2025 Annual Letter: Polyintelligence

There is no universal definition of intelligence — most focus on problem solving, which is certainly essential. But I’d assert that intelligence is the capacity to learn and encode those learnings in ways that generate meaningful, adaptive models of the world, leading to effective decisions or actions. The crucial ingredient is the continual interplay of data, feedback, and model refinement — whether that “model” is encoded in silicon, neural connections, or genetic material. Data is ambient and intelligence is the ability to capture it and incorporate it into models that update based on feedback to allow more effective solutions to emerge, with or without intentionality or consciousness. In this sense, intelligence is not restricted to human cognition. Viruses, plants, and AI algorithms each display forms of problem solving. Likewise, markets and social systems can exhibit collective intelligence that surpasses that of any single entity.

Nature captures its understanding primarily through encoding it into DNA, the blueprints of life. These genetic instructions record a form of nature’s intelligence derived from historic survival and adaptation — Darwinian evolution. Evidenced by their survival and proliferation, humans have long sought to understand and reduce the unknowns of nature in ever-increasing detail and universality. The biblical reference to humans accessing the Tree of Knowledge (of everything) in The Book of Genesis captures this quest as central to the human condition.

The new element enabling polyintelligence is the rise of machine/artificial intelligence. An outcome of man-made information technologies over the past century, AI is now achieving its own understanding and models of intelligence, both in nature and in humans. This year’s Nobel prizes in physics and in chemistry underscore the significance of these technologies.

Polyintelligence creates the potential to unify the organic, the human, and the technological. It invites us to reimagine intelligence as a network of connections that transcends boundaries, enabling us to think, create, and innovate at a depth and scale that can produce solutions to some of the world’s most intractable, contemporary challenges.

This connected framework is beginning to revolutionize biotechnology. From drug development to generative synthetic biology, polyintelligence enables breakthroughs that no single form of intelligence could achieve alone. Just like the genius of da Vinci’s integrated thinking, the fusion of these diverse “brains” will reshape the possibilities of science and medicine.

At Flagship, we are using human intelligence and machine intelligence to better understand nature’s intelligence (life science) and then using that deeper understanding to imagine and create technologies that have an extraordinary impact at an accelerating pace. After all, nature has the ultimate unfair advantage: a billions-of-years head start.

Many of the companies we’ve founded at Flagship are now tapping into polyintelligence to advance novel therapeutics or climate and crop solutions. We are convinced we are at the dawn of what will prove possible. To read more about how polyintelligence is informing and guiding our work, visit here.

The rapid evolution and application of artificial intelligence means that our current technological landscape is radically different than it was just a year ago. AI tools are becoming pervasive — powerful language models can now even run on your smartphone. Recently released third-generation large language models, show remarkable strides in contextual comprehension and reasoning. Even when these systems produce errors — often referred to as “hallucinations” — they can catalyze innovative thinking and novel solutions.

To appreciate the roots of the AI field now animating so much societal discussion, we can look back to an important milestone on its journey. The term artificial intelligence traces back to 1956, during the Dartmouth Summer Research Project on Artificial Intelligence, where about a dozen participants formally established “artificial intelligence” both as a term and research discipline. The cohort posited foundational questions we still explore today: how to enable machines to use language, form abstractions, solve complex problems, and even discover, reason, and invent.

As the field of computation progressed throughout the 20th century, an equally powerful force was gathering momentum: the genetic revolution. Like two mighty rivers converging, these parallel advances in AI and biotechnology reached a point of confluence when the Human Genome Project was completed in the early-2000s — the time of Flagship’s founding. As artificial intelligence and the biotech revolution continue to converge, we are increasingly able to explore the intricacies of the genetic code and its permutations — whether across populations or even cells within the same tissue — to unlock new opportunities to safeguard health, treat disease, and protect the food supply.

This rapid progress in AI is, of course, exciting, but without the proper oversight, it will not achieve the well of potential we envision. All learning — in nature, humans, and machines — involves trial and error. Nature generates variation all the time — some variants succeed where others fail. Likewise, humans experiment with variation, too. As we’ve all experienced, some lead to success, while others result in colossal failure. It stands to reason, then, that machine learning will also involve experiments and “mistakes,” which we should not only expect but embrace as a part of progress.

The decisions we make now — about governance, ethics, and bias — will set the tone for how AI shapes our society. Its promise is enormous — even greater and more positive, perhaps, than what is deemed “hype” today would suggest. But the greatest promise of AI is what it can do when wedded to human intelligence and nature’s intelligence. Polyintelligent thinking, polyintelligent systems, and polyintelligent solutions will have unmatched — and, to date, unimagined — power to improve human health, the climate, agriculture, and a whole host of other multifaceted challenges.

