An OpenAI whistleblower says that AI companies are “gambling with people’s minds” by creating social media-style models that encourage users to keep using their apps.
Zoë Hitzig resigned from OpenAI last month, saying its decision to run ads on ChatGPT created the risk that it would eventually “manipulate” its users by selling them ads based on “their most private thoughts”.
In an interview for The Observer’s New Review, Hitzig said she had joined the company to help shape AI tools at a moment when “the important decisions” were taken. After two years as a researcher, she could no longer justify her choice because she believed OpenAI was making the same mistakes that social media firms such as Facebook had made.
Last November, OpenAI was hit with seven lawsuits relating to people who had become psychotic or died by suicide after engaging with GPT-4o, which was criticised for its seductive and flirty writing style.
“I knew we’d make new mistakes with AI because it’s not the same as social media,” Hitzig said. “But there are some problems we could avoid, like let’s not maximise for engagement using psychometric profiles, because you know what that leads to: eating disorders, political manipulation, fake news and violence on the basis of fake news.
“Optimising for engagement before you understand the social and psychological consequences is really dangerous. It’s gambling with people’s minds.”
Hitzig, a poet and economist, said AI executives liked talking about long-term fears, such as AIs creating mass unemployment or biological weapons since they “feel far away, because they’re distracting us”. But she said there were more urgent and shocking events such as suicide and psychosis.
More from The Observer
Newsletters
Choose the newsletters you want to receive
View more
For information about how The Observer protects your data, read our Privacy Policy