Welcome to another edition of "Friday's Elk." I'm sending out this newsletter every other week, barring deadlines and such. It's free, but I will appreciate any support you may want to offer!

LifeGPT?

In early 2020, as the world plunged into the Covid-19 pandemic, I spent a lot of time thinking about what it means to be alive. The SARS-CoV-2 coronavirus was, depending on who you asked, a living thing or a non-living menace.

The scientists who argued that viruses were alive pointed to their remarkable power to evolve. We got to experience that power first hand, in waves of disease and death. SARS-CoV-2 evolved into Alpha, then Delta, Omicron, and then all of Omicron's highly adapted descendants.

Those who argued that viruses are not alive pointed to the virus itself. No metabolism going on in there. Just genes wrapped in a protein shell, waiting to bump into the right cell that could take it in and use it to make new viruses.

Working on my book Life's Edge, I noticed that scientists define life based on certain hallmarks. Evolution was one, and so was metabolism. In Life's Edge, I explored that hallmark in snakes, which put metabolism on spectacular display when they digest their prey whole. Homeostasis is another hallmark, one that I explored by descending into an abandoned mine in the Adirondacks to observe bats spending the winter in hibernation.

And then there's intelligence: not the brainy test-taking form that our culture prizes, but the fundamental capacity that even slime molds have to gather information about one's surroundings and make decisions.

But if you poll scientists, they will offer different sets of hallmarks as being really important to life. As a result, scientists have never settled on a single definition of life, and it's not clear if they ever will.

Recently I started to wonder if artificial intelligence might help biologists reach an agreement. I talked to scientists who are building programs similar to ChatGPT that work a little like virtual biologists. They look at vast amounts of data about cells and teach themselves the patterns in the information.

These programs can recognize distinct patterns of gene activity in certain types of cells, for example, enabling them to classify cells they haven't seen before. They can make predictions about what will happen if a drug enters a cell and shuts down a particular gene. It may not be very long before they can figure out the cocktail of chemicals required to reprogram cells into new types.

Last week, I wrote an essay for the New York Times about these so-called foundation models. Scientists are divided about how well they're performing at the moment, but it's important to bear in mind that they have just gotten started--the oldest of them are just a few years old.

It's hard to say how far they will develop in the next few years, but it's worth looking at another case study in A.I. to consider what is possible. For decades protein scientists struggled to predict how a given sequence of amino acids folds into a three-dimensional protein. Foundation models helped to solve the folding problem once and for all. And then they kept getting better: new versions could generate a sequence of amino acids that would fold into a desired final shape. This week, scientists reported they can use these models to design antibodies from scratch.

Foundation models can improve a lot in very little time. So how far could foundation models for cells go? The biggest speculative leap I encountered came from Steven Quake, a biologist at Stanford. He envisioned a computer inspecting all the activity of genes in billions of cells from a wide range of species. Ultimately, it might be able to join all the patterns in the data into a map of the possible. The map would represent different combinations of genes as points in a multidimensional space. It's conceivable that the points would form a single cloud: a space of genetic possibilities that permit a cell to live. Outside that cloud, no life can exist.

It's a provocative idea that wasn't even imaginable when I was writing Life's Edge just a few years ago. Now it's something that scientists can debate. Some of the scientists I talked to were skeptical about it: Quake implicitly assumes that life is made of cells that contain DNA and proteins. Perhaps a theory of life needs to be based on a deeper foundation than the biology that exists on Earth.

Fortunately, we won't have to wait too long to see how well this map of life will turn out. The computers are working hard right now. I might even try to write a revised edition of Life's Edge if something big happens--that is, assuming that writers like me haven't been replaced by A.I.!

More Stories

--Humans go through menopause. Some chimps do, and so do five species of whales. I wrote about why menopause is so rare, and why it evolved at all.

--On Wednsday, I also wrote about one of a remarkable archaeological site: it captures the life of hunter-gatherers 74,000 years ago as they lived through a giant environmental catastrophe. It revealed just how flexible modern humans were by then.

If you'd like to learn more about my books, you can visit my web site

"Friday's Elk" is free. If you'd like to support my writing, you can pay what you'd like for an optional subscription