Large language models (LLMs): Risks and policy implications

Edited by: JULIAN MANYIKA1 AND AUDREY BERTIN2

Although statistical language modeling has been a fairly active area of academic research in artificial intelligence (AI) since the 1980s, the recent explosion of both technological innovation and public attention around LLMs has dramatically changed the space of applications for this work. The unprecedented scale of these new models carries with it emergent capabilities—particularly in the form of generative models—that have made the technology generalizable to a broad range of use cases and accessible to the general public. Yet these rapidly developing features also pose novel risks, ranging from environmental impacts and copyright violations to misinformation and hate speech.

Traditional regulatory frameworks are not well-equipped to match either the rapid pace of innovation or the ability of LLMs to transcend the jurisdictions of individual governing bodies. Thus far, the governance gap has been filled by soft law frameworks such as private company standards, voluntary codes of conduct, and design guides from non-regulatory standard-setting organizations. Still, there is rising demand to solidify these soft guidelines into hard law. In addition to offering a historical overview of LLMs and highlighting some of the most pressing concerns involving LLMs today, we discuss legislative attempts to address these concerns and outline potential complicating factors.

Highlights

1. In recent years, rapid increases in the performance and accessibility of LLMs have presented new risks that are not comprehensively reflected in existing policy.

2. While these can be organized into general categories of risk, it is important that any risks also be assessed according to any potential complicating factors based on a particular model and its given use case(s).

3. Overall, the European Union (EU) has outpaced the United States (U.S.) in regulatory scope and speed of legislation. In the absence of government policy, industry standards and company initiatives are dictating the terms by which LLMs are currently being developed and deployed.

READ HERE:

This MIT Science Policy Review article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/ by/4.0/.