The Association for Computing Machinery today urged the providers of generative AI systems and developers using them to adopt safeguards for the technology, releasing a statement outlining a list of principles for its use and development.

Generative AI, according to the ACM, poses serious risks to society, and the group said that current guidelines for the safe use of technology are insufficient to ensure that generative AI will not cause considerable harm.

Today’s statement is based on a longer document on principles for the development, deployment, and use of generative AI technologies that the ACM issued in June. The ACM was founded in 1947 and claims to be the world’s largest scientific and educational computing society, with about 110,000 student and professional members.

In its statement about artifical intelligence, the group said that legal and regulatory schemes should be reviewed to take into account the particular dangers posed by generative AI misuse, urging that no “high-risk” AI should be operated without substantial human oversight and careful deliberation over whether the benefits of such systems outweigh their potential negative impacts.

ACM says intellectual property rules should be reviewed

Intellectual property ownership rules should also be reviewed, according to the ACM, in order to help settle major questions over the ownership of AI-created works and the legal status of data used to train AI models. Personal data used to generate information or train models should be subject to opt-out polices, and AI creators should maintain records of errors made by their systems, in an effort to be as transparent as possible about accuracy and correctability.

Today’s declaration stopped short of urging the adoption of specific legislation, however. Jeremy Epstein, the chair of the ACM’s US technology policy committee, said that the group is non-partisan and is not a lobbying organization.

“We do respond to legislative offices’ request for input directly, but don’t take public stands on bills or components of them,” he said.

Nor does the ACM take a particular stance on the issue of copyright law as applied to AI training data, which is the subject of several class action lawsuits filed in recent weeks against AI creators like Meta and OpenAI. Beyond a general call for careful consideration of the issue, Epstein said that the matter is likely to be resolved through litigation.

“The courts are going to have to determine this in the end, as legislating on changes to Section 107 of the Copyright Act has historically been, and is likely to remain, very difficult,” he said.

The ACM’s statement went on to reaffirm more general guiding principles for the development of technology, including auditability and transparency, ensuring a minimal impact on the environment, and strict attention to privacy and security issues that may arise.