AI learns what it means to be alive
In 1889, a French doctor named Francois-Gilbert Viault climbed down a mountain in the Andes, drew blood from his arm and inspected it under a microscope. Dr.’s red blood cells Viault, which carries oxygen, was up 42 percent. He had discovered a mysterious power of the human body: when it needs more of these crucial cells, it can make them on demand.
In the early 20th century, scientists theorized that a hormone was the cause. They named the theoretical hormone erythropoietin, or “red maker” in Greek. Seven decades later, researchers found actual erythropoietin after filtering 670 liters of urine.
And about fifty years later, biologists in Israel announced that they had found a rare kidney cell that produces the hormone when oxygen levels become too low. It is called the Norn cellnamed after the Norse gods who were believed to control human destiny.
It took humans 134 years to discover Norn cells. Last summer, computers in California discovered them on their own in just six weeks.
The discovery came about when researchers at Stanford programmed the computers to teach themselves biology. The computers ran an artificial intelligence program similar to ChatGPT, the popular bot that learned to speak a language fluently after training with billions of pieces of text from the Internet. But the Stanford researchers trained their computers with raw data about millions of real cells and their chemical and genetic makeup.
The researchers did not tell the computers what these measurements meant. They did not explain that different types of cells have different biochemical profiles. For example, they did not define which cells catch light in our eyes, or which cells make antibodies.
The computers processed the data themselves and created a model of all the cells based on their similarity to each other in a vast, multi-dimensional space. When the machines were ready, they had learned an astonishing amount. They were able to classify a cell they had never seen before as one of more than 1,000 different types. One of these was the Norn cell.
“That’s remarkable, because no one ever told the model that a Norn cell exists in the kidney,” says Jure Leskovec, a computer scientist at Stanford who trained the computers.
The software is one of several new AI-powered programs, known as basic models, that are setting their sights on the basics of biology. The models aren’t just tidying up the information biologists collect. They make discoveries about how genes work and how cells develop.
As the models grow larger, with more and more laboratory data and computing power, scientists predict they will make deeper discoveries. They can reveal secrets about cancer and other diseases. They can devise recipes to change one type of cell into another.
“A vital discovery about biology that biologists wouldn’t otherwise have made – I think we’ll see that one day,” said Dr. Eric Topol, director of the Scripps Research Translational Institute.
How far they will go is a matter of debate. While some skeptics think the models will hit a wall, more optimistic scientists believe fundamental models will tackle even the biggest biological question of all: what separates life from non-life?