This demonstration issue tests Franklin’s editorial format using public research materials on artificial intelligence, entrepreneurship, and information. It is not presented as a complete weekly scan. Each entry identifies the evidence used and links directly to its source.
Prototype issue · 1 source checked · 3 papers included
01Author working paperAuthor description
Training AI For When Humans Will Use It
Kevin A. Bryan and Joshua Gans
Economics of AIDecision theoryHuman–AI interaction
Question
How should AI be trained and valued when a human remains part of the decision process rather than accepting its output automatically?
Approach
The authors treat AI as one component of a composite experiment: a person can act on its recommendation, gather more information, consult other systems, or attempt verification. They study how this structure changes optimal training and the value of AI across decision-theoretic and game-theoretic settings.
Findings
The authors report that the usefulness of an AI system depends on the actions available to its human user, not only on standalone predictive performance. As a result, the best training objective and the value assigned to improved AI can vary with the surrounding decision environment.
Why it matters
The framework shifts AI evaluation toward the actual human workflow in which a model is used—a relevant distinction for firms deciding where better prediction will create value.
Caveat
Franklin used the authors’ public description rather than assessing the full paper. Specific assumptions, propositions, and boundary conditions therefore remain unreviewed.
02AEJ: Applied EconomicsFull-paper abstract
Information Frictions and Employee Sorting Between Startups
Would job seekers apply to higher-quality startups if they received credible information about the firms’ science and business models?
Approach
The researchers built a job board for 26 science-based startups and invited nearly 20,000 business-school alumni. Applicants were randomly shown coarse expert ratings of startup science quality, business-model quality, both, or neither; a second experiment examined related choices among MBA students.
Findings
Displaying expert ratings shifted applications toward better-rated firms. Positive science information raised application probability by 12 percent and negative information reduced it by 24 percent; corresponding changes for business-model information were +29 and −12 percent. Applicants nevertheless remained highly optimistic about startup success.
Why it matters
Credible outside signals may improve how skilled workers sort across young firms, where familiar indicators of employer quality are often missing.
Caveat
The main field experiment covers 26 actively hiring firms connected to one science-based entrepreneurship program. The size of the response may differ in other labor markets or startup populations.
03Journal of Economic LiteratureAuthor abstract
The Economic Impacts of AI: A Multidisciplinary, Multibook Review
Kevin A. Bryan
Economics of AITechnology adoptionPolicy
Question
What do prominent social-science accounts explain about AI’s economic effects, and which important questions do they leave unresolved?
Approach
The essay reviews seven books published over roughly twelve years, comparing their treatment of prediction, organizational adoption, data, implementation, and broader economic change.
Findings
The reviewed work offers useful accounts of AI as cheaper prediction, the organizational barriers to adoption, and the economic role of data. The author argues that it provides less guidance for scenarios involving rapid labor churn, accelerated science, major shifts between labor and capital, or existential risk.
Why it matters
The review identifies a gap between careful analysis of AI inside existing production systems and the policy questions raised by more transformative technological change.
Caveat
This is a review essay organized around seven selected books, not a systematic review of all empirical and theoretical research on AI’s economic effects.
Archive
A quiet record, week by week.
July 30, 2026Signals, judgment, and the economics of AI3 papers