Living with Algorithms: Human-AI Assemblages and Ethical Governance in West African Pastoral Conflict

The Big Picture: When Computers Try to Run the Village

Imagine a group of neighbors who have lived together for centuries. They have a complex, unwritten set of rules for sharing water, grazing land, and settling arguments. It's a system based on whispers, handshakes, and knowing exactly who owes whom a favor. It's messy, flexible, and deeply human.

Now, imagine a team of tech experts arrives and says, "We have a super-smart computer program that can solve all your problems faster and more fairly than your old ways." They install cameras, biometric ID scanners, and an AI that predicts where fights will happen.

This paper asks: Does this computer actually make things better, or does it break the delicate social fabric that keeps the peace?

The researchers studied this exact situation in two places: Plateau State in Nigeria and the Tillabéri Region in Niger. These are areas where farmers and herders often clash over land and resources.

The Main Discovery: The "Legitimacy Gap"

The study found a strange paradox. The computers were actually good at their technical job. They were faster at sending alerts and better at spotting patterns than the old human systems.

But, the people living there didn't trust them.

Think of it like a GPS that gives you the fastest route but drives you through your neighbor's living room. The GPS is technically "correct" (it's the fastest path), but you wouldn't use it because it violates social norms and makes your neighbors angry.

The paper calls this the "Legitimacy Gap." The system was efficient (fast), but it lacked legitimacy (it wasn't seen as fair or right by the community). When people don't trust the system, they stop cooperating, and the system eventually stops working.

Three Ways the System Broke Down

The researchers found three specific ways the "smart" system clashed with "real" life:

1. The "Rigid ID Card" Problem

• The Tech: The system used biometric ID cards to identify people. If you were a herder, the card said "Herder." If you were a farmer, it said "Farmer."
• The Reality: In these communities, identities are fluid. A man might herd cattle in the dry season and farm crops in the wet season. He might be part-Fulani and part-Berom.
• The Analogy: Imagine a library that only lets you check out books if your ID says "Fiction Reader." If you want to borrow a cookbook, the computer says, "No, your ID says Fiction." It doesn't understand that you are a whole person with many interests.
• The Result: The computer mislabeled people, leading to unfair decisions about who got help and who didn't.

2. The "Gaming the System" Problem

• The Tech: The AI predicted conflict based on data like herd sizes and movement.
• The Reality: People realized the computer was watching. So, they started tricking it.
• The Analogy: It's like a student who knows the teacher is grading on a curve. Instead of studying, the student figures out exactly what questions the teacher likes to ask and only studies those.
• The Result: Herders started hiding their GPS collars at night or lying about how many cows they had. This "gaming" made the computer's data messy and less accurate over time. The system got dumber because people were actively trying to outsmart it.

3. The "Quiet Rebellion" Problem

• The Tech: The AI used a "risk score" (a color-coded map) to decide who got emergency aid. If the rain was late, the computer said, "It's too risky to plant; no aid for you."
• The Reality: The farmers knew the computer was wrong because they knew the local seasons better than the satellite. Instead of arguing with the computer, they just ignored it.
• The Analogy: Imagine a strict parent who says, "If it's raining, you can't go outside." The kids know the rain is just a drizzle, so they go outside anyway and pretend they didn't.
• The Result: The computer thought everything was fine (because no one reported the planting), but in reality, the farmers were working hard. The computer was blind to the truth because the people refused to play by its rigid rules.

The Solution: The "Human-AI Hybrid Tribunal"

The paper argues that we can't just fix the code. We have to change how the computer and the humans work together. They propose a new system called a "Human-AI Assemblage."

Think of this not as a robot judge, but as a co-pilot system.

They tested a three-step fix:

1. Participatory Thresholds (Setting the Rules Together):
   Instead of the tech company deciding what counts as a "danger zone," they held a workshop with farmers, herders, and local leaders. They adjusted the computer's sensitivity together.

   • Result: They reduced false alarms by 50% because the locals knew the computer was being too paranoid.

2. Public Bias Audits (Checking the Scorecard):
   Every few months, they publicly checked if the computer was treating different groups fairly. If the computer denied aid to herders 18% more often than farmers, they flagged it immediately.

   • Result: This forced the system to correct its mistakes before they caused too much harm.

3. Hybrid Tribunals (The Human Override):
   They created a local court made of elected farmers, herders, and a tech expert. If the computer said "No Aid," this group could say, "Wait, we know this family is actually in trouble. Override the computer."

   • Result: People started trusting the system again because they knew a human could fix a computer error. Cooperation jumped from 58% to 79%.

The "Formalization Frontier"

The paper introduces a cool concept called the "Formalization Frontier."

Imagine you are trying to write down a recipe for a perfect stew.

• 0% to 55%: You write down the ingredients and cooking time. This helps a lot.
• 55%: You hit a wall. If you try to write down every tiny detail (the exact mood of the chef, the humidity in the kitchen, the specific brand of salt), the recipe becomes useless. It's too rigid.

The researchers found that once you try to turn more than 55% of local life into hard data, the system starts to lose trust. You can't code human flexibility perfectly.

The Bottom Line

The paper concludes that you cannot have a successful AI system without trust.

• Old Way: Build a faster, smarter computer and hope people use it.
• New Way: Build a system where the computer is a tool, but humans are the bosses. The computer provides data, but humans provide the judgment, fairness, and context.

The study proves that 63% of the system's success depends on "Ethical Legitimacy" (people trusting it), not just on how fast the computer calculates. If you ignore the human element, the computer will eventually be ignored by the humans.