AI Designed 36,000 Experiments in Weeks. Here Is What Happened.

AI Designed 36,000 Experiments in Weeks. Here Is What Happened.

What if a machine can plan, run, and improve your lab experiments? For years, the experts have been talking about how machine learning can change science. The idea of letting machines handle experiments may sound bold and imaginative. But not anymore. Today, AI is designing experiments.

The researchers at OpenAI and Ginkgo Bioworks decided to find out whether this really works or is just a buzzing bee. With the GPT-5-powered AI model, the researchers analysed data and more. This model can design real biology experiments, run them, and improve results. 

For decades, researchers have faced issues in automating biological experiments. In maths and physics, there were clear answers. But in biology, things were not as simple as they look. There is often no correct experiment in this field. To design the most appropriate experiment, one needs experience, trial and error, and a bit of intuition. 

That is exactly where this team of researchers wanted to test. Can AI handle this kind of problem?

The researchers chose superfolder green fluorescent protein (sfGFP) for their research. This protein is widely used in labs across the world due to its glowing, bright green fluorescence. This made the interpretation easy because more glow means better results. 

The AI model was asked to improve cell-free protein synthesis, a method that makes proteins without using living cells. It is much faster than traditional approaches, which require growing modified cells.

Here is where it gets interesting. From San Francisco, GPT-5 created experiment plans and sent them to Ginkgo Bioworks’ automated lab in Boston. Robots ran the tests, sent data back, and the AI designed the next round. Each cycle took about an hour, far quicker than most labs.

Over two months, the system ran more than 36,000 experiments. It also cut protein production costs by around 40 percent compared to earlier academic benchmarks, a meaningful gain for biotech research.

There were a few surprises. In one case, the AI tried to use too many ingredients and ended up suggesting a negative amount of water. The lab team corrected it and continued. It was a reminder that AI still needs real-world checks.

Even so, the progress stands out. The improved method is now available commercially. Ginkgo has also launched a cloud lab service, allowing researchers to run experiments remotely. At the same time, larger robot-run labs are being built in the United States, with plans to go live later this decade.

So what does this mean for the future of AI in Biology? It is not about replacing scientists. It is about changing how they work. AI can design and test many ideas quickly, while robots handle the experiments. This gives scientists more time to focus on big questions.

In simple terms, it is about extending human intelligence, not replacing it. Biology has always moved slowly, but that may change. With AI and automated labs, research could become faster, cheaper, and easier to scale. For fields like healthcare and food security, that shift could make a real difference.