AI Impact & Externalities.

Measure labour displacement

Jobs change or disappear, and the people affected are not part of the plan.

Measure productivity effects

Productivity gains are claimed without a measure of who got faster and who got more work.

Measure market concentration

AI advantage piles up with firms that already have data, compute, and distribution.

Measure distribution of economic gains

Gains accrue to owners and vendors, not to the workers and customers who bear the change.

Assess effects on communities

A community lives with the system effects and has no say in them.

Assess effects on institutions

Schools, courts, clinics, or newsrooms change how they work because of the system.

Assess social cohesion

AI-mediated interaction weakens trust between groups.

Detect changes in human behaviour

People change how they act because the system rewards or nudges that behaviour.

Measure AI energy consumption

Training and inference draw power that is not counted in the product cost.

Measure water consumption

Cooling and power generation use water that does not appear in the product account.

Improve compute efficiency

The same result costs more energy and hardware than it needs to.

Measure AI infrastructure footprint

Data centres, chips, and networks add an environmental cost outside the app.

Prevent cascading AI failures

One system failure triggers failures in systems that depend on it.

Map cross-sector dependencies

Critical sectors rely on the same models and fail together.

Measure concentration of AI capabilities

A few providers hold capabilities everyone else depends on.

Detect large-scale behavioural effects

A widely used system shifts behaviour across a population, not just one user.

Close the digital divide

People without reliable access, devices, or skills are left out of the benefit.

Reduce unequal access to AI

Price, language, or region decides who can use the system.

Measure concentration of AI benefits

Benefits cluster with groups that already had an advantage.

Expand access to AI infrastructure

Compute, models, and data stay with organisations that can pay for them.