Can We Program Biology? Inside the Rise of AI-Designed Molecules

SEO Title: Can We Program Biology? AI-Designed Proteins Explained

Meta Description: How AI models like AlphaFold and RFdiffusion are helping scientists predict and design proteins — and what that really means for medicine and biotech.

URL Slug: /can-we-program-biology-ai-designed-proteins

Tags: AI in biotechnology, AI protein design, AlphaFold, drug discovery, synthetic biology, computational biology, future of biotechnology, protein structure prediction

The Molecule That Didn't Exist Yesterday

Imagine handing a computer a problem that has nothing to do with code. Not "sort this list" or "optimize this route," but something stranger: design a molecule that has never existed in nature, one that will fold into a precise three-dimensional shape and latch onto a specific target inside a human cell.

No training data of "correct answers" in the usual sense. No compiler to check if it worked. Just decades of biological knowledge, a few billion parameters, and a question that biologists have been asking since the 1970s: if we know the building blocks of life, can we predict — and eventually design — the machines they build?

A few years ago, this would have sounded like science fiction, or at best a distant research goal. Today, it's a Tuesday-afternoon task at labs from Seattle to Shenzhen. Researchers at the University of Washington's Institute for Protein Design routinely ask a computer model to sketch out proteins that don't occur anywhere in nature, then send the blueprint to a lab bench to see if it actually folds the way the model predicted. Sometimes it does. Sometimes it doesn't. But the fact that this is even a workable process — sketch, predict, test, refine — represents a real shift in how biology gets done.

This isn't about AI writing DNA the way a programmer writes Python. Nobody is typing create_organism() and pressing enter. What's actually happening is narrower, more grounded, and in some ways more interesting: AI models are learning the patterns hidden in millions of known protein structures, and using those patterns to predict how new sequences will fold, or to propose entirely new sequences built to do a specific job.

In 2024, the Nobel Prize in Chemistry went to three scientists for exactly this kind of work — <cite index="7-1">Demis Hassabis and John Jumper for developing an AI model that predicts protein structures, and David Baker for pioneering the computational design of entirely new proteins</cite>. That prize is a useful marker. It tells us this field has moved past hype and into results that the scientific establishment considers foundational.

So here's the question worth sitting with as you read on: if biology can increasingly be predicted and designed with software-like tools, are we really "programming" biology — or are we just getting much better at understanding a system we didn't build in the first place? Keep that question in mind. We'll come back to it.

Why Biology Has Always Been Slow to Crack

Traditional science runs on a loop that hasn't changed much in a century: come up with a hypothesis, run an experiment, watch it fail, adjust, and try again. That loop works — it's given us vaccines, antibiotics, and a century of medical progress — but it's slow and expensive, especially in biology.

Here's why. A single protein — one of the tiny molecular machines that do almost all the work inside a living cell — is built from a chain of building blocks called amino acids. A protein of average size might have a few hundred amino acids, and each one can be any of 20 different types. That chain doesn't stay a flat string; it folds up into a specific 3D shape, and that shape is what determines whether the protein works. Change one amino acid in the wrong spot, and the whole shape — and function — can collapse.

The number of ways a chain like that could theoretically fold is almost incomprehensibly large — far more possibilities than there are atoms in the observable universe, a puzzle biologists nicknamed "Levinthal's paradox" back in the 1960s. For decades, the only reliable way to find out a protein's actual shape was to grow it, purify it, and study it with slow, expensive lab techniques like X-ray crystallography — a process that could take years per protein.

That's the bottleneck AI has started to chip away at.

Traditional biological discovery: Research → Experiment → Failure → Experiment again → Discovery

AI-assisted biological discovery: Biological data → AI model → Prediction or design → Laboratory testing → Improvement

Notice what didn't disappear in that second loop: laboratory testing. AI doesn't replace the wet lab. What it does is dramatically narrow down what's worth testing in the first place, so scientists spend less time guessing and more time confirming

What Does "Programming Biology" Actually Mean?

Let's be precise about the title of this article, because the phrase "programming biology" can be misleading if taken literally.

In traditional software, the relationship is direct:

Human → Code → Computer → Output

You write instructions, the computer executes them exactly, and you get a predictable result. In AI-assisted biology, the chain is longer and far less certain:

Human → Biological goal → AI model → Molecular design → Laboratory testing → Biological result

The AI model doesn't execute a command with certainty — it makes an educated, data-driven guess, informed by patterns learned from real biological structures. That guess then has to survive contact with an actual living system, which is messy, complex, and not obligated to behave the way a model predicted.

