Much of the fundamental mathematical work of artificial intelligence models was actually contemporaneous with Dino Buzzati, in the 1940s to 1960s. Although no-one was quite sure yet of the startling effectiveness of such models at scale, when the perceptron (a single layer neural network) was invented in 1958, it was already claimed to be the embryo of a computer that in the future would be “able to walk, talk, see, write, reproduce itself and be conscious of its existence.” [See link below] [1]

Endriade and the scientists in Buzzati’s novel build their artificial intelligence without language. “It doesn’t know any languages. Language is the worst enemy of mental clarity. In his desire to express his thought in words at all costs, man has ended up making such messes.” 

The novel mentions that Laura has “a soul,” called “the egg” – her, or its, personality and consciousness emanate from it. But we cannot isolate which parts of even our current non-conscious models – which attention heads, which single locality – contains specific capabilities or attributes or personalities. They are all distributed, and like in our brains, one neuron can be overloaded with many simultaneous uses. I cannot say if the soul is a capability or trait, but I should think that it is much more complex than what we think of as capabilities and traits in models that we currently have. The ability to write a line in iambic pentameter should not be as complex as having a soul.