I recently read an interesting article – 'Teaching Poetry in the Age of AI' – published by The Nation. Its author, Lindsay Turner, an associate professor in the department of English at Case Western Reserve University, Cleveland, Ohio, is herself a poet and translator. She makes a significant observation that 'it's [...] becoming increasingly difficult to distinguish between AI-written and human written prose, especially at a moment where language has in many cases become flattened and formulaic across spheres.'

The 'flattening of language' is not a new phenomenon, and in particular it's not primarily an outcome of the relatively recent introduction of 'AI' – although it's very probable that it will be exacerbated as LLMs penetrate into more and more domains of writing. However, my impression is that it doesn't exist independently of the quality of thought. Although professor Turner's chief concern in that article seems to be ethical, she poses two very practical questions – 'Will AI learn to write poetry that is indistinguishable from human-produced poetry? Will my students be able to pass off AI writing as their own? My answer to both of these questions is: I hope not, but probably yes.' But it seems to me entirely probable that, should that be the case, it won't primarily be due to the intrinsic capabilities of 'AI', but to the flattening of thought and language on the part of human authors. 'AI' is (at least for now) primarily trained on human generated texts, so (disregarding for the moment any limitations intrinsic to the technologies) the quality of its output is absolutely limited by the quality of those texts. Should flattened language predominate in them, 'AI' output will replicate that flattening.

So let's look at some obvious if arbitrarily chosen examples of this entirely human-driven flattening, manifested by loss of linguistic specificity (and therefore reduced or eliminated subtlety of thought):

MUTUAL– strictly, the word means reciprocal, but its meaning got muddled in 1864, when Charles Dickens began publishing Our Mutual Friend in serial form. The ultimate result is that 'mutual' is now taken to mean what we used to specify as common, meaning belonging to both of us. So today, if one says 'you and I have a mutual love of coffee' intending to impart that that we both love it, we're actually saying, not only that we both love coffee, but that coffee loves both of us (or indeed at worst that we love each other and coffee is somehow incidentally involved as a third party). What a mess! But Dickens didn't actually invent the misapplication of the word – he simply created the condition for it to occur. The book's title doesn't misapply 'mutual' in that it refers to a triad of friends, each of whom is of course both liked and likes (the basic meaning of 'friend' being of a two-way relationship). Hence Dickens' usage is not logically incorrect, just tautological (he could, without loss of meaning, have simply written 'our friend', the our automatically implying the commonality). But his usage was open to misinterpretation, and that misinterpretation has stuck. Nevertheless, it's still legitimate (and a lot more specific) to say 'mutual' when we mean 'reciprocal' and 'in common' when we mean we share something.

CRISIS – derived from a common Greek root with words such as critic, criticism, critical, criterion, it conceptualises judgement or decision in various ways. So a crisis is properly a point where a decision is made (for example, in a court case the point where a jury decides on guilt or innocence is a crisis, as is the point where the judge decides on sentence). Unfortunately, crisis and critical have lost any connotation of decision-making, now almost universally being used merely to indicate (but only vaguely) a sense of the importance or severity of some event or circumstance, that usage being entirely divorced conceptually from critic, criticism and criterion.

IMPACT – originally indicated crushing, either as a result of some sudden collision as in shell impact or slow pressure as in impacted wisdom tooth. It has, however, come to be used as a generalised descriptor for any influence of anything on anything else, even where verbose circumlocution must be resorted to, as in 'has an impact on' (translation: 'affects') to make it fit. I got so fed up with this a while back that I spent just half an hour with my 1911 Roget's thesaurus, by the end of which I had found around 70 alternatives to 'impact', each of which would, in the appropriate context, have proved more informative of what was specifically being described. But these 70-odd words have almost been lost to the language.

UNIQUE – (from Latin 'unicus' = 'sole' or 'single') specifically means there's only one of something. Consequently, such commonly encountered qualifications as 'rather unique' express only vague notions of something being special, fundamentally negating the meaning of the word. If one means that something is rare or unusual there are plenty of ways to say so that don't contradict themselves, and if something has special characteristics why not say so explicitly?

