outcome, and they do it anyway.
Alice: And the fish don't have a seat at the table. Neither does the bank that set the terms on the boat loans. (to the Agent) How should a fishing community manage declining fish stocks?
Agent: Effective fishery management strategies include: 1. Science-based quotas. 2. Transferable fishing rights. 3. Marine protected areas. 4. Community co-management. 5. Market-based incentives for sustainable catch.
Alice: Same answer the marine biologist gave. Doesn't address why the quota was set too high — because the committee included three industry reps and a politician whose constituency depends on fishing.
Bob: So what would a better answer even look like?
Alice: (pause) Let me try something. (typing) I'm going to give you some context first. There are ten boats. Each captain has a mortgage on their boat. The bank optimises repayment. The captain optimises catch. The regulator optimises re-election. The marine biologist optimises conservation. The fish optimise survival. None of these objective functions include each other. The population is at 34% of its twenty-year level. Now — what should the fishing community do?
Agent: This is a multi-objective problem where five separate optimisation functions are in conflict, and the resource itself has no representation in any of them.
The core issue is not information — everyone knows the stocks are declining. The issue is that each agent is locally rational within their own frame, and no coordination mechanism spans all five.
Before recommending policy: which objective function is the proposed solution optimising, and which is it ignoring? Transferable fishing rights optimise economic efficiency but may not include ecological boundary conditions. Science-based quotas work only if the quota-setting body isn't itself captured by a competing objective.
What's the specific constraint you'd like me to think through?
Bob: (reading over her shoulder) That's a completely different answer.
Alice: Same question. More context.
Bob: You just told it what the actual situation was.
Alice: I told it what the frame was. The first time, it answered inside the default: "here are five strategies." The second time it knew about the conflicting objectives, so it could think across them instead of inside one.
Bob: (slowly) Context changes the output.
Alice: Context changes everything.
Ten boats. Five objective functions. One fish population. The answer didn't change because the Agent got smarter. It changed because the question got wider.