Google DeepMind Veterans Launch Discovery Loop to Automate Scientific Discovery with AI
核心洞察
Four top AI researchers from Google DeepMind (搜索), including chief scientist Jeff Dean, have left to form Discovery Loop (搜索), a startup focused on automating the experimental loop across life sciences and other industries.
The company aims to build AI solutions that can automatically propose, run, and evaluate experiments at scale, targeting drug discovery, medicine, and health informatics.
Discovery Loop (搜索) is structured as a public benefit corporation and has secured initial funding co-led by Radical Ventures (搜索) and Khosla Ventures (搜索), with Google taking a stake and providing cloud compute.
A quartet of the most influential figures in artificial intelligence at Google DeepMind (搜索) has departed to launch Discovery Loop (搜索), a new startup with the ambitious goal of automating the scientific method itself. The founding team—Jeff Dean, Google's chief scientist; Sanjay Ghemawat; Oriol Vinyals; and Quoc Le—brings a combined working relationship spanning 14 to 30 years and a track record of building some of the world's most widely used AI infrastructure and models.
Discovery Loop (搜索), based in Palo Alto, California, is designed to "automate the experimental loop" across a broad spectrum of disciplines, including scientific discovery, engineering better medicines, and health informatics. In a social media post announcing the venture, Dean described the core approach: AI systems that can propose an experiment, run it, evaluate the results, and then repeat—thousands of times over—to accelerate breakthroughs in fields ranging from drug discovery to chip design and materials science.
"The four of us have worked together for 14 to 30 years, and have helped build some of the world's most used products, infrastructure and AI models," Dean stated. "We're excited to turn our attention to this ambitious endeavour."
The Lab-in-the-Loop Paradigm in Biopharma
The lab-in-the-loop approach that Discovery Loop (搜索) intends to pioneer is gaining significant traction in biopharmaceutical R&D. The methodology is being deployed to minimize the risk of clinical failure in new drug discovery projects by ensuring that therapeutic targets are robustly validated before any drug candidates directed against them enter clinical testing. This paradigm shift addresses one of the industry's most persistent challenges: the high attrition rate in drug development.
It remains unclear precisely how Discovery Loop (搜索)'s implementation will differentiate from other players in the category. The competitive landscape already includes technology giants such as NVIDIA, Amazon, and Google's own Isomorphic Labs (搜索), alongside specialized firms like Recursion Pharma (搜索), Protillion Bio, and Relation Therapeutics.
Corporate Structure and Investment
Discovery Loop (搜索) is structured as a public benefit corporation, a for-profit model that carries a legal obligation to balance financial returns for shareholders with a defined public or social benefit. The co-lead investors for the company's initial funding round are Radical Ventures (搜索) and Khosla Ventures (搜索). Notably, Google is also taking a stake in the startup and will provide cloud compute resources for its first year of operations—a signal that Alphabet CEO Sundar Pichai, who reportedly attempted to persuade the group to remain at Google, ultimately gave the exit his blessing.
"Jeff, Sanjay, Quoc and Oriol are starting by automating machine learning research, acting as their own first customer," Radical Ventures (搜索) noted in a blog post. "From there, they will direct these systems toward the physical world, targeting grand challenges in engineering, medicine, materials design, and clean energy."
Leadership Shifts at Google DeepMind (搜索)
The formation of Discovery Loop (搜索) coincides with a period of significant organizational change at Google DeepMind (搜索). Nobel Prize winner Demis Hassabis is relinquishing the chief executive role to become parent company Alphabet's chief scientist—a newly created position with responsibility for leading the company's artificial general intelligence (AGI) efforts—as well as DeepMind's chairman. Koray Kavukcuoglu, DeepMind's chief technology officer and Google's chief AI architect, is assuming operational responsibilities at the unit with the title of senior vice president.
The departures come amid suggestions that Google's AI push, built around its Gemini platform, is falling behind competitors such as Anthropic and OpenAI. Dean had been leading Gemini's development, which has experienced delays in launching its latest version originally scheduled for June. Analysts have suggested that the simultaneous departure of all four figures could have a significant impact on Google's AI trajectory.
Weeks before the official announcement, Dean previewed the concept behind Discovery Loop (搜索) to approximately 6,000 aspiring founders at Y Combinator's Startup School, describing the exact vision for automating scientific discovery without disclosing that it was already his own venture in development.