Biology and AI: Twin Engines Built for鈥�

While both AI and biology have made significant leaps in recent decades, the full potential of both fields is yet to be realized.

Artificial intelligence (AI) has long been closely tied to biology. In fact, the history of AI traces back to the 1950s when scientists from multiple fields began to form theories on how to create an artificial brain. In 1958, Frank Rosenblatt, a psychologist and neurobiologist at Cornell University, invented the perceptron, the first artificial neural network (ANN) that was built using biological principles. Since then, there has been a series of advances in AI research, leading to significant innovations, especially in image, speech and text recognition. Today, AI is a term used in almost every industry to denote the generation of complex insights that match and even surpass the abilities of the human brain.

While both AI and biology have made significant leaps in recent decades, the full potential of both fields is yet to be realized. We believe AI and biology are the principal and mutually enabling innovation engines of our generation, particularly so when combining forces. Applying machine learning to large biological datasets has begun to elucidate the cause-and-effect relationships underlying human genetics and disease. In the opposite direction, we have only scratched the surface in terms of leveraging biology to advance AI research. Biomimicry鈥攖he practice of learning from and imitating the strategies found in nature鈥攈as always been a source of inspiration for AI researchers. After all, humans are the most intelligent beings we know (thus far) and it鈥檚 only natural for us to try and match and surpass our own level of intelligence. Today鈥檚 AI models are nowhere near achieving strong AI 鈥� defined as fully autonomous digital beings that can perform any intellectual task that the human brain can 鈥� but the progress in this direction is astonishing.

Moving forward, the exponential increase in biological data generation, computing power and AI model performance will accelerate the pace of discovery and deepen our understanding of biological systems.

Leveraging AI for Biology

In recent years, AI has enabled several breakthroughs in biology, leading to new insights, new capabilities, such as the from amino acid sequences, as well as new therapeutic approaches.

The age of digital biotech (as it is sometimes called) has furthered applications of AI in new areas of innovation at Flagship, and is core to the pioneering efforts of many of our newer companies. Examples include:

Generate Biomedicine鈥檚 in silico
the at a single cell level as pioneered by Cellarity
Valo Health鈥檚
the prediction of the impact of known and potential future variants of different viruses, including identification of high-risk variants of concern by Flagship鈥檚 Apriori Bio

Moving forward, the exponential increase in biological data generation, computing power and AI model performance will accelerate the pace of discovery and deepen our understanding of biological systems.

The result? Myriad opportunities for innovation that advance the health of the human species and our planet. In human health, this will mean moving beyond making predictions at the molecular level to modeling and simulating the physiology of cells, tissues and whole organs. Such a breakthrough could enable us to identify novel therapeutic targets, develop better diagnostic tools and even forecast the unique trajectory of health from predisease to disease and its progression for a specific patient. We can also leverage AI to understand more precisely and much more deeply the impact of climate change and generate novel solutions to reduce our environmental impact and increase agricultural productivity.

Leveraging Biology for AI

Since its birth, the field of AI research has been inspired by our knowledge of nature and our own brain structures. One of the main inspirations has been leveraging biology to help solve complex optimization problems. In fact, many computing algorithms, including widely used genetic algorithms, have been developed based on the principles of evolution and observed behavior in biology. For example, the genetic bee colony algorithm was invented based on the behaviors of three types of bees (scouts, onlookers, employees) to perform multiobjective layout optimization, and has been . Biology has driven numerous other applications of AI research including developing car collision-prevention mechanisms inspired by the , incorporating attention mechanisms to to solving a given problem and leveraging the concept of to optimize the performance of algorithms.

In the next decade, we will see a symbiotic relationship emerge between biology and AI innovation. Biology will inform the design of better-performing AI systems, which will enable extraordinary leaps in our fundamental biological understanding, which in turn will advance AI research further.

Still, bio-inspired computation is only in its infancy. When it comes to modeling more complex concepts that can serve as trampolines for the next leaps in AI performance, we鈥檙e still at the starting gate. Developing AI models based on the defense mechanisms of the human immune system (e.g., T cells, antibodies) can help us develop novel techniques to . Applying the ingenious methods nature has created to encode our genetic information (e.g., RNA transcription, gene regulatory networks) can help us and develop novel .

Arguably, the largest opportunity lies in computational neuroscience. The AI research community has yet to successfully mimic and/or fully leverage a human being鈥檚 ability to continuously learn and transfer learning from one context to another. Neither have they been able to encode our remarkable capacity to learn new concepts without having cleanly labeled data, which is a consistent challenge for most AI systems. The holy grail is to create AI agents that can emulate human creativity and imagination to tackle massive scientific challenges that may otherwise be impossible to solve.

Conclusion

In the past decade, we have seen a flourishing of innovation in biology and AI that has occurred largely in separate spheres. In the next decade, we will see a symbiotic relationship emerge between biology and AI innovation. Biology will inform the design of better-performing AI systems, which will enable extraordinary leaps in our fundamental biological understanding, which in turn will advance AI research further. This will create a mutually reinforcing convergence of innovation in biology and AI.

Silos will continue to break down between the two fields as scientists realize that the problems they are each trying to solve have more in common than previously known. For example, as the complexity of deep learning models grows significantly, our ability to understand and explain how an AI agent comes up with conclusions will diminish. Then, in the same manner that biologists have had to reverse engineer the complexity of human life, AI researchers will need to reverse engineer the complexity of black-box models.

The overall implication is clear to us. As innovators in biology, we should not be focused solely on understanding how to best leverage AI to solve our problems. We should also be asking ourselves what we can do to drive innovation in AI because our health, and that of our planet, are at stake.

The authors would like to thank Rahi Punjabi, innovation associate, for his contribution to this essay.

Image credit: iStock. Vector illustration of virtual human body polygonal on circular anatomical future system based on health innovation and technology.

Story By

Armen Mkrtchyan

Armen Mkrtchyan is an Origination Partner at 蘑菇视频APP and leads Pioneering Intelligence, an initiative to institutionalize and expand the use of Artificial Intelligence (AI) across Flagship and portfolio companies, and to help drive鈥�

Karim Lakhani

Karim Lakhani joined 蘑菇视频APP as an academic partner in 2021. In this part-time role, Karim works with Flagship Founder and CEO Noubar Afeyan and Flagship leadership to help formulate, design and implement digital and machine learning鈥