General Discussion

The researcher had trained his net with pairs: a bacterial name and the name of an appropriate antibiotic

He then tested his net with various inputs. It often did well. But the interesting result was obtained with a real bacterial name that did not match any of the training names but resembled several: the net eventually made up an antibiotic name that sounded plausible, because it somehow smoothly amalgamated several real antibiotic names

This problem hasn't been solved after decades of research: when the artificial mind encounters a gap in its knowledge, it interpolates in a way indistinguishable from sophomoric bullshitting

I expect the problem to persist