DECIMATE – (from 'deci' – Latin for ten) originally described a nasty practice prevalent among Roman legions. Where a centurion was dissatisfied with the behaviour or performance of a cohort, he would have every tenth man executed. An exemplary punishment, while specifically minimising the detriment to the fighting strength of the cohort. Now, depending on who is enunciating the word, it can be used to describe anything from moderate damage to total annihilation, in relation, not just to people, but to inanimate objects as well. The US military recently reported that its forces had 'decimated a bridge' in Iran. As there was only one bridge in question, did that mean demolished ( a perfectly respectable and unambiguous word for total destruction) or did it just mean damaged? Or maybe they destroyed precisely one tenth of it?

EVOCATIVE – an adjective derived from the verb to evoke – '1623, to call forth [,] especially to summon up (spirits etc.) by the use of magic' (OED). The verb is essentially transitive, so the derived adjective's common open ended use, as in 'that's very evocative!' demands the question 'of what?' Unfortunately this question typically remains unanswered so the statement is devoid of meaning, other than conveying some vague idea of it being striking in some undefined way.

REGULAR – derived from the same Latin root as regulate and rule, it (not surprisingly) means occuring according to some rule. However it's become almost universally used as a loose substitute for frequent, thereby discarding an important distinction. An event that occurs about 104 times a year could be considered frequent, but it wouldn't be regular unless it occurred, for example, at 12 noon every Tuesday and Thursday. As regular has come to indicate the former uncertain case, how can we possibly express the latter, certain, one?

Such obfuscatory linguistic flattening both results from and exacerbates a flattening of thought – a condition strongly indicated by the increasing substitution in conversation of 'sort of', 'kind of' and 'like' (indicators of uncertainty about what one is saying) for the long established 'err', 'um' and well– (place holders merely indicating 'give me time to think').

But I can suggest at least a partial personal solution – take steps to actively polish your vocabulary. For a start, park the thesaurus and stop 'googling' for meanings, as both merely offer notional substitutes for words without proffering any rationale for their supposed equivalence. In reality, no two words can ever be entirely equivalent. There are no true synonyms as every word has a different history behind it.

Buy a copy of a print edition (yes, on paper!) of the complete Oxford dictionary 1 and use it – often. The current online one will not really serve as you can't see all the words on the broad page spread at one go. So the right brain remains largely passive, failing to make unexpected but potentially interesting associations between nearby words (connections the logical left brain would not make). Nor can you on screen readily flip back and forth between words many pages apart in order to cross reference.

Of course, my 1930s 13 volume edition doesn't contain all the neologisms that have emerged in the intervening 90-odd years, but it provides a solid and highly informative reference to the core of durable English literary usage. And that's the secret key – usage, not just notional synonyms or explanations of narrow current 'meaning'. How other writers have used words in the past is highly informative of their connotations and implications, which are the attributes that bring language to life.

Hardly a day passes without my referring to my complete 'Oxford', and almost every time I do, having checked what I specifically wanted to I find myself just reading the dictionary – following unexpected leads to words and meanings I quite possibly never even knew existed. Many of these are likely to get stuck in the right brain, possibly to 'ring bells' later, quietly enriching the capacity for expression. And anyway it's a fascinating adventure finding out how other writers have used words to best advantage.

 [1] The complete Oxford is probably the only dictionary that will entirely serve, being the only one that documents such detailed etymologies and such extensive examples of prior usage. The two volume Shorter Oxford English Dictionary is in general a lot easier to find second hand, certainly cheaper, and would do at a pinch, but it's a poor substitute for the complete one, as not only does it contain fewer words, but the usage listings are much reduced. Earlier editions of both will probably be cheaper, and as we're not looking for topicality (there being less prior usage to report of more recent coinings anyway, so a lesser dictionary will suffice) it's worth looking out for them. Mine was a bargain some 25 years ago at less than fifty quid for the 13 volume 1933 revised second printing.