highbeamSubscribe

Research

What we measure

Four areas where technology is reshaping people and public life. For each, the questions worth answering and the kinds of evidence that could answer them.

01

Agency and character

Does repeated use change people’s capacity to reason, exercise self-control, act honestly, or honor commitments?

What we would observe

  • Unaided task performance
  • Decisions under conflicting incentives
  • Adherence to self-chosen limits
  • Confidence calibrated against accuracy

02

Relationships and social life

Does use improve connection and cooperation, or displace and weaken them?

What we would observe

  • Subsequent repair behavior
  • Reciprocal human contact
  • Cooperation
  • Relationship quality
  • Effects reported by consenting interaction partners

03

Economic opportunity and power

Who gains capability, income, or bargaining power, and who becomes more dependent?

What we would observe

  • Skill retention
  • Distribution of earnings and gains
  • Switching costs
  • Worker discretion and bargaining conditions

04

Public life and institutions

Does use support informed judgment and participation, and how does it change control over information and decisions?

What we would observe

  • Evidence discrimination
  • Reasoned belief revision
  • Civic participation
  • Concentration of gatekeeping
  • Opportunities to contest automated decisions

These are research domains, not validated scales, and they overlap. Each study picks the few outcomes that matter for its question.

How we choose a study

“AI” is too broad to study. A specific use is not.

What

A specific product or feature

Why

Used for a particular purpose

Who

By a specified population

When

Over a defined period

We record versions, settings, exposure, alternatives and context, and we explain the good being pursued and the possible tradeoffs before observing results. A wellbeing measure is not automatically a measure of virtue; increased trust in institutions is not always warranted; changed political views are not themselves a harm.

See our first programHow we work