Scientists Battling a Robot
On February 19 Google announced their AI co-scientist platform that starts with a research goal and then utilizes seven AI agents:
- a generation agent,
- reflection agent,
- ranking agent,
- proximity agent,
- meta-review agent,
- evolution agent and a
- supervisor agent overseeing the whole effort.
In short, every scientist now has a staff of seven digital workers to help them with everything from ideation to discovery. It’s not just the additional labor that is exciting, but they have incorporated a “self-play” approach in which different ideas are debated among the agents. It evolves its ideas as it plays across different options and examines their rigor and facts available to the hypothesis.
Excellent Early Performance by AI Agents
When the ideas of the AI-co scientist was reviewed by real scientists in their own field, it was rated almost equally on the potential impact of their ideas as well as strong novelty. For example, it come up with novel repurposing drug candidates for acute myeloid leukemia (AML) which were then clincally tested and proved to be clinically relevant. It also had strong results concerning mechanims of antimicrobial resistance, and liver fibrosis causes.
Big Implications for Many Businesses
For many businesses an improvement in science has tremendous implications. For example, one of the critical things that SpaceX did was to push the boundaries of methods of aluminum welding so they could decrease the cost of building a rocket fuselage. SpaceX’s method needed fewer riveted panels, but still had high strength and low weight – all at a lower cost. Every manufacturer often benefits from a scientific breakthrough.
Likewise science-centric firms like pharmaceutical firms, or energy companies, or computing organizations can benefit greatly by faster, more accurate, and perhaps less expensive science. I’m reminded of an old quote told to me by the late Gordon Bell talking about something Bob Noyce, the founder of Intel once said at a board meeting: “Remember the scientists create all the value, you guys in marketing and finance just push it around.”
The other important thing to remember is that these large language models are build upon matrix math which allows very different types of data – images, words, sensor readings, etc., to all live in a shared knowledge space while it is being analyzed. I am optimistic that this cross data, cross context analysis will yield new insight and dramatic findings. We are only at the beginning.
Here are three practical actions a leader should take given Google’s new AI Co-Scientist tool:
1. Establish an AI-Augmented R&D Strategy
🔹 Why? The AI Co-Scientist accelerates hypothesis generation and research planning, reducing time to scientific breakthroughs.
🔹 Action:
• Create an internal AI research lab or partner with universities to pilot the AI Co-Scientist.
• Leverage AI to expand into adjacent scientific fields where your company has expertise but lacks depth.
2. Create an AI-Scientist-In-The-Loop Culture
🔹 Why? The AI Co-Scientist does not replace human expertise but enhances it. Companies that integrate AI into their research teams will outpace competitors.
🔹 Action:
• Train scientists and engineers to work with AI as collaborators, not just tools.
• Develop “AI-first” research workflows where teams validate and refine AI-generated hypotheses before committing resources.
• Encourage an experimental mindset, allowing AI to challenge traditional R&D assumptions and explore unorthodox solutions.
3. Hire Scientific Talent Who Work with and/or Build New Models
🔹 Why? Scientists who learn how to use and build these tools they will create value for all their colleagues.
🔹 Action:
• Work with academic institutions who train scientists and engineers to work with AI.
• Develop internal learning labs to share and grow internal expertise.
• Give small grants to academic institutions who specialize in those areas of science that interest your firm.
The Speed of Science is Increasing: Don’t Lose the Race
Remember that all your assumptions about the speed of scientific discovery may be true impediments to your discovery process. Agents and Large Language models change the generation and liquidity of knowledge radically. Most importantly, remember that the Silicon-based scientists from Google never tire, never sleep, and can always be exploring and improving – every hour of every day of every year.