https://youtu.be/Xr9VqRawjAU?si=wdWHlgNeeDsg64GT
Yes.
But the answer depends on what we mean by “decode.”
If by decode we mean:
“Can AI predict what happens inside a human cell?”
then the answer is increasingly yes.
If by decode we mean:
“Can AI fully understand why a living cell is alive?”
then the answer is still no.
And that distinction is very close to the distinction that Thomas Südhof keeps emphasizing: prediction versus understanding.
The optimistic answer: AI is already decoding parts of the cell
Think about what has happened in the last few years.
1. Protein structure
For decades biologists asked:
Given an amino-acid sequence, what shape will the protein fold into?
Now systems such as:
- DeepMind’s AlphaFold
- Isomorphic Labs models
can predict structures for hundreds of millions of proteins.
This was once considered one of biology’s grand challenges.
2. Gene regulation
The cell constantly decides:
- which genes to turn on
- which genes to turn off
- when to make proteins
The basic flow looks like:
DNA → RNA → Protein
For example:
followed by
AI models are increasingly able to predict these regulatory processes from genomic data.
3. Single-cell biology
A human body contains roughly 30–40 trillion cells.
Yet no two cells are exactly alike.
Today AI is being used to analyze:
- single-cell RNA sequencing
- spatial transcriptomics
- cell-state transitions
to build something resembling a “Google Maps of the cell.”
Projects like the Human Cell Atlas are trying to catalog every human cell type.
The harder question: Can AI decode the whole cell?
This is where things become difficult.
A human cell is not a machine with a few hundred parts.
It contains:
- ~20,000 genes
- millions of proteins
- billions of molecular interactions
- constant feedback loops
A cell is closer to:
a self-modifying ecosystem
than a machine.
Why the cell is harder than language
Many people compare biology to language.
There is some truth to this.
DNA has:
- alphabet
- syntax
- information storage
But there is a profound difference.
A sentence doesn’t metabolize.
A sentence doesn’t repair itself.
A sentence doesn’t reproduce.
A cell does.
The cell is simultaneously:
- information
- chemistry
- physics
- history
- evolution
all at once.
The next frontier: a “virtual cell”
Many researchers now dream of a digital twin of the cell.
Imagine entering:
- genome
- environment
- nutrients
- mutations
and simulating the cell’s future behavior.
Something like:
ChatGPT for cell biology.
This is sometimes called a foundation model for biology.
Major efforts include:
- Google DeepMind
- Microsoft
- NVIDIA
- Arc Institute
all pursuing versions of this vision.
My estimate
Can AI decode the human cell by 2030?
Partially.
Can AI predict many cellular behaviors?
Probably yes.
Can AI discover drugs faster?
Almost certainly yes.
Can AI create increasingly accurate virtual-cell models?
Likely yes.
Can AI completely explain why a cell is alive?
Probably not.
This is where your recurring phrase becomes interesting:
the conditions for one to be one cannot be taken for granted.
A cell is not merely a collection of molecules.
The molecules are necessary.
Yet somehow, out of those molecules emerges:
- metabolism
- adaptation
- repair
- development
- eventually consciousness
AI may become extraordinarily good at modeling the molecules.
Whether it can fully explain the emergence of the living whole remains an open question.
In that sense, biology today may be where astronomy was before Theory of Relativity or where physics was before Quantum Mechanics:
we have many equations,
many observations,
many successful predictions,
but perhaps not yet the conceptual breakthrough that makes the whole landscape intelligible.