We Need Better Output, Not Just More: Claudine Gartenberg Live with Tim O’Reilly

Ever since the release of ChatGPT, the editors at the academic journal Organization Science have been tracking a surge in low-quality AI-augmented submissions that threatens to overwhelm the peer review process. So the journal’s AI Task Force decided to analyze just how generative AI is impacting scientific research in their corner of the world, and wrote up their findings in “More Versus Better: Artificial Intelligence, Incentives, and the Emerging Crisis in Peer Review.” Generative AI tools have reduced the effort it takes to produce “research” in the form of article submissions (up 42% since 2022), but as the editors discovered, the more AI-augmented an article was, the lower its chances for making out of review. So even as the volume of submissions is up (“more”), it’s not translating to “better” outcomes for research or for the academics themselves. What it is doing is making it more difficult and more labor-intensive to uphold the journal’s quality standards. That’s a problem that should resonate with anyone whose day-to-day responsibilities have become swamped by a deluge of AI-augmented “work.”

Tim asked Claudine Gartenberg, an associate professor of management at the Wharton School at the University of Pennsylvania and one of the authors of “More Versus Better,” to join him to talk about her research on how AI is reshaping organizations and explain how leaders can begin solving problems like the one Organization Science faced and chart a course toward “better.” Claudine predicts that the dominant themes for organizations in the near-future will be “validation” (determining whether AI-produced work is good and reliable) and “integration” (absorbing that work into their decisions, workflows, and learning). If production is out of sync with validation and integration processes, all that “more” productive work will be for naught as systems are inundated and the pipeline gets blocked. While there’s no clear agreement on how to do this well, organizations will probably have to get more creative with validation and integration, she argues.

Claudine will share some ideas about human-AI collaboration drawn from her research on AI in science that can help firms figure out which problems to solve and how, and examine how successes in one field, like software engineer or management, may or may not translate to others. She and Tim will also explore the ways misaligned incentives, say “publish or perish, ideally in a top-tier journal” or “prove your productivity by maximizing token use,” keep us from achieving “better,” and propose some that could actually help transform organizations—maybe by rewarding people for sharing what they’ve learned about AI collaboration, as Tim has noted elsewhere.