The Facebook parent has spent thousands of hours stress-testing new AI chatbots but “can’t guarantee” they won’t say “something unwelcome.”
Among Meta Platforms Inc.’s suite of new artificial intelligence products is a chatbot trained to talk like a detective but look like Paris Hilton and an armor-wearing character called the Dungeon Master who’s based on Snoop Dogg.
They’re at the amusing end of a serious push by Meta to incorporate generative AI tools that can respond to messages from users into apps used by billions of people every month — and it falls on Nick Clegg, Meta’s president of global affairs, to make sure his employer does so responsibly after years of criticism that the company does not adequately safeguard its products.
“We’ve had thousands of hours of red teaming — stress testing — to make them as safe as we possibly can,” Clegg said on the latest episode of the Bloomberg Originals series AI IRL. “We can’t guarantee they won’t spew something inappropriate or unwelcome at any point, but we can keep retraining it.”
Clegg said he’s involved “right from the outset” when decisions are being made by senior leadership about the direction to take the company’s AI tools “to anticipate some of the legal, regulatory, moral, ethical, cultural issues which these new products might bump up against.”
Meta has used AI under the hood of its consumer-facing products for years, particularly in determining which posts and ads to show users. It has also relied heavily on AI to help detect and combat the spread of misinformation. But now, AI isn’t just a solution to the company’s problems; it’s also a big potential headache. Meta must confront a new set of concerns posed by AI tools that may go wildly off script with a user or spread misinformation. Some users have reportedly found that Meta’s AI-generated stickers, announced alongside the celebrity chatbots last month, can create concerning imagery.
Clegg, a former member of the European Parliament and one-time UK deputy prime minister, was hired by Meta in 2018 to run its global lobbying efforts when the company was grappling with various crises around privacy issues and election meddling. Meta and others are now developing AI systems under intense scrutiny from governments, regulators, and ethicists.
So far, the US has only sought voluntary agreements from tech companies to safeguard their AI products. Clegg joined President Joe Biden at the White House in July to debut some of those agreed upon. Europe, meanwhile, is moving faster to regulate AI than the US. The European Parliament approved a draft law in June that would set boundaries on how AI technology can be used — and massive penalties for violations.
Clegg said Meta’s use of transparent, open-source licenses for some of its datasets and large-language models — the technology that underpins AI chatbots — would be key to safeguarding ethical use.
“The more open we can be about how these new technologies operate, the more open we can be about the risks and benefits, what the controls are for users, and how we need to work with others,” he said.