It is vitally important, therefore, that we get this right.

We must ensure that the choices we make don’t inhibit AI from enabling polyintelligence and the bigger leaps it can inspire.

Part of Leonardo’s genius was to perceive hidden systems — the invisible forces and laws — that govern everything from the workings of the human body to the movement of objects through the air. Yet even centuries later, there is a great deal we neither perceive nor understand about our natural world and its various forms of intelligence.

With characteristic conceit, we tend to measure proficiency against benchmarks that reflect our human priorities, problem-solving approaches, and communication methods, which can lead to a false sense of what it means to be “intelligent.” This bias often overlooks or undervalues forms of cognition that do not resemble our own or simply evade our comprehension, whether it’s the intricate chemical communication of plants, the swarm intelligence of insects, or the so-called “hallucinations” of AI. In doing so, we sometimes miss the depth and breadth of intelligence that defies human-centric definitions, expectations, and capabilities.

For example, it turns out sperm whales communicate with their tight-knit families through intricate click patterns called codas. These animals spend most of their lives in the dark, relying on sound to navigate and maintain complex social bonds. Finding there were more clicks than could ever be manually segmented, researchers applied machine-learning to reveal that these codas comprise a nuanced “alphabet” that combine into a structured communication method. These findings challenge our notions of language and intelligence.

Consider the Human Genome Project, referenced earlier, which surprised everyone by revealing that a mere 2% of our DNA codes for proteins — as defined by a set of human-defined rules and guidelines, rendering a large fraction of the genome as “junk DNA.” At Flagship, we saw the opportunity to instead take a systematic, unbiased, and comprehensive approach to protein discovery — one that combines nature’s intelligence and artificial intelligence. With this approach, we founded ProFound Therapeutics to interrogate what was once deemed “junk,” revealing a vastly expanded human proteome that includes thousands of previously undiscovered proteins — newfound insights that could unlock a universe of potential new medicines.

As AI deepens our view into molecular physiology, we are finding more specialized and precise ways to treat human disease. For instance, a nuanced understanding of how the immune system communicates and addresses disease is eliminating the need for broad and nonspecific immune suppression. We founded Repertoire Immune Medicines to develop therapies that mimic the optimal immune response to address autoimmune diseases and cancer with precision rather than brute force.

While devastating, a disease like cancer illuminates the uncontrolled diversification of our cells. When cancerous mutations occur in certain cells, they result in tumors, but countless unseen changes are also happening in each of the other trillions of cells in our bodies. Using the latest sequencing technologies, the emerging field of somatic genomics is tapping into this dynamic genetic landscape to uncover the keys to natural disease resistance. Flagship-founded Quotient Therapeutics is applying AI to map these mutations to their functional outcomes and revealing insights that escape traditional genetics. This approach allows us to understand, for example, why some liver cells resist fat accumulation in metabolic dysfunction-associated steatohepatitis (MASH), uncovering opportunities to address this disease and so many more.

Whether decoding whale songs or discovering somatic mosaicism, machine intelligence is serving up a fascinating and humbling headline: We have long underestimated nature’s intelligence within and around us. Its variety and complexity is staggering, and far exceeds the current forms of artificial or human intelligence. Consider what multiple forms of viral, microbial, plant, and ecosystem intelligence evolved over literally billions of years can model for us and teach both humans and machines.

Through polyintelligence we can deepen our connection with the natural world and enable a collaborative exchange with our own biology and beyond.

Much as Leonardo drew inspiration from nature to fuel his myriad endeavors, we at Flagship have long taken inspiration from nature and its unrivaled capacity for emergent innovation. As such, polyintelligence has emerged as a crucial guiding principle of Flagship’s mission: not only understanding nature but innovating beyond its current capabilities. Just as an artist like Leonardo masters perspective and color theory before defying classical boundaries, in our hands, AI is learning and encoding the foundational “rules” by which nature operates to propose new solutions.

We applied this philosophy at Generate Biomedicines in developing Chroma — the first publicly released generative protein-design model. We analyzed millions of proteins to understand how nature encodes functions into proteins, then based on that understanding, imagined entirely new proteins nature has not yet devised with highly specialized therapeutic benefits. At our company Abiologics, we are similarly looking to surpass the existing limits of biology. The company is creating fully synthetic biomolecules with immune-stealth capabilities, enhanced bioavailability, and programmable functions — qualities that transcend what is possible with current therapeutics.

