Demis Hassabis, CEO of Google Deepmind, has a clear warning as artificial intelligence wipes out everyday life. Don't let social media fall into the same trap that made them toxic.
Speaking at the Athenian Innovation Summit with Greek Prime Minister Kiliakos Mitotakis last Friday, Hassavis said AI is one of the most transformative technologies in history, but its development calls for a much more cautious approach than what he described as the old “old” in Silicon Valley.
“We should learn lessons from social media. This attitude led to the 'moving fast, breaking things' attitude ahead of our understanding of what the resulting secondary and tertiary effects would be. ” Hassavis said.
He warned that AI was primarily designed to maximize user engagement (as social media has done for over a decade) and there is a risk of amplifying the same issues, from attention to mental health issues.
Social media algorithms “will keep an eye on your attention more and more, but not necessarily in a way that is beneficial to you as an individual,” he said.
Instead, Hassavis urged regulators and engineers to adopt a scientific method approach. This is a thorough test and understanding of the system before unleashing billions of people.
He argued that AI should be built as a tool to serve people rather than manipulate them.
Hassavis said the goal is to strike the right balance – “being bold on opportunity, but taking responsibility for reducing risk” is a “continuous” tension that he believes will “continue to AGI.”
Related Stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
AI already shows some of the same cracks
AI researchers have already seen AI replicate some of the toxic patterns of social media.
In a study published in August, researchers from the University of Amsterdam provided their own stripped social networks to 500 chatbots, splitting into creeks, raising extreme voices, and seeing small groups dominate the conversation without advertising or recommended algorithms.
The team tested six interventions to break the cycle, from hiding follower counts from the time series feed, but nothing worked. They concluded that dysfunction works deeper than algorithms and is burned into the way social platforms reward emotionally charged sharing.
Meanwhile, AI is embedded in social media, with virtual influencers becoming mainstream, warning that brands could test AI's faces and voices, and that some creators could undermine their careers by licensing their likenesses forever.
Openai CEO Sam Altman argues that addictive social media feeds can be more harmful to children than AI itself, but Reddit co-founder Alexis Ohanian suggests that AI can provide more control over what users are seeing online.
