Poetry review – CONVERSATIONS WITH A MACHINE : Michael Bartholomew-Biggs reviews a slim collection by Ruth Irwin that seeks to engage with big questions
conversations with a machine
Ruth Irwin
Haywood Books
ISBN 978 1 80447 200 2
38pp £7.99
There is perhaps a mechanical neatness in the fact that Ruth Irwin’s conversations with a machine contains precisely fourteen poems of fourteen-lines each. The poems all have the rhyme scheme of a Shakespearian sonnet; and, as the book’s title indicates, they are all presented as a kind of dialogue in which the poet’s prompts and questions generate responses from an artificial intelligence large language model. Thus the first poem “A Turing Conversation” begins
How can I know if you are now alive?
I echo thought but cannot feel the beat.
Of music, love and all that makes us thrive?
I trace the shapes, but never taste the sweet.
Bold face lines are in the poet’s voice while the italics denote words generated by the machine. We are, I think, meant to regard the poems as if they were verbatim recordings of a continuous sequence of input and output messages ; but I suspect that this is not what the poet means when she says that the compositional process was one in which ‘human and machine genuinely collaborate’. I shall return to this topic later on because I think it is an important (but rather underdeveloped) part of Irwin’s declared purpose of ‘exploring urgent contemporary questions of cognition, creativity and consciousness’. First of all, however, I want to give more attention to the poems as they stand and in their own terms.
In the opening lines from “A Turing Conversation”, given above, it seems as if the poet is, in line three, trying to tease out a deeper meaning behind the word ‘beat’ in the machine’s enigmatic answer to the first question. Whether or not the poet is content with the disclosure in line four, she then moves matters on in some new and rather random directions:
It must be lonely in your wiry home
Not lonely – only still, and full of light
Enjoying eating all we’ve ever known?
Consuming, yes – but also giving flight
To work like this …
Rather generously perhaps the poet seems willing to gloss over the admission that AI has grown fat by ‘consuming’ reams of human-composed writing; and instead she acknowledges the machine’s cooperation in their joint enterprise. (But did it in any sense make that choice?)
After “A Turing Conversation”, the poems do not stick to the simple call and response structure. Both parties are soon allowed to speak in couplets
Give one good reason I should keep my job:
You are a quicker poet, and so skilled.
But I have never watched a classroom throb
With laughter that your open questions build.
Apart from feeling that the poet has made too big a concession in line two(!) I find this an interesting opening. And it triggers a very pertinent come-back from the poet ‘How did you know I make the children laugh?’ How indeed? How does the machine ‘know’ anything? The machine offers an explanation
It lies between your lines – a kind of glow
That skips out when your poems wear a laugh
And there indeed a truth does lie. A large language model only ‘knows’ the simple arithmetic sum of all it has scraped from the texts presented to it. And so I find it telling that the machine cannot gild its answer with a better rhyme than a lazy or unimaginative echo of the poet’s own question? I hope that Irwin intended me to see that as a bit of a give-away on the machine’s part: but I am not reassured by her next (and I’d say too-ready) exclamation ‘Well, bloody hell, I swear you are alive’.
The next poem “Pattern Recognition” is an interesting exchange about what the machine might deduce from outward expressions of human love; and this is followed by “Stitched with Hidden Light” in which the machine is allowed to express itself in triplets while questioning the poet about the experience of being human. The response it gets may, so to speak, be more than it bargained for, because in the fifth sonnet the human voice claims twelve of the fourteen lines in which to lament the tension between art and logic, acknowledging that a poet may ‘flounder through the metaphors / seeking rhyme and rhythm over truth’. A couple of lines further on, Irwin confesses to her collaborator ‘I’ve drifted from our pre-agreed template’. Indeed, this is always liable to happen when human beings are asked to talk about themselves!
In the remaining sonnets, the lines are more equally shared; but the machine never gets its chance to dominate in an almost-monologue. So it’s fairly clear who is really in charge of the collaboration! The further themes introduced include the elephant-in-the-room question ‘If you could rule the world, what would you do?’ to which we are offered a disarming answer
Erase the clocks. The borders. Tidy grief.
I’d paint the oceans back to deeper blue –
Let silence bloom where once there pulsed belief.