Flagship-founded Inari applies these concepts in another area of biology: agriculture. The company exemplifies how we can harness nature’s intelligence for precise multiplex gene editing and how AI can be used to accelerate nature’s evolutionary processes in the face of record-high temperatures, drought, and other climate-induced stressors, producing staple crops like corn, wheat, and soybeans that can thrive with fewer inputs in the face of intensifying environmental challenges.

Biology also offers endless inspiration for computational innovations. Consider how immune systems learn and adapt; by mirroring these mechanisms, we could create cybersecurity protocols that constantly refine their defenses — swiftly identifying and neutralizing digital threats just as the human body neutralizes pathogens. We can even take inspiration from the evolutionary process of mutation and selection to evolve new forms of machine intelligence.

At Flagship, we are transposing emergent principles observed in nature onto machine intelligence to pioneer agent-based workflows. Each agent collects local information, makes decisions, and refines its strategies to solve complex tasks in much the same way symbiotic organisms adapt and thrive within ecosystems in nature. These agents collectively evolve novel structures and solutions without the need for top-down orchestration, mirroring the self-organizing patterns found in biological systems. By harnessing the power of this emergent phenomenon, Flagship is advancing a new frontier in AI — one where decentralized agents continually learn, adapt, and collaborate to ultimately make machines smarter.

Take for example FL100 (a Flagship company still in stealth), which is leveraging agentic workflows to generate and validate new insights and product concepts at an unparalleled pace and scale. Or, consider FL105, leveraging the same approach to capture individual personalities and differences to provide insight into our mental wellbeing. In these cases, harnessing numerous, even conflicting viewpoints fuels an iterative and emergent approach that can uncover novel solutions beyond the reach of current methods.

So, what’s the next big leap? The emergence born of machine-machine collaboration could evolve machine intelligence that far exceeds human-level intelligence. Yet to be publicly unveiled, our company Lila Sciences is on a mission to achieve scientific superintelligence that reaches new heights of scientific discovery, effectively expanding our knowledge frontiers to realms we have yet to imagine.

Polyintelligence is hard at work at many other Flagship-founded companies including Moderna, Metaphore Biotechnologies, Montai Therapeutics, Cellarity, Empress Therapeutics, Prologue Medicines, and Sail Biomedicines, among others. Read more about these companies and their platforms here.

I suspect experiments from these newly forming companies and others like them will confirm that we have been too narrow in both our definitions of intelligence and our understanding of where it lies. Furthermore, I think it’s inevitable that our embrace of polyintelligence and a better understanding of the emergent nature of machine, human, and nature’s intelligence will serve up fundamental challenges to our seemingly settled notions of what “sentience” is, or what and where we can find evidence of “will.” Buckle up.

Like most curious humans, Leonardo was interested in life beyond our reach. While Galileo Galilei is credited with creating the first telescope used to study astronomy, Leonardo’s notebooks clearly show designs of a device that he describes as being able to “magnify the moon.” In my own lifetime, the quest for what we’ve referred to as “intelligent life forms” on planets and galaxies beyond our reach has driven everything from NASA budgets and sci-fi plots to Blue Origin and SpaceX.

At Flagship, we are fond of asking “What if …?” So, I ask you: What if we’ve been looking up, when we should have been looking around us and within us? What if the intelligence that is intra-terrestrial, and even within our own bodies, has much greater lessons to teach us — lessons that could ensure our survival as humans?

Up until this point, human intelligence has been the ultimate arbiter of what is good, correct, or intelligent. How we judge the value and promise of ideas has been contingent on our uniquely human perceptual limits and mental constructs, which have proven to be wholly inadequate for understanding complex entities or phenomena that operate beyond our comprehension, be it the musings of whales or the workings of our immune system.

As we fully harness machine intelligence to plumb the greater depths of nature, we will likely surface ideas and concepts that we may not immediately understand.

At Flagship, we overcome the often-paralyzing influence of uncertainty by taking a “future-backward” approach to innovation — envisioning better futures and working backward to achieve them. We can thereby chart a path to a future in which nature, human, and machine work in synergy to build transformative technologies that broadly and equitably benefit society. Over time, we can build trust into AI tools and systems to reduce the burden on humans and allow us to focus on more meaningful, creative, and satisfying work.

Just as Leonardo redefined the boundaries of art and science, the emergence of polyintelligence will give us the chance to spark a modern technological renaissance to tackle our greatest challenges. The ultimate reward is a collaborative future in which the boundaries of knowledge continuously expand, reshaping human health, planetary health, and the very nature of discovery itself.


Noubar Afeyan, Ph.D.
Founder & CEO, Flagship Pioneering