Neural networks can cope with large amounts of information, supplied as written text, facial features, industrial processes, and so on. They can recognize restricted sets of patterns better than we can. But what they do is the tapping of a tin drum compared with the symphony orchestra of natural environments.
Such coupling of environmental parameters through internal circuits allows a cell to predict future events. These circuits contain, implicitly, the probability of certain life-changing events.
If the detailed chemistry of the cell is simply the outcome of a historical ragbag of ad hoc interactions, then it will be no more predictable than the weather.