With the rapid advancement of generative AI tools and their increasing use in finance research, fundamental questions are emerging about how research will be conducted and assessed going forward. Perspectives vary widely. Some emphasize the pace of technological progress and the possibility that AI could play an increasingly central role in the research process. Others are more cautious and view AI primarily as a tool for mostly well-defined tasks. A more realistic view likely lies between these positions. The research process involves elements such as taste, judgment, values, and intuition, which are not supposed to be delegated to machines. At the same time, the growing capabilities of AI tools suggest that they can play a meaningful and productive role if used thoughtfully. Finally, value tends to accrue to what is relatively scarce, and our systems for evaluating research should adapt to shifts in the relative scarcity of different inputs to and outputs from the research process.
In this context, the profession would benefit from more systematic evidence on what AI systems can produce in a research setting. A clearer understanding of this frontier helps guide the responsible use of these tools and clarify the nature of scholarly contribution. Against this backdrop, we invite submissions to a special session at the 2027 AFA Annual Meeting aimed at showcasing and evaluating the capabilities of AI in finance research, and based on that evidence, reflecting on appropriate norms for their use and the unique aspects of human contribution. Our objective is not to frame the question in terms of replacement, but rather to better understand where AI tools/agents are most effective within the research process and how they can be combined with human insight to enhance research outcomes.
We invite submissions that are predominantly generated by AI. Authors should submit entirely new projects in which AI systems play a systematic role beyond modular, tool-like functions such as variable construction or analytical assistance. We are particularly interested in work where AI contributes across multiple stages of the research process, with limited human involvement. Authors are required to document their initial inputs, data access, model configuration, and full workflow. Submissions will be evaluated by human reviewers using standard academic criteria, including the novelty and importance of the research question, the incremental contribution to the literature, and the quality of execution. In addition, this session will place specific emphasis on the extent to which the research process is carried out by AI systems, with limited human intervention.
From the submission pool, four papers will be selected for presentation in the special session based on both the research contribution and the design and effectiveness of AI use in the workflow. The expectation is not zero human involvement after the initial prompt, but the highest possible research quality per unit of human expertise and effort. In addition, all high-quality submissions will be included in an AFA 2027 online collection to provide a broad view of current AI capabilities across multiple dimensions.
Human discussants will be explicitly tasked with assessing both scientific merit and the nature of human labor embodied in each paper. They will also articulate what AI-generated finance research lacks relative to the profession’s current standards of originality and excellence. These discussions are intended to inform a broader conversation about how evaluation standards should evolve in response to ongoing technological advances.
The organizers will not use non-public information from the submission and review process for their own research. This initiative is intended as a public forum rather than a setting for data collection.
The ground rules for paper submissions are as follows:
- There is no submission fee.
- Papers can be on any topic appropriate to a top journal in finance.
- No humans perform any work on the project, other than the named authors.
- Each author can only submit one paper; and each author can be listed on at most three submissions.
- The investigation of a paper should begin on or after June 1, 2026 with an initial prompt provided to an LLM or AI agent. The submission deadline is August 31, 2026.
- All conversations with AI, including the initial prompt, must be documented and submitted along with the final paper.
- Humans are allowed to make direct contributions to the project outside of prompts to the AI, but these contributions should also be documented.
- The ideal form of documentation is (1) a text of all conversations with AI, (2) a time log of human activity, and (3) a report on the fraction of project lines (code, writing, and documentation) that were contributed by humans as opposed to AI.
These rules intentionally leave authors wide latitude in how to structure the workflow, and in which tasks the humans adopt given their constrained time. Indeed, this variation is exactly the object of the exploration. The ideal paper is not one with no human involvement, but with maximal quality per unit of human effort.
Submission can be made through the AFA website starting in early June.
Organizers:
Itay Goldstein (Wharton School)
Wei Jiang (Goizueta Business School; Session Chair)
Robert Novy-Marx (Simon Graduate School of Business)
William Mann (Goizueta Business School)