DistantNews
Support us
๐Ÿ‡ฐ๐Ÿ‡ท South Korea /Technology

'Google Legend' Jeff Dean Embarks on New Challenge: 'Our Goal is the Automation of Science'

From Hankyoreh · () Korean

Translated from Korean, summarized and contextualized by DistantNews.

At a glance

News Sources not specified Context piece
  • Google AI legend Jeff Dean is leaving Google to co-found a new company, 'Discovery Loop.'
  • The new venture aims to automate scientific discovery and accelerate research in science and engineering.
  • Alphabet, Google's parent company, will be an early investor and cloud partner for Discovery Loop.

Jeff Dean, a towering figure in artificial intelligence and a legendary programmer at Google, is embarking on a new venture. Just a week after announcing his departure from the tech giant after 27 years, Dean revealed his plans to co-found a new company called 'Discovery Loop.' The company's ambitious goal is to automate machine learning and the processes of science and engineering, thereby accelerating discovery and disseminating its benefits more widely.

Dean, who joined Google in 1999, the year after its founding, has been instrumental in developing many of the company's core AI technologies. He made the announcement at the 'International Conference on Knowledge Discovery and Data Mining' (KDD) in Jeju, South Korea, where he delivered a keynote speech on the trends in AI. His departure was not a complete break from his former employer; Alphabet, Google's parent company, will serve as an early investor and cloud partner for Discovery Loop.

Our goal is to automate machine learning and science and engineering, to discover more, faster, and to spread the benefits of that discovery widely.

โ€” Jeff DeanJeff Dean outlines the mission of his new company, Discovery Loop.

In his keynote, Dean outlined the trajectory of AI, from computers that could not 'see or understand speech' to systems capable of 'performing science autonomously.' He compressed the last 15 years of progress into a narrative arc, touching upon the training of large neural networks, the development of Google's Tensor Processing Units (TPUs), knowledge distillation, transformer models, and advanced reinforcement learning. He humorously recounted his transition, stating, "I announced my retirement at a video conference at 8:30 AM on Wednesday, had a farewell party with 1,500 Google employees at 2:30 PM, played soccer at 3:30 PM, and returned to find 400 messages. I became 'unemployed' at 11:59 PM Thursday and a co-founder of Discovery Loop one minute later."

I announced my retirement at a video conference at 8:30 AM on Wednesday, had a farewell party with 1,500 Google employees at 2:30 PM, played soccer at 3:30 PM, and returned to find 400 messages. I became 'unemployed' at 11:59 PM Thursday and a co-founder of Discovery Loop one minute later.

โ€” Jeff DeanJeff Dean humorously describes his rapid transition from Google employee to co-founder of Discovery Loop.

Dean elaborated on the concept of 'automating science,' explaining it as the automation of the experimental loop, which typically involves hypothesis, implementation, execution, and evaluation. By automating these steps, Discovery Loop aims to significantly speed up the pace of innovation. He highlighted the inefficiencies of traditional scientific experimentation, which are often labor-intensive and slow. Large-scale experiments require breaking down tasks for smaller teams, leading to coordination costs and a lack of flexibility. Discovery Loop's approach seeks to streamline these processes, enabling the development of higher-quality models focused purely on scientific and engineering principles.

Furthermore, Dean emphasized the company's commitment to societal contribution. "We strongly believe that technology must contribute to society, and we must make it work that way," he stated, announcing that Discovery Loop is established as a public benefit corporation. While profit-driven, the company will also uphold a sense of public mission, focusing on developing technologies that address significant societal challenges and ensuring accessibility to their outcomes. Responding to a question about AI outpacing human oversight in automated experiments, Dean acknowledged it as a challenge, suggesting that humans would likely focus on selecting the most promising experimental results rather than scrutinizing every outcome.

We strongly believe that technology must contribute to society, and we must make it work that way.

โ€” Jeff DeanJeff Dean discusses the public benefit mission of Discovery Loop.
DistantNews Editorial

Originally published by Hankyoreh in Korean. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.