| Hi, it's Julia in San Francisco. A Nobel Prize raises some questions about the spirit of research at Google. But first... Three things you need to know today: • AMD unveils its new line of AI chips to challenge Nvidia
• Northvolt's battery-making boom fizzles out in Sweden
• Elon Musk showed off Tesla's new robotaxi called Cybercab Google's remarkable Nobel Prize haul this week raises a question: Is there a disconnect between shiny prizes awarded by sober-minded scientists and the scrappy day-to-day reality of corporate artificial intelligence research in the post-ChatGPT era? Demis Hassabis, Google's AI chief, and John Jumper, another researcher at the company, on Wednesday shared the award in chemistry for AI research into proteins, which has the potential to unlock breakthroughs in areas from drug development to the environment. A former Googler, Geoffrey Hinton, on Tuesday received the prize in physics for work that has become the foundation of modern AI. When my colleague Shirin Ghaffary and I spoke with Hassabis by Zoom, just hours after he received the good news, he was beaming. The famously competitive researcher started racking up an impressive array of trophies as a teenage chess champion. Now, he'd won arguably the biggest prize of all. The awards cemented AI's status as a pillar of science and a tool that can solve some of humanity's biggest problems, if deployed with care — a cause to which Hassabis has devoted his life. "I would have done this no matter what, even if it turned out to be 50 years, you know, too early," Hassabis said. "Because this was always my passion, to investigate intelligence." But Wall Street has also woken up to the opportunities presented by AI. And that rush of corporate investment and competitive pressure has changed the culture of AI research — especially at Google parent Alphabet Inc., where researchers once enjoyed the freedom to pursue passion projects with generous funding and only minimal oversight. The roots of AlphaFold, Hassabis and Jumper's pioneering work on protein folding, stem from that bygone era, when DeepMind was an independent sister company of Google. Over the past decade, Google researchers have forged some of the biggest breakthroughs in the field of AI. The "T" in ChatGPT, OpenAI's popular chatbot, is named for a technique invented in Google labs. Translating these advances into hit products of Google's own was the problem. To remedy the situation, Google last year merged its two premier AI research labs to form a new super unit, Google DeepMind, and fortified its ties with product teams. The lab is still working on applications of AI for climate science and medicine. Yet its primary focus is Gemini, Google's flagship AI model. It's a change for researchers, some of whom feel the mobilization around Gemini has left less room for experimentation, as we reported earlier this year. Some of Google's biggest contributions to the field came from small, self-directed teams. Google's famous paper in 2017 on machine learning, "Attention is all you need," had eight authors, each of whom has become a celebrity in the insular world of Silicon Valley. More than 1,200 contributors were listed on a paper for Gemini published last year. In the new world order, some people familiar with the company's operations told us they fear pure research is getting short shrift — and after the merger of Google Brain and DeepMind, some teams focused on scientific applications of AI were concerned their projects would be scrapped entirely. "If the DeepMind folks were starting today, would they have achieved the same groundbreaking scientific work?" said Oren Etzioni, an AI researcher who founded TrueMedia.org, a nonprofit combating political disinformation. "It's a very fair question and an undeniable risk that they don't do the same level of fundamental research that we've become accustomed to." Yet Hassabis sees no conflict between his mandate to win the AI race and his passion for science. The march toward artificial general intelligence, Hassabis said, means progress in the underlying technology can be applied to a host of domains. And to hear him tell it, what's good for business and what's good for society don't have to be in tension. "If we can revolutionize drug discovery, that will be enormously commercially valuable, as well as incredible for the world," he said. "So I try to do things that have dual purposes like that." Hassabis is known for working long hours, especially since the merger. (In our interview, he estimated he works 100 hours a week, spending 10%-15% of his time on AlphaFold and related projects.) In his public appearances, he often sports a uniform of a royal blue sweater and matching wire-rimmed glasses, as if seeking to preserve all his brain power for more important matters. Going forward, maybe Google's taste of Nobel stardom will give its researchers a little more time to chase the next prize.—Julia Love Ride-hailing companies Uber and Lyft have used a loophole to block drivers from accepting rides at certain periods of the day, even during high demand. The driver lockouts have cost individual drivers hundreds or thousands of dollars. Comcast's CEO isn't all that worried about the increasing numbers of "cord-cutters." Netflix orders a new live variety show with comedian John Mulaney. Disinformation stirred by Donald Trump and Elon Musk is making recovering efforts from hurricanes in the US more difficult. | 
No comments:
Post a Comment