the biology revolution: when we can run a thousand experiments at a time

March 19, 2023

imagine a world where biology operates with the efficiency of silicon. where, instead of biologists working on a single experiment for months, we deploy AI-driven agents running thousands of parallel trials at once, combing through the space of possibilities in search of the next big discovery. what happens when biology scales like machine learning? let’s talk about how a world where we can iterate as fast as we can compute changes everything.

the world model: biology, but predictable

we know how neural nets get better as we scale them. throw enough data and compute at them, and suddenly, they’re capable of generating insights and writing code. now imagine an AI “world model” for biology—a kind of hyper-aware virtual lab that’s built on simulations of everything we understand about genetics, cell biology, and molecular interactions. at first, it’ll be crude and approximate, like a first-gen chatbot that stumbles through sentences. but as we feed it more data, it will refine itself, until this model isn’t just running hypothetical experiments—it’s accurately predicting real-world outcomes.

in practice, this world model means biology moves from empirical guesswork to something approaching engineering precision. need to understand how a new gene-editing approach will affect cell viability? plug it into the model. want to see what happens if you upregulate one pathway while downregulating another? run the simulation.

think of the possibilities: design proteins and gene therapies with more success than failure. predict drug interactions without animal testing. it’s a way to scale biology without scaling the suffering and trial-and-error that’s built into current biological research.

hypothesis generation: ai, the scientist

right now, hypothesis generation is an art, a talent that lives in the minds of scientists who draw on years of experience and intuition. but with AI, this process becomes scalable and, dare i say, better. imagine if every new insight into the human genome, every recent finding on cellular aging or cancer metabolism, could automatically spark thousands of new hypotheses. hypotheses that are generated not by a human piecing together disparate observations but by an AI trained on the entire corpus of biological knowledge, with a mandate to connect the dots we can’t see.

this AI wouldn’t just propose one-off experiments; it would propose entire research programs, complete with anticipated obstacles and potential downstream applications. maybe it hypothesizes a new metabolic pathway to help fight obesity, or suggests a series of genetic tweaks to extend human lifespan by 10 years. the AI’s hypotheses would come with embedded probability scores based on world-model predictions. like a self-directing scientist, the AI would focus on the highest-likelihood paths, wasting less time on dead ends.

experimentation: scaling up in silico and in vitro

now comes the magic: in silico experiments, on a massive scale. think of every lab in the world connected through a single digital platform, running virtual experiments in parallel, tweaking variables one by one, generating terabytes of data every day.

this isn’t some sci-fi lab from ex machina where AI robots do the pipetting. it’s more subtle. it’s biologists sending their ideas to the cloud, where an army of AI agents test out genetic permutations, cell cultures, drug interactions, and complex treatments at speeds a human could never match. this virtual testing short-circuits the classic bottlenecks of time and cost.

and when the AI finds promising leads? then come automated wet labs. we’re already seeing this with high-throughput genetic screens and automated chemical synthesis. imagine a lab that, given an AI output, can set up, run, and analyze hundreds of petri dishes, tweaking variables and watching for the best results, without a single human hand involved. a biologist’s job might shift to simply interpreting AI recommendations, guiding AI exploration in the directions that align with human goals (or whims).

the quest: what do we find?

so where does this all lead? what happens when science accelerates from months per experiment to thousands per second?

  1. genetic engineering without bounds: we’re already working on gene therapies, but what if we had AI designing entirely novel genes? if we reach the point where AI can understand genetic pathways at scale, then it could engineer organisms with capabilities we can barely imagine. think biofuels that are actually efficient. bacteria that consume plastic and pump out clean water. plants that thrive in salty soil.

  2. aging and longevity: biology’s holy grail—understanding and even reversing aging—might actually become accessible. aging isn’t a single process but a tangled mess of cellular degradation, genetic mutations, and metabolic slowdowns. with scaled experimentation, we’d untangle these interactions piece by piece. combine this with AI hypothesis generation, and it’s not wild to think we could start editing our genomes to slow down the clock, or even replace aging cells on a regular schedule.

  3. targeted drugs and real-time therapies: imagine medicines not just customized to your genetics but tailored to your current biological state. have an infection? in the future, a simple blood test could feed your unique cellular data into an AI that identifies, in real-time, the best molecular structure to attack the pathogen, produces it synthetically, and gets it into your bloodstream within hours.

  4. mind-machine interfaces: neurology is biology’s final frontier, and it’s filled with unknowns. ai scaling could give us a practical understanding of the human brain, mapping out neural pathways and understanding neurochemistry in ways that open up brain-computer interfaces. if we reach this point, concepts like memory storage or neural augmentation become less science fiction and more a matter of “when” rather than “if.”

  5. synthetic ecosystems and environmental repair: we talk about geoengineering as a high-risk solution for climate change, but imagine synthetic biology at scale. you could program entire ecosystems, designing microbes that consume carbon dioxide, plants that thrive in drought, or even organisms that sequester heavy metals. AI could suggest which genetic configurations would stabilize ecosystems, and high-throughput experiments could test these ecosystems in controlled environments.

but what about the ethics?

of course, the inevitable questions come up. who decides what research is pursued? when biology is scalable, do we start developing “biology monopolies,” where only the best-resourced labs hold control over the data and models? and, most of all, what happens to biology as an art, a craft practiced by human hands, when much of it can be done better by machines?

these aren’t questions to take lightly. the field of biology is at an inflection point, not unlike physics was before the atomic age. when we finally reach the point where AI scientists can run experiments, hypothesize new genes, and analyze complex systems faster than any human could, biology will become something entirely different. it’ll be driven by curiosity and the relentless push of computation rather than the slow, meticulous nature of lab work.

we can look forward to a future where biology is predictable, scalable, and shockingly efficient. but whether that’s a future we actually want? i suppose that remains to be seen.