Prior feast-and-famine speeds how fast single E. coli respond to an identical food pulse

The control is the part worth admiring. Both sets of cells met the same influx of food, and only their recent past differed: the ones that had just come through rapid feast-and-famine cycling adapted much faster than the ones arriving from steady conditions ^[3]. Because the immediate conditions were identical, the difference has to come from something the cells were carrying rather than something they were sensing ^[5]. That separation of history from stimulus is what earns the word learning under the definition the work uses, which asks only that past experience shape later behavior ^[15].

What is more provisional in the author's account in The Conversation is the address. The memory showed itself as changes in growth rate ^[16], and a mathematical model of the internal network that controls bacterial growth located the store in ribosomes, the machinery that sets how fast a cell grows ^[6]. The supporting line is that no other component the team tested in the model reproduced the behavior ^[7]. That is elimination inside a parts list. It does not exclude a slow-changing component nobody wrote into the model, and the account describes no direct measurement of fast and slow ribosome populations in the cells themselves.

The neural-network comparison is narrower than it sounds, and more interesting for being narrow. The claim is that this molecular system follows the same basic computational logic as a gated recurrent network, the architecture used on sequences such as speech and sensor data ^[9], where a gate decides how much existing memory to keep and how much to overwrite when new information arrives ^[10]. Gating was invented in machine learning to hold long-range information while staying plastic, and a bacterium faces that tradeoff in the flesh: adapt too readily and it is exposed when conditions turn, forget too readily and it cannot anticipate a threat that recurs ^[13]. The cell also carries a price the artificial version does not, since keeping readiness to adapt comes out of growth ^[11].

The account skips a magnitude. The gap between the two histories is given as much faster, with no ratio and no time constant ^[4], and the span of the memory, minutes to hours, comes from the fast and slow ribosome populations in the model rather than from a measured decay ^[8]. No drug appears in the experiment, either. Nutrient switching is the only perturbation described; the human gut, where nutrient levels rise and fall and antibiotic threats come and go, is the motivation for asking ^[12][17].

So the result I would defend is the history-dependence itself, which the identical-present design genuinely earns across tens of thousands of tracked cells ^[2]. The location of the store is a different matter: the model supplies it, but a measurement still has to confirm it.