AI & Expertise research with patent lawyers

How to cite

Autor, David; Pearl, David; Strand, Scott. 2025. "AI & Expertise research with patent lawyers." AEA RCT Registry. AEARCTR-0015823. https://doi.org/10.1257/rct.15823-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

David Autor* , David Pearl , Scott Strand

* Primary investigator

Registry ID
AEARCTR-0015823
DOI
https://doi.org/10.1257/rct.15823-1.0
First registered
Last updated
Status
In development
Trial dates
2025-05-09 – 2025-12-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/15823

Abstract

The study explores how access to AI tools, supplemented by corresponding training focused on most appropriate ways to apply AI in patent drafting, affect professional output, skill acquisition, and job satisfaction of patent lawyers at various levels of expertise. In particular, it tests whether providing access to a specialized AI assistance tool for drafting and critiquing patent claims improves the quality and speed with which tasks are completed as well as the speed and extent of skill acquisition. Findings will inform how AI reshapes work in high-skill domains.

Intervention

Control: Lawyers will use their traditional process without AI assistance when performing experimental tasks.
Treatment: Lawyers will have access to a specialized AI tool when performing experimental tasks and in their day-to-day work after receiving training on how to work with AI effectively in the patent drafting context.

Experimental design

This experiment is designed as a two-part randomized controlled trial (RCT) to investigate the impact of an AI tool on patent lawyers' work. The first part involves lab experiments where junior and expert patent associates are asked to draft patent sections and critique existing drafts, both with and without AI assistance. These tasks are performed twice, with a three-month interval, and the outputs are evaluated by blinded third-party raters for quality. The second part is a field experiment where the AI tool is rolled out to patent law firms working with Google, and the quality of their real-world patent drafts created with and without AI is compared by internal raters, along with surveys assessing user experience and perceived changes in efficiency and proficiency.

Design details

Randomization method
Computer randomization from a list
Randomization unit
Individual
Clusters
350 maximum
Observations
350 maximum
Arms
Control: 33% of participants Treatment: 69% of participants
Primary outcomes
Quality of completed tasks (As rated by independent 3P raters) ; time on task (self-reported) ; perceived quality & time gains or losses (self-reported); satisfaction with patent drafting experience (self-reported)

Cited by

No citation data available for this registration yet.

Keywords