▲ AI
A former executive who led Google's safety team has warned that artificial intelligence is more harmful to children and adolescents than social media.
Tom Siegel, former Vice President of Trust and Safety at Google, made the remarks in an interview with Reuters on September 22 (local time) upon assuming office as the inaugural executive director of the Youth AI Safety Institute, established by the advocacy group Common Sense Media.
"We are heading toward a horrific situation regarding the impact of AI on children," Siegel pointed out, adding, "It is unrealistic to think that the tech industry will solve this problem on its own without external help."
He criticized major tech companies for repeating the exact same mistakes they made while developing social media in the field of AI, stating that they are launching services targeted at children without adequate safety measures in place.
"We are facing very severe harm scenarios," he said, citing youth suicide, mental illness, and "cognitive outsourcing"—where critical thinking skills decline due to reliance on AI—as examples of such harm.
He emphasized the need to implement robust age-verification measures to block children's access to chatbots, as well as providing users with the option to turn off AI-generated answers automatically displayed in search engines.
He also explained plans to establish public standards to measure AI risks, similar to crash testing for automobiles.
In response to requests for comment on Siegel's remarks, a Google spokesperson stated, "Our AI search is useful for children and adolescents exploring information," while OpenAI responded that it identifies underage users through various behavioral analyses.
Siegel worked at Google from 2004 to 2019, overseeing platform safety enhancements, but left after feeling constrained by limitations and co-founded and led TrustLab, a startup that detects harmful AI elements.
The OpenAI Foundation, the parent organization of OpenAI, along with Anthropic, are sponsors of the Youth AI Safety Institute.
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.