The future has arrived with a budget for studying what it might do to you. This is, on balance, better than arriving without one.
In July, Anthropic announced a $200 million commitment to external research through its Economic Futures agenda. The stated priorities include understanding labor-market effects and exploring responses to economic change. The announcement acknowledges uncertainty about how capabilities will diffuse and what consequences will follow.1
A supplier funding research into its own technology’s effects creates a practical test of independence: can the resulting work say something the supplier would rather not hear?
That question is not an allegation that the fund will dictate conclusions. Independence is something to design and demonstrate, not something to assume present or absent from the identity of the donor alone.
A grant is not an impact assessment
The announcement can fund inquiry. The affected worker still has to wait for the findings, then for someone to decide what to do with them.
Who gets to define the question?
Research agendas matter because they determine what receives attention. A study of productivity gains asks something different from a study of bargaining power. A study of occupational change asks something different from a study of household security.
An agenda can become narrow without anyone falsifying a result. It might repeatedly examine what is easy to measure while giving less attention to the people or outcomes that are difficult to observe.
A program needs room for competing questions and inconvenient results. Otherwise, even rigorously executed studies can leave the same unexamined corner of the room outside every frame.
The forecast should include its uncertainty
AI’s economic consequences depend on more than model capability. They involve adoption, prices, demand, organizational choices, complementary investment, and the way institutions distribute gains and losses.
The research earns trust by showing how its evidence supports its conclusions and where it runs out. A capability demonstration should not arrive in the employment debate wearing the credentials of a labor-market study.
A worker whose task changes may experience a benefit, a burden, or both. Higher output does not automatically tell us who receives the gain. An aggregate improvement can coexist with painful losses for particular people. Those are not contradictions to be edited out; they are part of the subject.
Independence has practical features
The credibility of funded work can be strengthened by clear publication rights, transparent methods, disclosure of support, access to appropriate data, and the ability to publish inconvenient results.
These are general standards we would look for, not assertions about the specific terms of every Economic Futures grant. The program’s actual arrangements should be assessed on their own documentation.
Follow the question through the method and data to the conclusion. Keep the funder’s name visible along the way; it is relevant information, not a substitute for reading the work.
The people in the footnotes
Research can make these concerns more legible. It cannot substitute for a public conversation about what should happen in response.
The evidence should be allowed to arrive off message.