CSET

What We’re Reading on AI Regulation

Danny Hague

June 7, 2023

The first in a series of posts, this article provides a reading guide on topics relevant to AI regulation. This article shares 101 resources on generative AI, writings on human-AI interactions, and a list of official documents and commentary on U.S. AI policy.

Recent advances in generative AI, including the popularity of ChatGPT, have focused interest on the guardrails that need to be enacted to protect society and secure systems while spurring continued innovation. U.S. policymakers have held high-profile congressional hearings, written front page op-eds, and conducted White House summits with industry leadership. Many are wondering what form the next concrete actions should take. 

While AI regulation is having its moment, a closely related area is that also merits attention is AI safety and security, which can include any number of topics ranging from determining liability for harm caused by AI to auditing AI systems before they are deployed to developing internationally agreed-upon guardrails on when and how AI will be used. 

In this post—the first of in a series—CSET has collected recently published resources to help policymakers and the general public learn more about AI regulation. This post focuses on large language models (LLMs), human-AI interactions, and existing U.S. frameworks and guidelines. 

In some cases, we have also linked to CSET publications on the topic or asked our team to  provide additional context and analysis based on their expertise. We hope readers find this resource useful. Inclusion in this list is not necessarily a CSET endorsement of the views in the piece itself. 

Future posts in this series will explore potential AI governance approaches, current and potential military applications of AI, common practices in AI assessment, and other topics of interest. 

Resources on Proposed U.S. or International Governance Bodies/Structures 

CSET Senior Fellow Heather Frase responding to oft-discussed regulatory body frameworks: “AI is too complex for a single AI agency, domestic or international. I don’t think the IAEA is a good analogy for the government agencies needed. Atomic energy is too focused with limited applications and involved entities. AI is much broader. The financial system is a better analogy, with multiple U.S. regulators and international bodies involved. There are regulators and oversight for specific sectors (e.g. casinos, credit unions, money service businesses, etc.) and for higher-level needs (central bank, consumer financial protection, etc.), along with international organizations (e.g. Egmont, SWIFT, and FATF). We need a network of government and international agencies. However, establishing domestic and international groups to help the System-of-System level of design for those national and international agencies, regulators, and oversight entities would be a useful first step.”

AI, Generative AI and Large Language Model 101s

Human-AI Interactions 

  • The Luring Test: AI and the engineering of consumer trust, a blog post authored by Michael Atleson and published by the U.S. Federal Trade Commission to provide guidance on responsible activity at the intersection of generative AI and advertising.
  • Performative Power, a paper authored by Moritz Hardt, Meena Jagadeesan, and Celestine Mendler-Dünner. This paper was published as part of the 2022 NeurIPS conference and introduces a method for measuring an algorithm’s ability to persuade users. 
  • Artificial Influence: An Analysis Of AI-Driven Persuasion, a paper authored by Matthew Burtell and Thomas Woodside examining how AI may be able to persuade users. 
  • Co-Writing with Opinionated Language Models Affects Users’ Views, a February 2023 paper by Maurice Jakesch, Advait Bhat, Daniel Buschek, Lior Zalmanson, Mor Naaman that evaluates how much language model writing assistants influenced users opinions in their subsequent writings.
  • Trusted Partners, a February 2021 CSET issue brief authored by Dr. Margarita Konaev, Tina Huang and Husanjot Chahal discussing human-machine teaming and the future of military AI. 

Selected U.S. Government AI Frameworks and Guidelines 

We hope you find these resources helpful in thinking through regulatory approaches to AI. In our next post, we will highlight selected international AI regulation efforts, potential governance approaches, and common practices in AI assessment.