OpenAI, Alphabet, Meta Decide to Watermark AI-Generated Content material for Safer Tech, Assures White House

Top AI corporations together with OpenAI, Alphabet, and Meta Platforms have made voluntary commitments to the White House to implement measures resembling watermarking AI-generated content material to assist make the expertise safer, the Biden administration mentioned.

The corporations – which additionally embody Anthropic, Inflection, Amazon.com, and OpenAI associate Microsoft – pledged to completely check programs earlier than releasing them and share details about methods to scale back dangers and put money into cybersecurity.

The transfer is seen as a win for the Biden administration’s effort to control the expertise which has skilled a growth in funding and client reputation.

Since generative AI, which makes use of knowledge to create new content material like ChatGPT’s human-sounding prose, grew to become wildly well-liked this 12 months, lawmakers world wide started contemplating methods to mitigate the risks of the rising expertise to nationwide safety and the economic system.

US Senate Majority Chuck Schumer in June known as for “comprehensive legislation” to advance and guarantee safeguards on synthetic intelligence.

Congress is contemplating a invoice that will require political adverts to reveal whether or not AI was used to create imagery or different content material.

President Joe Biden, who’s internet hosting executives from the seven corporations on the White House on Friday, can be engaged on growing an government order and bipartisan laws on AI expertise.

As a part of the trouble, the seven corporations dedicated to growing a system to “watermark” all types of content material, from textual content, pictures, and audio, to movies generated by AI in order that customers will know when the expertise has been used.

This watermark, embedded within the content material in a technical method, presumably will make it simpler for customers to identify deep-fake pictures or audio that will, for instance, present violence that has not occurred, create a greater rip-off or distort a photograph of a politician to place the individual in an unflattering mild.

It is unclear how the watermark will likely be evident within the sharing of the knowledge.

The corporations additionally pledged to concentrate on defending customers’ privateness as AI develops and on guaranteeing that the expertise is freed from bias and never used to discriminate towards susceptible teams. Other commitments embody growing AI options to scientific issues like medical analysis and mitigating local weather change. 

© Thomson Reuters 2023