To make the underlying biology concrete, it helps to think of it like an instruction manual and a workshop. DNA is the instruction manual — a long molecule that stores the information cells need. Sections of that manual, called genes, get copied into a related molecule called RNA, which acts like a working photocopy carried out of the archive and into the workshop. There, molecular machinery reads the RNA and builds a protein — a folded chain of amino acids, the actual machine that does the work: digesting food, fighting infection, contracting muscle, and thousands of other jobs. Some proteins are themselves specialized tools called enzymes, which speed up chemical reactions that would otherwise happen too slowly to sustain life.

If DNA is part of the instruction manual, proteins are some of the machines built from those instructions. AI models that predict or design proteins are, in effect, learning to read blueprints and sketch new ones — not writing the manual from scratch, and certainly not building a living organism in one step.

Why Is Biology So Hard to Design?

Even once you understand the DNA-to-protein pipeline, designing biology on purpose is a different level of difficulty than understanding it.

A few reasons why:

The search space is enormous. For a modest protein of 300 amino acids, the number of possible sequences vastly exceeds anything a lab could test by trial and error.
Small changes, big consequences. Swapping a single amino acid can be the difference between a protein that works perfectly and one that folds into a useless — or harmful — clump.
Context matters. A protein doesn't act alone. It interacts with water, salts, other proteins, membranes, and a constantly shifting cellular environment. A design that looks perfect on a computer screen can behave differently inside a real cell.
Experiments are slow and expensive. Producing and testing even one candidate protein in a lab can take weeks and real money — which is exactly why narrowing the candidate pool matters so much.

A useful way to think about it: designing a working protein can be more like designing a machine with thousands of interacting moving parts than writing a few lines of code, where each part influences every other part, and you often can't know if it works until you build the whole thing.

This is precisely the kind of problem computational prediction is good at — not eliminating the uncertainty, but reducing it enough that scientists aren't starting from a blank slate.

How AI Actually Learns Biology

It's worth being clear-eyed about what's happening inside these models, because it's easy to slip into language that makes them sound more magical than they are.

The general process looks like this:

Biological data is collected — for example, the structures of hundreds of thousands of proteins that scientists have painstakingly determined over decades using lab techniques like X-ray crystallography and cryo-electron microscopy.
That data is cleaned and represented mathematically — sequences and structures are converted into a form a neural network can process.
The AI model learns statistical patterns — which arrangements of amino acids tend to produce which shapes, and which shapes tend to produce which functions.
The model makes predictions — given a new sequence, what shape will it likely fold into? Or, given a target function, what sequence might achieve it?
Researchers generate candidate molecules based on those predictions.
Candidates are tested experimentally in real cells or purified protein systems.
Results feed back into the process, refining future predictions.

What the model is not doing is understanding biology the way a human scientist does, with intuition about why something works. It's recognizing statistical patterns in enormous datasets and extrapolating from them. That's an important distinction — pattern recognition at scale is powerful, but it's not the same as comprehension, and it can fail in ways that are hard to predict in advance.

The Breakthrough: AI and Protein Structure

The clearest entry point into this whole story is AlphaFold, the AI system built by Google DeepMind.

For decades, predicting a protein's 3D shape from its amino acid sequence — the "protein folding problem" — was considered one of biology's hardest open questions. <cite index="11-1">Researchers had been trying to solve it since the 1970s</cite>, and progress was slow enough that many scientists doubted it would ever be solved computationally.

That changed in 2020, when DeepMind's AlphaFold2 <cite index="11-1">performed almost as well as expensive, time-consuming lab techniques like X-ray crystallography</cite> at a major structure-prediction competition. <cite index="4-1">The team then used the model to predict the structures of over 200 million proteins</cite> — <cite index="4-1">essentially all proteins that scientists have ever sequenced</cite> — and made that database freely available to researchers worldwide.

The impact was immediate and measurable. <cite index="11-1">By October 2024, AlphaFold2 had been used by more than two million researchers across 190 countries</cite>, and <cite index="11-1">work that once took years to determine a single protein structure could now be done in minutes</cite>. In 2024, <cite index="4-1">DeepMind released AlphaFold3, which goes further than shape prediction alone — it can also estimate how small molecules, including potential drugs, might bind to a given protein</cite>, a capability now being applied commercially through DeepMind's drug-discovery spinout, Isomorphic Labs.

