A Carnegie Mellon and Georgia Tech study finds E. coli store memory the same way AI models do
If you dropped the exact same pulse of food onto two identical bacteria, you'd expect the exact same response. That's not what happened. In a study published in PRX Life in May 2026, researchers at Carnegie Mellon University and Georgia Tech found that E. coli exposed to unstable, feast-and-famine conditions adapted to new food far faster than bacteria pulled from a stable environment, even when the immediate conditions in front of them were identical.
You may think that one cell has no place to keep things such as memories. Scientists, including Josiah Kratz, observed tens of thousands of individual E. coli bacteria through a special microfluidic apparatus and found that the cause was ribosomes, the components of the cell responsible for building proteins and determining the growth rate.
Some ribosomes were responsive to nutrient changes right away, whereas other ribosomes took their time; in combination, they created memory for each cell ranging from minutes to hours.
Here's the detail that should catch your attention if you follow AI at all: the researchers found this ribosome network follows the same computational logic as a gated recurrent neural network, a model architecture used to process sequences like speech.
Bacteria use molecular "gates" to decide how much of an old memory to keep versus overwrite, functionally identical to the gates modern AI systems use to balance memory against new information, except here, the gate is chemistry, not code.
One would expect that having memories is something that comes without any cost, but this is not so. The ability to be ready for adaptation takes resources from growth, so all bacteria make the trade-off between stability and flexibility at all times, just like engineers make when developing learning AI systems.
The scientists state that a lot of pathogenic bacteria survive only due to the fact that they adapt to the changing environment in the body.