AI companies agree to voluntary safety measures

Artificial intelligence companies on Tuesday agreed to a set of voluntary safety guidelines, reaffirming their commitment to developing AI technology in a way that encourages the safety of their platforms and the public.

At the AI ​​Seoul Summit, hosted by the South Korean and UK governments, 16 leading companies signed the non-binding Frontier AI Safety Commitment, which aims to reduce AI risks while increasing accountability and transparency.

Companies that have agreed to this initiative include OpenAI, Google, IBM, Amazon, Meta, and xAI.

The companies agreed to assess technology risks at every stage of new AI system development and focus on mitigating risks associated with potential misuse of the technology. It was agreed that the results of internal and external assessments would be taken into account “as appropriate” to determine the safety of the system. External evaluations may be conducted by an independent third party, the home government, or “any other body that the government deems appropriate.”

The companies also agreed to set an acceptable, predetermined threshold for the maximum level of risk posed by a particular AI system. They agreed to clearly state how to ensure risk is kept below that threshold and what to do if the AI ​​system exceeds that threshold. This “includes a process to further develop and deploy systems and models only if we assess that residual risk remains” below the threshold. ”

“In extreme cases, the organization commits not to develop or deploy any models or systems where mitigation measures cannot be applied to reduce the risk below a threshold,” the non-binding agreement states in part. There is.

The two-day summit, which followed the first such meeting in November, saw several world leaders similarly build appropriate guardrails to develop AI safely and create a “common understanding of AI safety.” They pledged to share resources to “build AI research efforts and coordinate AI research efforts.” ”