Comment: Warnings from AI’s ‘godfather’ should have come sooner

By Parmy Olson / Bloomberg Opinion

It is hard not to be worried when the so-called godfather of artificial intelligence, Geoffrey Hinton, says he is leaving Google and regrets his life’s work.

Hinton, who made a critical contribution to AI research in the 1970s with his work on neural networks, told several news outlets this week that large technology companies were moving too fast on deploying AI to the public. Part of the problem was that AI was achieving humanlike capabilities more quickly than experts had forecast. “That’s scary,” he told the New York Times.

Hinton’s concerns certainly make sense, but they would have been more effective if they had come several years earlier, when other researchers who didn’t have retirement to fall back on were ringing the same alarm bells.

Tellingly, Hinton in a tweet sought to clarify how the New York Times characterized his motivations, worried that the article suggested he had left Google to criticize it. “Actually, I left so that I could talk about the dangers of AI without considering how this impacts Google,” he said. “Google has acted very responsibly.”

While Hinton’s prominence in the field might have insulated him from blowback, the episode highlights a chronic problem in AI research: Large technology companies have such a stranglehold on AI research that many of their scientists are afraid of airing their concerns for fear of harming their career prospects.

You can understand why. Meredith Whittaker, a former research manager at Google, had to spend thousands of dollars on lawyers in 2018 after she helped organize the walkout of 20,000 Google employees over the company’s contracts with the U.S. Department of Defense. “It’s really, really scary to go up against Google,” she tells me. Whittaker, who is now CEO of encrypted messaging app Signal, eventually resigned from the search giant with a public warning about the company’s direction.

Two years later, Google AI researchers Timnit Gebru and Margaret Mitchell were fired from the tech giant after they released a research paper that highlighted the risks of large language models, the technology currently at the center of concerns over chatbots and generative AI. They pointed to issues like racial and gender biases, inscrutability and environmental cost.

Whittaker rankles at the fact that Hinton is now the subject of glowing portraits about his contributions to AI after others took much greater risks to stand up for what they believed while they were still employed at Google. “People with much less power and more marginalized positions were taking real personal risks to name the issues with AI and of corporations controlling AI,” she says.

Why didn’t Hinton speak up earlier? The scientist declined to respond to questions. But he appears to have been concerned about AI for some time, including in the years his colleagues were agitating for a more cautious approach to the technology. A 2015 New Yorker article describes him talking to another AI researcher at a conference about how politicians could use AI to terrorize people. When asked why he was still doing the research, Hinton replied: “I could give you the usual arguments, but the truth is that the prospect of discovery is too sweet.” It was a deliberate echo of J. Robert Oppenheimer’s famous description of the “technically sweet” appeal of working on the atomic bomb.

Hinton says that Google has acted “very responsibly” in its deployment of AI. But that’s only partly true. Yes, the company did shut down its facial recognition business on concerns of misuse, and it did keep its powerful language model LaMDA under wraps for two years to work on making it safer and less biased.

(In its blog post announcing research into conversational AI models that morphed into LaMDA, Google said it wasn’t releasing a research demo of the technology to tackle “safety and bias in the models.”) Google has also restricted the capabilities of Bard, its competitor to ChatGPT.

But being responsible also means being transparent and accountable, and Google’s history of stifling internal concerns about its technology don’t inspire confidence.

Hinton’s departure and warnings hopefully will inspire other researchers at large tech companies to speak up about their concerns.

Technology conglomerates have swallowed up some of the brightest minds in academia thanks to the lure of high salaries, generous benefits and the huge computing power used to train and experiment on ever-more-powerful AI models.

Yet there are signs some researchers are at least considering being more vocal. “I often think about when I would quit [AI start-up] Anthropic or leave AI entirely,” tweeted Catherine Olsson, a technical staff member at AI safety company Anthropic on Monday, in response to Hinton’s comments. “I can already tell this move will influence me.”

Many AI researchers seem to have a fatalistic acceptance that little can be done to stem the tide of generative AI, now that it has been unleashed to the world. As Anthropic co-founder Jared Kaplan told me in an interview published Tuesday, “the cat is out of the bag.”

But if today’s researchers are willing to speak up now, while it matters, and not right before they retire, we are all likely to benefit.

Parmy Olson is a Bloomberg Opinion columnist covering technology. A former reporter for the Wall Street Journal and Forbes, she is author of “We Are Anonymous.”