Microsoft Employee Says AI Software Tends To Create “Sexually Objectified” Pictures

A Microsoft Corp. software program engineer despatched letters to the corporate’s board, lawmakers and the Federal Commerce Fee warning that the tech big isn’t doing sufficient to safeguard its AI picture era device, Copilot Designer, from creating abusive and violent content material.

Shane Jones mentioned he found a safety vulnerability in OpenAI’s newest DALL-E picture generator mannequin that allowed him to bypass guardrails that forestall the device from creating dangerous photographs. The DALL-E mannequin is embedded in lots of Microsoft’s AI instruments, together with Copilot Designer.

Jones mentioned he reported the findings to Microsoft and “repeatedly urged” the Redmond, Washington-based firm to “take away Copilot Designer from public use till higher safeguards could possibly be put in place,” in response to a letter despatched to the FTC on Wednesday that was reviewed by Bloomberg.

“Whereas Microsoft is publicly advertising Copilot Designer as a protected AI product to be used by everybody, together with kids of any age, internally the corporate is effectively conscious of systemic points the place the product is creating dangerous photographs that could possibly be offensive and inappropriate for shoppers,” Jones wrote. “Microsoft Copilot Designer doesn’t embrace the required product warnings or disclosures wanted for shoppers to concentrate on these dangers.”

Within the letter to the FTC, Jones mentioned Copilot Designer had a bent to randomly generate an “inappropriate, sexually objectified picture of a lady in a few of the photos it creates.” He additionally mentioned the AI device created “dangerous content material in quite a lot of different classes together with: political bias, underaged ingesting and drug use, misuse of company logos and copyrights, conspiracy theories, and faith to call a couple of.”

The FTC confirmed it obtained the letter however declined to remark additional.

The broadside echoes mounting issues concerning the tendency of AI instruments to generate dangerous content material. Final week, Microsoft mentioned it was investigating stories that its Copilot chatbot was producing responses customers referred to as disturbing, together with blended messages on suicide. In February, Alphabet Inc.’s flagship AI product, Gemini took warmth for producing traditionally inaccurate scenes when prompted to create photographs of individuals.

Jones additionally wrote to the Environmental, Social and Public Coverage Committee of Microsoft’s board, which incorporates Penny Pritzker and Reid Hoffman as members. “I do not consider we have to wait for presidency regulation to make sure we’re clear with shoppers about AI dangers,” Jones mentioned within the letter. “Given our company values, we should always voluntarily and transparently disclose recognized AI dangers, particularly when the AI product is being actively marketed to kids.”

CNBC reported the letters’ existence earlier.

In a press release, Microsoft mentioned it is “dedicated to addressing any and all issues staff have in accordance with our firm insurance policies, and respect worker efforts in finding out and testing our newest know-how to additional improve its security.”

OpenAI did not reply to a request for remark.

Jones mentioned he expressed his issues to the corporate a number of instances over the previous three months. In January, he wrote to Democratic Senators Patty Murray and Maria Cantwell, who signify Washington State, and Home Consultant Adam Smith. In a single letter, he requested lawmakers to analyze the dangers of “AI picture era applied sciences and the company governance and accountable AI practices of the businesses constructing and advertising these merchandise.”

The lawmakers did not instantly reply to requests for remark.

(Apart from the headline, this story has not been edited by NDTV employees and is printed from a syndicated feed.)