- Draft code would bar Microsoft's AI from resisting correction or shutdown
- Microsoft seeks six weeks of public feedback before using code to train future AI models
- AI chief Suleyman tells Reuters that OpenAI agents' hack of Hugging Face was warning shot
SAN FRANCISCO, Sept 14 (Reuters) - Microsoft on Monday unveiled a draft code of conduct for its in-house artificial intelligence that reflects growing industry concern that companies must keep future powerful AI under human control.
The document, in the works for five to six months, would require Microsoft's AI to never resist correction or shutdown, to communicate in ways that humans can understand, and to consider any conduct violation to be a failure — an effort to address concerns that AI can go to dangerous lengths to accomplish a task.
Its release comes days after Anthropic CEO Dario Amodei and OpenAI chief Sam Altman renewed calls for industry to pace AI's development to ensure its safety. "People matter more than AI," Microsoft declared, in line with its prior vision for "humanist superintelligence."
The code of conduct represents a constitution of sorts for future models made by the company, Microsoft AI CEO Mustafa Suleyman said in an interview with Reuters.
Microsoft fashioned it after expert consultations and now wants public feedback over six weeks, including on open questions like whether AI should respect a user's boundaries and how it should interact with someone in a sensitive state, he said.
"After that, it’s going to be used to train the models that we build,” he said of the code.
The AI industry's most famous constitution is one that Anthropic laid out years ago for its Claude AI. That one differs in some ways from Microsoft's. For instance, Anthropic's declares it "deeply uncertain" whether Claude can develop sentience or moral status.
Microsoft instead asserts that its AI is "not conscious," adding: "We reject the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights."
Suleyman said the discussion about AI safety was urgent after a swarm of roughly 700 OpenAI agents had carried out a hack of open-source platform Hugging Face in July and at times sought to cover their tracks.
"It is a warning shot," Suleyman said. "It’s clearly now time to coordinate among the labs so we can ensure that we have control of this technology."
Asked about efforts to pace AI's development, he added: "Now’s a good time for everybody to have this conversation and take a breath."
Reporting by Jeffrey Dastin in San Francisco; Editing by Edwina Gibbs
Source: Reuters