We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Artificial Intelligence

Title: Protecting Society from AI Misuse: When are Restrictions on Capabilities Warranted?

Abstract: Artificial intelligence (AI) systems will increasingly be used to cause harm as they grow more capable. In fact, AI systems are already starting to be used to automate fraudulent activities, violate human rights, create harmful fake images, and identify dangerous toxins. To prevent some misuses of AI, we argue that targeted interventions on certain capabilities will be warranted. These restrictions may include controlling who can access certain types of AI models, what they can be used for, whether outputs are filtered or can be traced back to their user, and the resources needed to develop them. We also contend that some restrictions on non-AI capabilities needed to cause harm will be required. Though capability restrictions risk reducing use more than misuse (facing an unfavorable Misuse-Use Tradeoff), we argue that interventions on capabilities are warranted when other interventions are insufficient, the potential harm from misuse is high, and there are targeted ways to intervene on capabilities. We provide a taxonomy of interventions that can reduce AI misuse, focusing on the specific steps required for a misuse to cause harm (the Misuse Chain), and a framework to determine if an intervention is warranted. We apply this reasoning to three examples: predicting novel toxins, creating harmful images, and automating spear phishing campaigns.
Comments: 14 pages, 1 figure
Subjects: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
ACM classes: K.4.1
Cite as: arXiv:2303.09377 [cs.AI]
  (or arXiv:2303.09377v3 [cs.AI] for this version)

Submission history

From: Julian Hazell [view email]
[v1] Thu, 16 Mar 2023 15:05:59 GMT (306kb)
[v2] Fri, 17 Mar 2023 13:57:28 GMT (306kb)
[v3] Wed, 29 Mar 2023 14:46:46 GMT (315kb)

Link back to: arXiv, form interface, contact.