Summary of Key Points
Jeff Dean, a former senior AI executive at Google (where he worked for 27 years), along with three other prominent Google AI experts, have left to establish a new company called Discovery Loop. Their goal is to use AI to automate the core cycle of scientific discovery: formulating questions, conducting experiments, analyzing results, and iterating, rather than creating chatbots. Google attempted to retain them but failed; instead, it became an investor in the new company and provided cloud computing resources. In the initial stages, the company will use AI to improve its own algorithms and later expand into areas such as chips, biology, and pharmaceuticals. However, for now, AI still needs to collaborate with humans to generate novel ideas.
Two Key Reasons for Leaving to Start a Company
There are two main reasons why Jeff Dean and his colleagues decided to leave Google:
1. Greater Focus: Large companies like Google have many teams and diverse businesses, meaning executives often have to manage multiple tasks (such as attending meetings and coordinating departments), which prevents them from focusing solely on one thing. In contrast, small companies focus entirely on a single goal without distractions.
2. Cloud Computing Breaks the Power Monopoly: In the past, only giants like Google and Microsoft could afford the supercomputers required for training AI models. Now, with sufficient funding, small teams can access similar computing power by renting cloud services (such as Google Cloud or AWS). This has made it possible for smaller teams to achieve what was once only possible for larger organizations.
Business Direction: Avoiding Chatbots, Focusing on Accelerating Scientific Discovery
Many AI companies are developing chatbots like ChatGPT, but Discovery Loop is taking a more ambitious approach: using AI to optimize the entire scientific research process. The core of scientific research involves a cycle of formulating questions, conducting experiments, analyzing results, and forming new questions. However, many experimental processes (e.g., in biology) are time-consuming (taking months). Discovery Loop aims to automate this cycle:
- AI will help researchers design experiment plans.
- It will run multiple experiments simultaneously, much faster than humans can.
- It will automatically analyze the results and suggest the next steps for researchers.
- Most importantly, AI has the ability to integrate knowledge from different disciplines (biology, chemistry, materials science), which is difficult for a single expert to achieve, allowing for the discovery of new directions.
Jeff Dean’s Technical Approach
When asked how to identify technological trends, Jeff Dean shared a method: instead of reading a single paper in detail, he quickly reviews the abstracts of dozens or even hundreds of papers. This helps him create a “technical map” in his mind, showing what is being researched in various fields and what new breakthroughs have occurred. By connecting these scattered pieces of information, he can identify areas worthy of long-term investment. For example, although AI was previously used separately in biology and materials science, combining the two fields could lead to the development of new drugs or materials.
Practical Challenges
Although their goal is ambitious, they recognize that AI currently has limitations: it cannot generate truly novel and viable ideas on its own. While AI can analyze experimental data, it may not come up with innovative suggestions (e.g., using a certain material to treat cancer). Therefore, in the initial phase, humans and AI will work together to develop new ideas, with AI gradually learning to think independently. Their first step is to use AI to improve its own machine learning algorithms. Once this capability is matured, they plan to expand into areas such as chip design, drug discovery, and materials science.
Google’s Approach
Although Google wanted to retain Jeff Dean and his team, it ultimately became an investor in Discovery Loop, providing cloud computing resources. This indicates that Google recognizes the potential of using AI to accelerate scientific discovery. Instead of trying to stop them, it decided to participate and benefit from this innovative approach.
Conclusion
This news story highlights how top AI professionals are leaving large companies to pursue a new opportunity in automating scientific research. Their strengths lie in their technical expertise and access to cloud computing power. The challenge is to teach AI to think innovatively. Google’s investment shows that the industry sees this direction as promising, which could potentially change the speed and methods of scientific research in the future.