AI Ethics: Pausing AI? The Ethics, History, Epistemology, and Strategy of Technological Restraint

If the development of certain technologies, such as advanced AI systems, is as dangerous as some suggest, is there any hope that we might pause, restrain, or contain their development? To many, this might look deeply implausible, given the considerable commercial and strategic interest in AI. Yet there is a surprisingly diverse historical track record of technological delay and restraint, even for strategically promising technologies that were seen as near-inevitable in their time.

Epistemic hurdles to studying undeployed and cancelled technologies make us prone to underestimating the frequency of restraint; to misinterpret its causes; and to underestimate its future feasibility. This does not mean that restraint for AI is currently desirable, that it would be an easy or wise or competitive strategy. However, it suggests that investigating when and why restraint might be needed and viable is an interesting, legitimate and useful part of any toolset to enable responsible and deliberate choice in steering technological development.

Matthijs Maas is a Senior Research Fellow in Law & AI at the Legal Priorities Project, and a Research Affiliate at the Centre for the Study of Existential Risk (CSER), University of Cambridge. His work focuses on mapping theories of change for long-term AI governance.