MIT's AI Tool Exposes Hidden Biases in Power Grid Decisions

MIT researchers have developed an AI-driven system to improve decision-making in power distribution. The tool, called SEED-SET, aims to balance cost efficiency with fairness in electricity supply. Early tests show it can uncover biases that leave disadvantaged areas more vulnerable to outages. The system works in two parts. First, it generates a wide range of power distribution scenarios. Then, it identifies the most informative cases for human review, ensuring decisions align with stakeholder values.

In trials, **SEED-SET** produced over twice as many optimal test cases as traditional methods in the same timeframe. It also exposed cases where cost-cutting strategies favoured higher-income neighbourhoods, increasing outage risks for poorer areas. To capture human preferences, the system uses a large language model as a stand-in for stakeholder input. Researchers designed **SEED-SET** to handle multiple objectives, even when ethical criteria are unclear. While tested mainly in power grids and transport systems, the team plans to refine the model for larger, more complex problems. The goal is to make AI-driven decisions both efficient and fair.

The new method highlights trade-offs between cost savings and equity in power distribution. By flagging biased outcomes, it helps decision-makers adjust strategies before real-world implementation. Further development could expand its use in other high-stakes sectors.