AI optimists crowd out doubters at TED

Driving the news: Industry giants painted bold pictures of utopian breakthroughs.

• Google DeepMind CEO Demis Hassabis suggested that while today's science sees only a few limbs of the tree of knowledge, AI could allow us to see "the entire tree" one day.
• Mustafa Suleyman, Hassabis' co-founder at DeepMind and now a top Microsoft executive, went further, contending that we should think of AI as "a new kind of digital species."
• Investor Vinod Khosla, meanwhile, promised AI would deliver free medical advice and education to all.

These discussions were thin on plans for how the industry might overcome the technology's many current limitations, including inaccuracy and bias. But they did raise one alarming scenario in which unlimited competition among both companies and countries could lead to dangerous, even deadly outcomes.

• The risk, known as a "Moloch trap," is that individuals or organizations will take actions that make sense for themselves but end up harming society. (Poker champion Liv Boeree traced this dilemma in a TED AI talk last year.)
• Even Hassabis acknowledged that competitive pressures brought on by the release of ChatGPT pushed Google to release generative AI tools earlier than it otherwise would have.
• He argued that worked out OK: "It turns out that even with those flaws, many tens of millions of people still find them very useful," he said. "That's pretty exciting in many ways."

Yes, but: Hassabis also said he thinks AI builders will need to shift from competition to collaboration as they get closer to artificial general intelligence (AGI) to make sure the technology has safe foundations.

• "I see a sort of kind of bottleneck that we have to get humanity through, which is building safe architectures as the first types of AGI systems," Hassabis said. From there, he said, more robust competition could resume, built off those consensus systems.
• He did not specify how such cooperation would work or what process would make it happen, though he hinted at a bigger role for regulation.

The other side: A few more sober voices came from outside of the companies developing the technology.

• Researcher Helen Toner, who left OpenAI's board after the failed ouster of Sam Altman, called for AI companies, especially those at the leading edge, to open themselves to outside audits and disclose their ongoing work.
• Rumman Chowdhury, the AI ethics specialist who conducted the adversarial "red team" testing at last year's DEF CON, argued for something akin to a "right to repair" for AI, allowing people to find and fix the flaws in systems.

Zoom out: AI's influence extended beyond the talks themselves.

• There was "Call Me Ishmael," an AI-powered recreation of a 1950s pay phone that let attendees call various literary characters. (My conversation with Winnie-the-Pooh was underwhelming.)
• Inworld showcased how AI can be used to make entertainment more interactive, while also providing for consistent characters and storytelling.
• TED itself is using generative AI to translate its talks into more languages. Organizers showed a video of a TED speaker's talk seamlessly switching from English to French to German to Italian, all while maintaining the speaker's voice and facial expressions.
• The Misalignment Museum used whimsical gadgets to showcase the darker side of AI, including a "spam bot" made out of a real can of Spam.

Flashback: It's the second year in a row that AI has dominated the discussion at TED.