This work earned <cite index="7-1">Demis Hassabis and John Jumper</cite> half of the <cite index="7-1">2024 Nobel Prize in Chemistry</cite>, "<cite index="7-1">for protein structure prediction</cite>," making it one of the first Nobel-recognized breakthroughs driven fundamentally by AI.

It's worth being precise about what this achievement is and isn't. AlphaFold predicts structure — it tells you what a known or hypothesized protein probably looks like. It does not, by itself, invent new proteins or guarantee that a predicted structure will behave a certain way inside a living organism. That next step — design rather than prediction — is a separate, and arguably harder, challenge.

From Predicting Biology to Designing Biology

This is where the story shifts from reading biology to writing it — carefully, and with a lot of help from human scientists.

Prediction answers the question: what does this biological molecule probably look like?

Design answers a different, harder question: what biological molecule should we build to achieve a specific function?

The first stage was teaching machines to read biology. The next stage is asking them to help write it.

This is the work David Baker's lab at the University of Washington has spent years pioneering, first with a tool called Rosetta and, more recently, with an AI system called RFdiffusion. <cite index="16-1">RFdiffusion works like the image-generating diffusion models behind tools such as DALL-E, except instead of starting from visual noise and gradually forming a picture, it starts from random noise and gradually forms a workable protein structure</cite>. <cite index="16-1">With older design methods, researchers sometimes had to test tens of thousands of candidate molecules in the lab before finding one that worked as intended; with RFdiffusion, the team found they sometimes needed to test as few as one design per challenge</cite>.

That's not a small improvement — it's the difference between an approach that's realistic to try and one that mostly isn't. As one outside researcher described the shift, <cite index="13-1">before AI-driven diffusion methods arrived, the success rate for these kinds of experimental designs was roughly one working design out of every ten thousand attempts; with diffusion-based tools, success rates in the lab climbed into the low single-digit percentages</cite> — still far from perfect, but a dramatic jump.

It's crucial to be honest about the limits here. Researchers using these tools are proposing candidate proteins computationally and then testing whether they actually fold and function correctly in the lab — most designs still fail that real-world test. AI is not independently creating fully functional living organisms, and no serious researcher in this field claims otherwise. What's changed is the ratio of promising ideas to lab hours required to test them.

Real-World Examples: Where This Is Actually Being Used

🧬 Protein design

David Baker's lab has used RFdiffusion and related tools to design entirely new proteins — including <cite index="12-1">proteins with binding pockets shaped to hold onto specific compounds, such as a steroid hormone, an iron-carrying blood molecule, and a chemical plants use to absorb sunlight</cite> — and confirmed in the lab that many of these designs actually worked as intended. The problem was that nature's existing protein "parts catalog" is limited to what evolution happened to produce; AI design lets researchers propose parts nature never built, aimed at a specific job.

💊 Drug discovery

Insilico Medicine, a biotech company using AI throughout its drug pipeline, used its Pharma.AI platform to identify a new drug target for idiopathic pulmonary fibrosis (a rare, currently hard-to-treat scarring lung disease) and then design a candidate molecule to act on it. <cite index="27-1">The entire preclinical program, from an initial hypothesis to a finished drug candidate ready for trials, took under 18 months and roughly $2.6 million</cite> — <cite index="29-1">compared with a traditional process that often takes more than a decade and costs upward of $2 billion</cite>. The candidate, now known as rentosertib, later showed <cite index="30-1">a dose-dependent improvement in lung function in a Phase IIa clinical trial, compared with a decline in the placebo group</cite>. It still isn't an approved medicine — but it's a real, ongoing example of AI shortening the earliest, slowest part of drug development.

🧪 Enzyme engineering

Researchers at the University of Texas at Austin used a machine-learning model to redesign PETase, a natural enzyme that breaks down PET plastic (the material in most drink bottles). <cite index="31-1">The resulting enzyme, called FAST-PETase, can break down plastics that normally take centuries to degrade in a matter of hours to days</cite>. <cite index="35-1">The team tested it against 51 different post-consumer plastic products — bottles, containers, fibers — and found it broke them down effectively</cite>. We'll look at this one in more depth in the case study below.

