🔬 Research summary by Jonas Schuett, Policy Research Intern at DeepMind | Research Fellow at the Legal Priorities Project | PhD Candidate in Law at Goethe University Frankfurt
[Original paper by Peter Cihon, Jonas Schuett, Seth D. Baum]
Overview: How can different actors improve the corporate governance of AI in the public interest? This paper offers a broad introduction to the topic. It surveys opportunities of nine types of actors inside and outside the corporation. In many cases, the best results will accrue when multiple types of actors work together.
Private industry is at the forefront of AI research and development. Corporate AI researchers publish more than their academic counterparts, and their research is often productized into AI systems deployed around the world. As AI research and development has an increasingly outsized impact on the world, it is essential to ensure that the governance of the field’s leading companies supports the public interest.
Opportunities to improve the corporate governance of AI
The opportunities to improve AI corporate governance are diverse. The paper surveys opportunities for nine different types of actors:
- Management can establish policies, translate policies into practice, and create structures such as oversight committees.
- Workers can directly affect the design and use of AI systems, and can have indirect effects by influencing management.
- Investors can voice concerns to management, vote in shareholder resolutions, replace a corporation’s board of directors, sell off their investments to signal disapproval, and file lawsuits against the corporation.
- Corporate partners can use their business-to-business market power and relations to influence companies, while corporate competitors can push each other in pursuit of market share and reputation.
- Industry consortia can identify and promote best practices, formalize best practices as standards, and pool resources to advance industry interests, such as by lobbying governments.
- Nonprofit organizations can conduct research, advocate for change, organize coalitions, and raise awareness.
- The public can select which corporate AI products and services to use, and also support specific AI public policies.
- The media can research, document, analyze, and generate attention to corporate governance activities and related matters.
Coordination and collaboration
In many cases, the best results will accrue when multiple types of actors work together. The paper shows this via extended discussion of three running examples:
- First, workers and the media collaborated to influence managers at Google to leave Project Maven, a drone video classification project of the US Department of Defense. Workers initially leaked information about Maven to the media, and then signed an open letter against Maven following media reports.
- Second, nonprofit research and advocacy on law enforcement use of facial recognition technology fueled worker and investor activism and public pressure (especially the 2020 protests against racism and police brutality) that ultimately pushed multiple competing AI corporations to change their practices.
- Third, workers, management, and industry consortia have interacted to develop and promote best practices concerning the publication of potentially harmful research.
Between the lines
The paper will be of use to researchers looking for an overview of corporate governance at leading AI companies, levers of influence in corporate AI development, and opportunities to improve corporate governance with an eye towards long-term AI development.