10 Highlights From Stanford University's Generative AI Report

This weekend I had a chance to read the new report on generative AI from Stanford University Human-Centered Artificial Intelligence (HAI). Stanford HAI is committed to studying, guiding, and developing human-centered AI technologies that enhance human productivity and quality of life. Stanford HAI leverages the university’s strength across all disciplines including law, medicine, genomics, business, engineering, technology, economics, literature, biology, chemistry, neuroscience, and philosophy. The new report includes essays from Stanford leaders on the potential impact of generative AI and how to best apply generative AI to advance various industries. In today's newsletter I'm sharing 10 highlights from the report. Click here to read to full report on the HAI website.

10 Highlights From The Report

"Although the exact figure is disputed, death due to medical error in the US is a significant problem. Generative AI models could assist healthcare providers in seeing potential issues that they may have otherwise missed. Furthermore, if the mistakes are due to minimal exposure to rare situations, generative AI can create simulated versions of this rare data to further train the AI models or the healthcare providers themselves."

Fei-Fei Li, Sequoia Capital Professor in the Computer Science Department; Denning Co-Director of Stanford HAI (page 4)

"It is often difficult to get large numbers of patients in clinical trials and it is crucial to have a realistic group of patients who do not receive a therapy in order to compare outcomes with those who do. This is one area within biomedical research where generative AI offers great opportunities...This could make trials potentially smaller, faster, and less expensive, and lead to faster progress in delivering new drugs and diagnostics."

Russ Altman, Kenneth Fong Professor in the School of Engineering; Professor of Bioengineering, of Genetics, of Medicine, and of Biomedical Data Science; Associate Director of Stanford HAI (page 6)

"Generative AI’s well-reported challenges with factual correctness are particularly problematic in medicine, where inaccuracies can cause serious harm. Recent problems in medicine include incorrect differential diagnosis and invalid scientific citations. We are working to improve the factual correctness of medical explanations from these models so they can achieve an accuracy that is suitable for safe clinical use."

Curt Langlotz, Professor of Radiology, of Biomedical Informatics Research, and of Biomedical Data Science; Director of the Center for Artificial Intelligence in Medicine and Imaging, Associate Director of Stanford HAI (page 7)

"Generative models of proteins can allow us to efficiently explore the space of complex three-dimensional protein structures, thereby aiding in the search for proteins with novel and useful functions, including new efficacious medicines... AI generative models are well poised to deliver considerable insights into nature itself, across biological, physical, and mental realms, with broad implications for solving key societal problems.

Surya Ganguli, Associate Professor of Applied Physics; Associate Director of Stanford HAI (page 8)

"Imagine a smart tutor that is always patient and understands the level of knowledge the student has at any point in time on any subject. These tutors will not replace teachers, but instead will augment the student learning experience – giving students a more personalized interaction, focusing in areas where they might be weaker."

James Landay, Anand Rajaraman and Venky Harinarayan Professor in the School of Engineering and Professor of Computer Science; Vice Director of Stanford HAI (page 10)

"Generative AI may simply automate a highly reductive notion of both the creative process and of the learning process itself... Consider the slowed-down, recursive reading and interpretive skills required to understand any piece of writing by Toni Morrison... Just try the experiment of my students, who submitted an excerpt of Toni Morrison’s?Beloved?to Grammarly, which attempted to correct her exquisite prose for what sociolinguists term “standard English,” and quickly saw how even deeply rich meaning can be rendered impotent."

Michele Elam, William Robertson Coe Professor in the School of Humanities and Sciences and Professor of English; Associate Director of Stanford HAI (page 11)

"Just as legal databases such as Westlaw and Lexis revolutionized legal research, there is the potential for generative AI to help individuals prepare legal documents, attorneys in legal research and writing, and judges to improve the accuracy and efficiency of painfully slow forms of adjudication. While the industrial organization of legal search could get in the way, generative AI could help level the legal playing field."

Daniel E. Ho, William Benjamin Scott and Luna M. Scott Professor in Law at Stanford Law School and Director of the Regulation, Evaluation, and Governance Lab (RegLab); Associate Director of Stanford HAI (page 13)

"It is absolutely critical that we benchmark these foundation models to better understand their capabilities and limitations as well as use these insights to guide policymaking. Toward that end, we recently developed HELM (Holistic Evaluation of Language Models). HELM benchmarks over 30 prominent language models across a wide range of scenarios (e.g., question answering, summarization) and for a broad range of metrics (e.g., accuracy, robustness, fairness, bias, toxicity) to elucidate their capabilities and risks."

Percy Liang, Associate Professor of Computer Science; Director of Stanford Center for Research on Foundation Models (page 15)

"At Stanford Digital Economy Lab, we are cataloging the list of economic activities likely to be affected by generative AI and estimating what share of the economy they represent... In cases where generative AI can be a complement to labor, particularly for knowledge workers and the creative class, wages could increase even as output increases... These technologies have the potential to speed up the rate of innovation itself, by facilitating invention, design, and creativity. Thus they may not only increase the level of productivity but also accelerate its rate of change."

Erik Brynjolfsson, Jerry Yang and Akiko Yamazaki Professor at Stanford HAI; Director of Stanford Digital Economy Lab (page 16)

"Generative language models provide a massive opportunity to reinvent how work is done inside all sorts of companies and industries: Marketing, sales, product development, customer support, and even human resources will all change... In nearly all cases, the AI system will help humans to get work done. As such, it continues the story of new technologies and automation making things easier and improving quality of life."

Christopher D. Manning, Thomas M. Siebel Professor in Machine Learning at the School of Engineering; Professor of Linguistics and of Computer Science; Director of Stanford AI Lab; Associate Director of Stanford HAI (page 18)

Subscribe and Comment

I'm interested in your feedback - please leave your comments. To subscribe please click subscribe at the top of this article.

Copyright ? 2023 Margaretta Colangelo. All Rights Reserved.

This article was written by?Margaretta Colangelo. ?Margaretta is a leading AI analyst who tracks significant milestones in AI in healthcare. She's consulting at AI healthcare companies and she writes about some of the companies she's consulting with. Margaretta serves on the advisory board of the AI Precision Health Institute at the University of Hawai?i?Cancer Center?@realmargaretta