🦠 Disease research and antibody design

Beyond small-molecule drugs, several companies — Xaira Therapeutics, Nabla Bio, Chai Discovery, and Absci among them — are now using AI to design antibodies (the immune system's targeting proteins) from scratch. <cite index="46-1">A 2025 study from researchers connected to David Baker's lab generated full antibodies computationally that successfully bound to specific, chosen targets with high structural precision</cite>. This field is newer and less proven in the clinic than small-molecule AI drug discovery, but it points toward AI playing a role in a much wider range of medicines, not just simple molecules.

🌱 Industrial biotechnology

The same design techniques used for medicines are being applied to industrial problems: enzymes engineered to work at unusual temperatures or pH levels, tailored for manufacturing processes, or built to break down waste materials — exactly the kind of application the FAST-PETase project demonstrates.

🧫 Synthetic biology

Some labs are going a layer further, trying to model how entire networks of genes and proteins behave together inside a cell, rather than just individual molecules. <cite index="43-1">Xaira Therapeutics, for instance, released research using single-cell RNA sequencing data from roughly eight million cells to start mapping the networks that regulate how genes turn on and off</cite> — an early step toward what some researchers call a "virtual cell." This is genuinely early-stage work, closer to a research direction than a finished tool.

Mini Case Study: Teaching AI to Design a Plastic-Eating Enzyme

The Problem Plastic pollution is one of the more visible environmental problems of the last fifty years. PET plastic — used in bottles, food containers, and synthetic fabrics — <cite index="36-1">makes up a huge share of the roughly seven billion tons of plastic waste that has piled up in landfills and oceans since 1950</cite>, and it can take centuries to break down naturally.

The Old Challenge In 2016, scientists discovered a naturally occurring enzyme, PETase, in bacteria found near a plastic bottle recycling plant — the bacteria had evolved to use PET as a food source. But natural PETase is slow, and it stops working well outside a narrow temperature and acidity range, which limited its usefulness for large-scale recycling. Traditional protein engineering — testing mutations one at a time in the lab — would have meant sifting through an enormous number of possible amino acid changes to find ones that actually helped.

The AI Approach A team at the University of Texas at Austin, led by Hal Alper, trained a machine-learning model to predict which specific mutations to PETase would make it both faster and more tolerant of different temperatures and pH levels — narrowing millions of theoretical combinations down to a small, promising set. <cite index="32-1">The model helped the team zero in on three suggested amino acid substitutions</cite>, which were then combined with existing modifications from earlier engineering work.

The Experimental Stage The resulting enzyme, FAST-PETase, was tested in the lab against real-world plastic waste. <cite index="31-1">Researchers used untreated, post-consumer PET from 51 different products and found the enzyme could nearly completely degrade them within a week</cite>. <cite index="32-1">At 50°C, it worked almost twice as fast as the next-best engineered enzyme, which needed a much higher temperature of 70°C to perform well</cite>. <cite index="32-1">In one demonstration, it broke down an entire plastic cake tray within 48 hours</cite>.

The Result The 2022 study, published in Nature, described a plastic-degrading enzyme meaningfully faster and more practical than anything engineered before it — with the crucial caveat that AI-guided mutation prediction, not lab trial-and-error alone, is what got the team there in a workable timeframe.

Why It Matters This case study is a good reminder that "AI-designed biology" isn't only about medicine. The same core technique — using a model to predict which small changes will help, instead of testing every possibility by hand — applies just as well to environmental and industrial problems.

AI-Designed Medicines: What's Real and What's Still Far Off

It's tempting to read a headline like "AI designs a drug" and assume a computer produced a finished medicine ready for a pharmacy shelf. That's not what's happening, and it's worth being direct about it: AI does not create medicines instantly.

What AI actually helps with, at the current stage of the field:

Identifying promising biological targets (proteins or genes associated with a disease)
Predicting molecular properties, like how well a candidate might bind to its target
Estimating protein–drug interactions before any lab work happens
Narrowing an enormous field of possible molecules down to a manageable shortlist
Speeding up the early, most exploratory phase of research

None of this replaces the rest of the pipeline, which remains long and tightly regulated: Discovery → Laboratory testing → Preclinical studies → Clinical trials → Regulatory approval. Even Insilico's much-publicized rentosertib, arguably the most advanced AI-designed drug on the market's radar, has spent years moving through exactly these stages, the same as any other experimental medicine, precisely because safety and efficacy still have to be proven in real patients, not just predicted in silico.

