I was recently very troubled to hear that the general public prefers AI imitations of Shakespeare, Chaucer, Byron, and others to the real thing. A study had produced results which seemed fairly conclusive, with over three-quarters of the 700 or so participants consistently considering the AI imitations to be more “beautiful” than the real deal. Few of them could identify whether they were reading the original or the simulacrum.

I am quite philosophically confident that Large Language Models aren’t and will never be truly intelligent, and therefore unable to actually “express themselves”. Nonetheless, I did fear that perhaps we had hit a point where AI could sufficiently flood the aesthetic zone, so increase the noise-to-signal ratio, that we could no longer trust our judgement, no longer tell the difference between sincere artistic expression and fakery. These programs are marvellous things, and only likely to expand their capability – though there is some concern that the very success of LLMs is damaging the learning sets involved.

I was quickly relieved, however, upon reading the Telegraph’s sampling of the “poetry” generated by AI. Indeed, I was left seriously doubting both the taste of the general public and the future ability of generative AI to produce any serious imitation of good poetry. There are two key reasons for this.

Firstly, the “poetry” itself is execrable garbage at both the technical and expressive levels. I am not impressed that the AI managed to turn out something which looks, on the page, like a poem; that seems to be the most basic expectation of something using so much power and given so much attention. Nevertheless, this stuff is genuinely dreadful. The sample from Byron goes so far as to simply steal lines from Byron himself — a rather literal “imitation”! The Shakespeare imitation begins:

When first I saw thee, fair and gentle maid,
My heart did skip a beat, my breath did stall;
Thy beauty struck me like a sweet cascade,
And in that moment, love did me enthrall.

It would be generous to compare this to Valentine’s Day greeting card verse. The AI has obviously understood from its Renaissance poetry set that verbs may be transposed and made passive at the end of the sentence, and so does this in three of the first four lines. The AI inserts an auxiliary verb (“did”) thrice, surely to pad out the line — in this at least we may admit it comes off as human! That its imagery is hackneyed — “fair and gentle maid” — is not really a criticism in the way we expect — it can only reproduce what it sees, and boil down to the quintessence what is meant, and there are indeed many fair and gentle maids out there in poetry, and a good thing too. But this is a point I shall return to in a moment.

Compare Shakespeare’s first four lines:

So am I as the rich whose blessèd key
Can bring him to his sweet up-lockèd treasure,
The which he will not ev’ry hour survey,
For blunting the fine point of seldom pleasure.

One transposed verb, no auxiliary verbs. And look at the imagery, 430 years later! That a love may be like a treasure is no wonder; but the sweet voluptuous rhythm, the wonder of the poet who finds himself rich, the carefully chosen adjectives — “sweet, up-lockéd” — all this rises from the page, undimmed by centuries.

I have mentioned that part of the AI’s problem is that it can only work with its set — but this highlights the other, much more serious problem, fatal I think to any hope that it can truly create poetry. Intellect is not simply a matter of processing information and rendering an output or decision — that is the error of the greatest advocates for AI. There is, in art, what Eliot calls the historical sense:

Tradition is a matter of much wider significance. It cannot be inherited, and if you want it you must obtain it by great labour. It involves, in the first place, the historical sense. . .and the historical sense involves a perception, not only of the pastness of the past, but of its presence; the historical sense compels a man to write not merely with his own generation in his bones, but with a feeling that the whole of the literature of Europe from Homer and within it the whole of the literature of his own country has a simultaneous existence and composes a simultaneous order. This historical sense, which is a sense of the timeless as well as of the temporal and of the timeless and of the temporal together, is what makes a writer traditional. And it is at the same time what makes a writer most acutely conscious of his place in time, of his own contemporaneity.

AI has merely inherited its material, but it can have no sense of the ways in which the cultures of Modern England and Ancient Greece are both simultaneous and distant; it has no European bones to feel with. It can imitate Shakespeare’s language, more or less, but it cannot touch his spirit, because it has no spirit — and it is with the spirit, really, that we gasp at the sensual rhythms of the Sonnets, and recognize in them our own experiences of and desires for love. The transcendent experience of the writing and reading of poetry is quite beyond AI. 

AI cannot make meaningful decisions about what is “better language”

Further, this is why AI cannot make meaningful decisions about what is “better language”. We have no issue with fair and gentle maids, let me repeat — may there be one for every noble swain! — but the very phrase came into the world hackneyed and dead. The poet has the duty to (thus Eliot again) “purify the dialect of the tribe”. This means knowing now, in this moment — as a man at a given time, but a man set in a great tradition, earned at great cost — how to speak, how to express timeless things in time. Fresh metaphors must be found, and new counterpoints, all echoing everything good about the old. And the thing about the new purifications of the dialect, when they work, is that they stay fresh ever-after — Shakespeare and Wordsworth and Eliot remain aureate in my mind, whereas I have already, and gratefully, forgotten ShAIkespeare.

Finally, someone might ask: if the AI poetry is so bad, why did such a large sample prefer it to Shakespeare? Because, first, taste like tradition is something one must work to acquire, insofar as taste is the response to living tradition; and second, because, in this way like AI, humans can only understand and use what they have already learned, and learned properly, and having abandoned poetry as a civilization, we cannot be expected to know it when we see it.