← Back to After AGI After AGI

Science

AI researchers working 24/7. Discovery accelerates. AlphaFold was just the beginning — what comes next changes the pace of human knowledge itself.

How science changes

Science has always been bottlenecked by human cognition. That bottleneck is being removed.

The scientific method hasn't changed its fundamental shape in 400 years: hypothesis, experiment, analysis, conclusion. A human scientist does all four steps, sequentially, limited by their own mental capacity, working hours, and access to literature. The entire enterprise of modern science is built on the assumption that thinking takes time.

AI breaks that assumption at every step. It generates hypotheses by finding patterns across every paper ever published. It designs experiments by simulating millions of possibilities. It analyses results by processing data at scales humans can't comprehend. It writes up conclusions in perfect prose. And it does all of this 24 hours a day, in parallel, across every field simultaneously.

This isn't "AI helps scientists." It's a structural shift in the rate at which knowledge can be produced.

The numbers

200M Protein structures solved by AlphaFold (vs 200K before)
1000x Faster materials discovery with AI simulation
3 yrs Time saved in drug design per candidate molecule

These numbers represent the early phase — narrow AI applied to specific problems. The coming wave is general: AI systems that can work across biology, chemistry, physics, and materials science simultaneously, transferring insights between domains in ways humans can't.

Fields being transformed

Biology and medicine

Protein folding, drug discovery, disease mechanism identification, personalised treatment design. This is the most advanced AI application in science. AlphaFold solved a 50-year problem. The next generation doesn't just predict structures — it designs proteins for specific therapeutic functions. Clinical trials are the bottleneck now, not discovery.

Materials science

Battery chemistry, catalysts, structural materials, superconductors. The traditional process — synthesise, test, iterate — takes years. AI can simulate millions of candidate materials and identify the most promising ones for synthesis. The time from discovery to deployment could compress from decades to years. This has direct implications for climate tech, electronics, and manufacturing.

Physics and cosmology

Particle physics data analysis, cosmological simulations, quantum system design. AI finds patterns in datasets too large for human analysis. It designs experiments and predicts outcomes. In cosmology, it models galaxy formation and dark matter distributions. In quantum physics, it designs new experimental setups. The LHC generates petabytes of data that humans can only analyse with AI assistance.

Climate and environment

Climate modelling, carbon capture material design, renewable energy optimisation, ecosystem monitoring. AI improves prediction accuracy, runs faster simulations, and identifies intervention points humans wouldn't spot. The scale of climate science — thousands of variables interacting in complex systems — is exactly what AI excels at.

Computer science research

AI improving AI. Novel architectures discovered by automated search. Algorithm optimisation. Security vulnerability detection. This field is uniquely positioned because the AI researcher and the research subject are the same thing — AI systems exploring the space of possible AI systems. Self-improving loops could accelerate advancement unpredictably.

Neuroscience

Brain mapping, neural decoding, consciousness research. AI helps interpret neural recordings at unprecedented resolution. It identifies patterns in brain activity that correspond to thoughts, memories, intentions. The connection between biological and artificial intelligence becomes a two-way street: AI helps us understand brains, and understanding brains helps us build better AI.

Timeline

Now
AI is a standard tool in every major lab. Literature review, experimental design, and data analysis are AI-assisted. Drug candidates are being discovered by AI that would have taken years of human research. Materials with novel properties are being predicted and synthesised. The first AI-discovered drugs are in clinical trials.
2027
Autonomous labs — AI systems that design experiments, run them with robotic equipment, analyse results, and iterate without human intervention. The "AI scientist" becomes a functional reality for well-defined research problems. Human scientists shift to framing questions, validating results, and exploring the unexpected.
2028+
Breakthrough cascade. AI discoveries enable more AI discoveries. The rate of scientific progress becomes limited not by human intellect but by physical constraints — lab capacity, compute availability, clinical trial speed. Fields that were stuck for decades (longevity, fusion, Alzheimer's) see sudden acceleration. The map of knowledge expands faster than humans can read.

The risks

Accelerated science isn't automatically good. It creates new categories of risk.

Dual-use discovery

If AI can design better drugs faster, it can also design better pathogens. The same capability that accelerates vaccine development can accelerate bioweapon development. Laboratory safety protocols and publication norms were designed for human-paced research. AI-paced research breaks those norms.

Reproducibility crisis amplified

AI generates papers faster than humans can verify. The flood of AI-produced research already straining peer review becomes a tsunami. Fraud detection becomes harder when AI can generate convincing fake data. The signal-to-noise ratio of scientific literature degrades just as its volume explodes.

Black box knowledge

AI may discover things that work but that humans don't understand. A new cancer drug that outperforms everything else — but we can't explain why it works. A better battery chemistry that nobody can derive first principles for. We enter a world where we trust AI findings without understanding them. This is epistemically uncomfortable and practically risky.

Human obsolescence in research

Grad students, postdocs, research assistants — the roles that train the next generation of scientists — are vulnerable. AI does literature review, experimental design, and data analysis. The apprenticeship model of scientific training depends on junior researchers doing this work. Remove it and the pipeline for future senior scientists narrows.

What it means

Cures accelerate

Cancer, Alzheimer's, Parkinson's, ALS — the diseases that have resisted decades of research may yield to AI-driven hypothesis generation and drug design. The timeline compresses from "maybe in 20 years" to "likely in 5-10." The first AI-discovered drugs reaching patients will be a signal of whether the acceleration is real.

Climate breakthroughs

Better batteries, efficient solar cells, carbon capture materials, fusion reactor design — AI accelerates progress across the entire climate tech portfolio. The question shifts from "can we solve climate change" to "can we deploy solutions fast enough" — a policy and engineering problem rather than a fundamental science one.

Longevity research

Aging is a biological process that AI may help understand and intervene in. Not just treating age-related diseases, but the fundamental mechanisms of aging itself. The implications for society — retirement, healthcare costs, population structure — are as significant as the medical ones.

New domains

AI might discover entirely new fields of science — phenomena, relationships, and laws that humans never thought to look for. The most transformative discoveries may not be "solved problems" but "questions we didn't know to ask." That's the genuinely unpredictable part of accelerating science.

The AI scientist

Imagine a researcher with perfect memory, infinite attention, and no need for sleep.

An AI researcher that has read every paper ever published in every field. That generates hypotheses by connecting insights across biology, chemistry, physics, and materials science. That runs experiments in simulation 1,000 times faster than real time. That writes up results, generates figures, and prepares submissions. That does all of this simultaneously across every research domain.

Early versions exist now. AI systems that design, execute, and analyse experiments with minimal human oversight. By 2027, these systems will be running entire research programs in well-defined domains. The bottleneck shifts from "can we think of this" to "can we build and test this."

This changes science as fundamentally as the invention of the scientific method itself.

The bottom line

Science is about to get faster. Much faster. The rate at which we discover new knowledge will accelerate in ways that most scientists themselves haven't fully absorbed. Some problems that have resisted human effort for decades will suddenly yield. New problems — of oversight, verification, and dual-use risk — will emerge.

The age of discovery isn't over. It's accelerating.