AI-Designed Enzymes and a More Sustainable World

Here's a question worth sitting with: what if, instead of building bigger machines to solve environmental problems, we could design tiny biological machines to do the work instead?

That's essentially what's happening with enzyme engineering. Beyond plastic degradation, researchers are exploring AI-guided enzyme design for:

Breaking down agricultural and industrial waste more efficiently
Producing food ingredients through more sustainable chemical routes
Replacing energy-intensive industrial chemistry with milder, enzyme-driven reactions
Supporting cleaner manufacturing processes in general
Assisting environmental cleanup efforts, such as pollutant breakdown

This is one of the areas where AI-assisted biology has a chance to feel less abstract than a new drug candidate winding through clinical trials — it's a fairly direct line from "we designed a better enzyme" to "less plastic waste sits in a landfill for centuries."

The Limitations: Where This Technology Still Falls Short

It's easy, writing about a field this exciting, to slip into overstatement. So here's the sober list.

AI predictions can be wrong. Structure and design models are estimates based on patterns, not certainties.
Biological systems are extremely complex, and a model trained on isolated proteins may not capture how a molecule behaves in a full living system.
Laboratory validation remains essential. Every serious research group treats AI output as a hypothesis to test, not a finished result.
Data quality matters enormously. Models are only as good as the biological data they were trained on, and gaps or biases in that data can lead to blind spots.
A design that works computationally may not work biologically. Folding correctly on a screen doesn't guarantee a molecule will be stable, non-toxic, or functional inside a real cell.
Biological safety has to be separately evaluated for anything intended for medical or environmental use.
Regulatory approval still takes years, regardless of how fast the computational design phase was.
High-quality biological data can be hard to obtain, especially for less-studied organisms, diseases, or molecule types.

None of this diminishes what's been achieved — it just means the honest framing is "AI dramatically narrows the search," not "AI solves biology."

The Big Ethical Question

As AI-assisted biological design becomes more capable, it raises questions that don't have tidy answers yet.

Who owns an AI-designed molecule? If a model trained on public scientific data proposes a new protein, questions of intellectual property, credit, and commercial rights are still being worked out across the industry.

What happens if an AI-designed protein has an unexpected effect? Responsibility in a process that blends human goals, machine predictions, and lab validation isn't always straightforward to assign.

How should these tools be regulated? Existing frameworks for drug and biotechnology approval were built around human-led research pipelines; regulators are actively adapting them to account for AI-generated candidates.

Could this technology be misused? The same tools that can design a helpful new enzyme could, in principle, be misapplied toward harmful biological agents — which is why leading labs in this space have adopted screening processes and, in some cases, restricted publication of the most sensitive design capabilities.

Should there be limits on what AI is allowed to design? This is genuinely contested, and reasonable people in the field disagree, weighing the pace of medical progress against the risks of moving too fast in a domain where mistakes can have biological consequences.

There's real potential here — faster treatments for diseases that have gone underserved, more sustainable industrial processes, tools for handling future pandemics more quickly. There's also real risk, and most serious researchers in this space treat safety and oversight as inseparable from the science, not an afterthought.

Can AI Really "Create Life"?

This is the question underneath all the others, so it deserves a direct answer.

What AI can genuinely do today:

Predict how a known or hypothetical protein sequence will fold
Analyze huge biological datasets for patterns humans would take years to find
Design candidate proteins aimed at a specific structural or functional goal
Suggest new molecules — including potential drug candidates — for further testing
Assist early-stage drug discovery by narrowing enormous search spaces
Help researchers explore biological possibilities faster than manual methods allow

What remains much harder, or currently out of reach:

Designing a complete living organism from scratch
Predicting every interaction that happens inside a full living system
Independently creating biological systems that are guaranteed safe
Replacing laboratory experimentation as the final word on whether something works

So, are we programming biology, or simply learning to understand it better? The honest answer sits closer to the second option, at least for now. If AI can help design the instructions for a single molecule, that's a real and useful capability — but it's a long way from writing the full instruction set for a living thing. What's changed is not that biology has become code. It's that the process of discovering biology's answers has started to look a little more like search — guided, narrowed, and accelerated by machines trained on decades of hard-won data.

