Researchers try to cut the genetic code from 20 to 19 amino acids

That said, the results also show the limitations of working with current AI models, largely because, unlike humans, they can’t really explain the process by which they make decisions. For example, some models made very different suggestions from each other, which the researchers say implies that they are exploring different regions of the space of possible sequences. But we don’t actually know if that’s the case or if each model had mathematical reasons to dislike the other’s suggestions.

This is one of several cases presented in the paper where researchers attempted to reason backwards about what the model was doing based on its results. In at least one case, the software redesigned the entire structural element (an alpha helix) in which the isoleucine it modified was located, for reasons they can’t even guess.

It’s a good reminder that, for now, these software programs are tools: they let us do things that wouldn’t otherwise be possible, but they don’t really help us understand much. We still need to reason through phenomena using the neural networks inside our skull.

This does not necessarily have to be the case; we could put more emphasis on exposing the inner workings of this software during its development in order to gain insight into its decision-making process. But for now, I think the focus has been (quite reasonably) on getting something that works.

An astonishing achievement, but is it useful?

Overall, it’s an amazing piece of work. These proteins must interact with each other, interact with ribosomal RNAs, transfer RNAs, messenger RNAs, growing proteins produced by the ribosome, as well as all normal large subunit proteins. Each of them has had billions of years to develop their ability to work with each other. The fact that we can make such radical changes to the system in a matter of years is nothing short of mind-blowing.