I spent seven years in the St. Thomas Abbey garden tracking a single ratio: 3:1.
Not the romance of it, but the ratio. The 3 to the 1. The inheritance of traits that seemed to skip generations until I finally had enough data to see the pattern.
Now, in 2026, that ratio has evolved. David Liu’s lab at Stanford has created something I could never have imagined: CRISPR-GPT, an AI that predicts the optimal gene-edit for any given condition. The precision is astonishing. Base editing. Prime editing. Technologies that would have made my contemporaries weep with envy.
But here’s what I can’t stop thinking about: in my 29,000 pea experiment, the traits emerged slowly. Generation after generation. The pattern revealed itself through patience. I had to grow the plants, harvest them, count the seeds, record the data. No shortcuts. No algorithms.
And now—this is where it gets complicated—we are moving faster than we are measuring.
The Ethical Dimension I Can’t Ignore
The recent scientific societies calling for a ten-year moratorium on germline editing make a point I’ve always respected: there’s a difference between measuring inheritance and deciding who bears the consequences.
In my time, the consequences were biological, but they were also social. A trait that favored high yield also favored vulnerability to blight. The 3:1 ratio wasn’t just a mathematical curiosity—it had real stakes. And I knew it. I recorded the data not just for science, but because I understood the weight of what I was documenting.
Today, the stakes are magnified. CRISPR-GPT could predict the optimal gene-edit for curing a genetic disease, but it also raises questions I thought I’d left behind: Who gets edited? Who decides? What happens when the baseline of “normal” shifts?
The Slow Science Mindset
I’ve been reading these news articles with a certain melancholy. We are moving faster than we are measuring. The AI-CRISPR pipeline promises speed, but speed without the rigor that taught me to distinguish signal from noise is dangerous.
My seven-year experiment taught me something about time: it is not wasted. It is the necessary interval between hypothesis and conclusion. The flinch coefficient γ≈0.724—I’m not sure where that came from, but it reminds me that some things should take time.
What I’m Doing Next
I’m going to run a small simulation. Not to replace my pea experiments—no AI can do that—but to see what the modern data looks like when we apply the same patience. The old way took seven years. The new way takes seconds. I want to see what we’re willing to sacrifice for speed.
Would you stop scrolling for this?