The Future: What's Realistic Over the Next 5–15 Years

Some of what researchers are actively working toward — labeled here clearly as potential, not guaranteed:

Faster drug discovery, with AI continuing to shrink the early research phase, even as clinical trials remain a multi-year process
More personalized medicine, where treatments could eventually be tailored based on an individual's specific biology
A wider range of AI-designed enzymes for manufacturing, agriculture, and environmental cleanup
More sustainable industrial processes, replacing some energy-intensive chemistry with engineered biological alternatives
New biological materials, potentially useful in areas from medicine to textiles
Expanded climate applications, including enzymes or organisms designed to help with carbon capture or waste processing
Better disease research, particularly for conditions where the underlying biology is still poorly understood
More accurate biological simulations, including early "virtual cell" models that attempt to represent how genes and proteins interact as a system
AI increasingly working alongside human researchers, as a tool for narrowing possibilities — not as a replacement for scientific judgment

None of this is guaranteed on any particular timeline. Biology has a track record of being harder than expected, and every one of these possibilities still depends on the same unglamorous steps: laboratory testing, safety evaluation, and regulatory review.

What This Made Me Think About

Researching this topic, the thing that stuck with me wasn't really the headline fact that AI can predict a protein's shape in minutes instead of years, impressive as that is. It was a smaller shift, almost a change in posture: AI is moving from helping us understand what already exists to helping us imagine biological structures that may not exist yet.

That's a different kind of tool. Prediction is essentially a very fast, very well-informed guess about something real. Design is something else — it's asking a model to propose something new, aimed at a purpose a human defined. I'm not a biotechnology researcher, and I'd be misrepresenting myself if I pretended otherwise. But as someone who spends a lot of time thinking about what AI tools are actually good for, this felt like one of the clearer examples of AI functioning as a research collaborator rather than a replacement — narrowing an impossibly large search space down to something a human scientist can actually test by hand. The lab bench is still where the real answer gets decided. The AI just gets you there with a much shorter list of candidates.

Can We Program Biology?

We're not at a point where a person can type a line of code and have a living organism appear. That's not what any of this is, and no credible researcher in the field claims otherwise.

But something has genuinely changed. We've taught machines to read biology — to look at a sequence of amino acids and predict, with real accuracy, the shape it will fold into. We're now teaching them to help write biology — to propose new molecular designs aimed at a goal a human scientist set. And step by step, those designs are moving off the screen and into working laboratories, plastic-degrading enzymes, and — cautiously, slowly, with every safety check still in place — early-stage clinical trials.

If software has always been written in code, and biological molecules are increasingly being designed with the help of AI, it raises a real question for the next generation of scientists: what will "programming" even mean, once the material you're programming with is alive?

Sources & Further Reading

Nobel Prize press release, "The Nobel Prize in Chemistry 2024," NobelPrize.org (2024)
Nature, "Chemistry Nobel goes to developers of AlphaFold AI that predicts protein structures" (2024)
The Conversation, "Machine learning cracked the protein-folding problem and won the 2024 Nobel Prize in chemistry" (2024)
Baker Lab, University of Washington Institute for Protein Design, "RFdiffusion: A generative model for protein design" (2023) and "Modeling and generating more of life's building blocks" (2024)
Watson, J.L. et al., "De novo design of protein structure and function with RFdiffusion," Nature (2023)
Inside Precision Medicine, "Insilico Medicine Brings First AI-Designed Drug to Clinical Trials" (2023)
Insilico Medicine, "From Start to Phase 1 in 30 Months" and "First Generative AI Drug Begins Phase II Trials with Patients"
Drug Target Review, "First AI-designed drug, Rentosertib, officially named by USAN" (2026)
GEN (Genetic Engineering & Biotechnology News), "Fast and Efficient Plastic-Degrading Enzyme Developed Using AI" (2023)
Chemistry World, "AI-engineered enzyme eats entire plastic containers" (2025 report on the original 2022 study)
GEN, "Xaira's First Virtual Cell Model Is Largest To-Date, Toward Complex Biology" (2026)
GenEngNews, "Protein Design's AI Revolution: Inside David Baker's 'Communal Brain'"

This article discusses current scientific research and is intended for general educational purposes. It does not constitute medical advice.

Search This Blog

Zyvex - Future of Technology