Researchers find widespread weaknesses in AI safety, performance tests: Report

LONDON

Experts have discovered flaws in hundreds of tests that check AI safety and effectiveness, which could "undermine the validity of resulting claims."

Researchers have found weaknesses, some serious, in many tests used to check the safety and effectiveness of new artificial intelligence models being released into the world, according to a report by The Guardian on Tuesday.

Computer scientists from the British government’s AI Security Institute, and experts at universities, including Berkeley and Oxford, examined more than 440 benchmarks that provide an important safety net as part of a new study.

They found flaws that "undermine the validity of the resulting claims," that "almost all … have weaknesses in at least one area," and resulting scores might be "irrelevant or even misleading."

Andrew Bean, a researcher at the Oxford Internet Institute, said that many of the benchmarks are used to evaluate the latest AI models released by the big technology companies.

The study came amid mounting concern over the safety and effectiveness of AI models, which are being released at a high pace by competing technology companies.

Google recently withdrew one of its latest AIs, Gemma, after it made up unfounded allegations about a US senator having a non-consensual sexual relationship with a state trooper, including fake links to news stories.

"There has never been such an accusation, there is no such individual, and there are no such new stories," Marsha Blackburn, a Republican senator from Tennessee, said in a letter to Sundar Pichai, Google’s chief executive.

"A publicly accessible tool that invents false criminal allegations about a sitting US senator represents a catastrophic failure of oversight and ethical responsibility," she said.

Google said its Gemma models were built for AI developers and researchers, not for factual assistance or for consumers.