Investing in the exponential: AI in Sciences

The future is here: AI in Biology, Chemistry and Physics.

Image generated by DALLE

Artificial Intelligence continues to penetrate various aspects of of our lives — from automating routine tasks to developing treatments for previously uncurable diseases and discovering breakthrough materials for a cleaner future.

As an AI investor, I’m most excited by founders leveraging the latest breakthroughs in AI to tackle hard, non-obvious problems with a transformative impact. The application of generative AI toward scientific discovery across biology, chemistry, and physics represents precisely such an opportunity.

In 2023, Microsoft Research’s paper The Impact of Large Language Models on Scientific Discovery: a Preliminary Study using GPT-4 examined GPT-4’s capabilities in scientific research and evaluated the model’s proficiency in understanding concepts and performing tasks in these domains. It also provided a framework for what AI for Scientific Discovery comprises, including but not limited to drug discovery, biology, computational chemistry, materials design, and partial differential equations (PDEs).

Source: Microsoft (Microsoft Research AI4Science and Microsoft Azure Quantum) Report on The Impact of Large Language Models on Scientific Discovery: a Preliminary Study using GPT-4.

The field has advanced rapidly since, and AI-accelerated scientific discovery has entered the mainstream. In 2024, Demis Hassabis and John Jumper won the Nobel Prize in Chemistry for their groundbreaking work on protein structure prediction. Radical portfolio company Nabla Bio announced JAM, their generative model for antibody design, demonstrating clear evidence of AI’s ability to create novel antibodies with drug-like properties. Another Radical portfolio company, Orbital Materials, released Orb, the fastest and most accurate AI model for simulating advanced materials, beating models from Google and Microsoft in accuracy and speed. We’ve also seen progress toward AI transforming the scientific method itself — Sakana AI published their work on an AI Scientist automating hypothesis generation, experiment design, and data analysis, while Stanford research showed how LLMs aid researchers by suggesting novel ideas and acting as creative collaborators in generating research directions.

Three reasons why I see breakout potential in this space

Scarce Talent: Top talent equally conversant in AI and scientific disciplines is rare and creates inherent value. Teams with this type of talent are able to productize and commercialize innovations within accelerated timeframes. I believe technically differentiated teams prioritizing domain expertise will build generational companies in this space, and today, many of these teams are leaving leading big tech labs, research institutes, and industry to launch companies in this space. Notable examples of best in class teams building companies in this category are Orbital Materials, founded by DeepMind alumnus Jonathan Godwin; Nabla Bio, founded by Surge Biswas, author of the 2019 UniRep paper that pioneered the application of language models to protein engineering and Frances Anastassacos, Ph.D. in Biological and Biomedical Sciences from Harvard; and Latent Labs, founded by Simon Kohl who led AlphaFold’s protein design team. 

Increased Industry Appetite: Industry appetite for AI solutions is maturing, particularly in life sciences, which has been a pioneer in adopting AI solutions. Major pharmaceutical companies have existing discovery and joint development arrangements to work with AI companies, enabling testing and scaling of the best AI solutions. Examples of these agreements include Genesis Thereapuetic’s $670M collaboration with Eli Lilly and Aspect Biosystem’s $2.6B deal with Novo Nordisk. Energy and industrial sectors are following suit, as evidenced by Occidental Petroleum’s purchase of carbon capture company Carbon Engineering for $1.1B.

Exponential Value Creation: I believe that AI applied to the sciences is poised to create some of the largest value categories in the future. Schmidt Futures pegged the bio-economy between $4T to $30T globally. These are trillion-dollar industries waiting to be created, and the work has already begun. The foundation is set - leading labs like Google DeepMind have paved the way with breakthroughs like AlphaFold and GNoME.

The timing is optimal for building these companies despite their complexity, capital requirements, and risks. After all, what better challenge than combining human and artificial intelligence to solve previously intractable problems?

I’ll close with a quote from Radical portfolio company founder Charles Fisher (CEO of Unlearn). He says:

“Society will get a bigger lift from creating AI to solve problems human beings can’t. That is the domain of narrow superhuman intelligence.”

“While there’s a lot of AI research on automating things humans do pretty well, there’s much more to be gained by using AI to solve unsolved problems. I think we’ll soon find that some of the problems we thought were so complex that they are impossible aren’t that difficult for a narrow superhuman intelligence.”

If you are building or working at the intersection of AI and Sciences, please reach out to connect, provide feedback, brainstorm, and collaborate.

Resources for additional reading:

• Google DeepMind’s podcast covers much of the organization’s contribution to accelerated scientific discovery.

• Microsoft Research Report on The Impact of Large Language Models on Scientific Discovery: a Preliminary Study using GPT-4.

• University of Toronto’s Acceleration Consortium.

• AI transforming the scientific method: Sakana AI & Stanford research.