Structural bias in multi-objective optimisation

Researchers have discovered that algorithms used for multi-objective optimization, which balance competing goals like cost and performance, can have a built-in 'structural bias'. This bias makes them favor certain solutions regardless of the actual problem, similar to a scale that's tilted before you weigh anything. The team created a new testing method to isolate and measure this hidden bias, which has been overlooked until now. Their work provides tools for building fairer and more reliable AI systems.