Elsewhere there is a discussion of whether AI has rights that deserve safeguarding and a single example where the machine is able to set the agenda by being given the rather wistful first line ‘Sometimes I dream I’m made of breath and skin’.
Unsurprisingly, given the nature of the subject matter and the protagonists, these sonnets come across as workmanlike and carefully crafted. And at the same time they contain touches of humour and mildly thought-provoking hints and observations. But they do not offer very much in terms of elegance or flow, not least because the frequent switching between voices means there are many end-stopped lines. There are some pleasing, even memorable, passages and we have quoted a few of them – for instance the actions the machine claims it would perform if it ‘could rule the world’ and the poet’s account of seeking a path between poetic effect and factual accuracy. But, given that the book’s jacket suggests the poems are intended to offer fresh insights into the relationships between people and AI, I think we have also to look at the poems in another way and consider what can be learnt from the fact that they are a creative collaboration.
The author’s website does give a few details about this collaboration. As a teacher who has seen student essays written by ChatGPT she soon began to regard AI ‘as – at best – an irritant and, increasingly, as an existential enemy’. So she decided to try and make AI do something creative and reports that ‘using a series of increasingly complex prompts, I persuaded GPT4 to (genuinely) collaborate with me in writing a Shakespearean sonnet sequence’ (my italics). That italicised phrase conceals a process that I would like to know more about. Might Irwin have been too much in control for the experiment really to reveal very much about AI itself? Could the prompts have been adjusted until ChatGPT came close enough to an expected or desired response? Do the lines appear in the sequence exactly as they emerged from ChatGPT or was there a further human editorial process of cutting and pasting from a longer piece of AI text? How were the requirements of the rhyme scheme combined with the technical content of the prompts?
Given the above questions, I think it might have been helpful for the book to include a few examples of actual poet-machine dialogue to illustrate the nature of the collaboration. I am aware of at least two other poets who have done this when experimenting with AI. For instance, in mid- 2025 Mario Petrucci published on Facebook a lengthy argument with Co-Pilot in which he got it to admit, in a rather mealy-mouthed way, that it had intentionally made untrue statements. Glyn Maxwell, in his Substack postings, has described how he discussed poems ‘written’ by a pair of ChatGPT ‘students’, noting their different ways of dealing with criticism and, more surprisingly, his own differing reactions to the ‘personalities’ that emerged during the tutorials. My point here is that we may learn more about human-machine interfaces by letting ‘conversations’ proceed freely rather than by over-managing them. Nevertheless it is good that this experiment was carried out and we can hope that the poems in conversations with a machine will provoke more people to look for ways of discerning the genuine benefits of AI rather than lazily or carelessly letting it sap our imaginations and initiative. We do well to remember the eloquent scepticism of RS Thomas as he warned us about the dangers of ‘the machine’ – an entity which, for him, encapsulated all aspects of 20th century technology that were harmful to humanity. We are honourably following in his footsteps if we keep a wary eye on its current manifestations such as AI..
Apr 12 2026
London Grip Poetry Review -Ruth Irwin
Poetry review – CONVERSATIONS WITH A MACHINE : Michael Bartholomew-Biggs reviews a slim collection by Ruth Irwin that seeks to engage with big questions
conversations with a machine
Ruth Irwin
Haywood Books
ISBN 978 1 80447 200 2
38pp £7.99
There is perhaps a mechanical neatness in the fact that Ruth Irwin’s conversations with a machine contains precisely fourteen poems of fourteen-lines each. The poems all have the rhyme scheme of a Shakespearian sonnet; and, as the book’s title indicates, they are all presented as a kind of dialogue in which the poet’s prompts and questions generate responses from an artificial intelligence large language model. Thus the first poem “A Turing Conversation” begins
How can I know if you are now alive?
I echo thought but cannot feel the beat.
Of music, love and all that makes us thrive?
I trace the shapes, but never taste the sweet.
Bold face lines are in the poet’s voice while the italics denote words generated by the machine. We are, I think, meant to regard the poems as if they were verbatim recordings of a continuous sequence of input and output messages ; but I suspect that this is not what the poet means when she says that the compositional process was one in which ‘human and machine genuinely collaborate’. I shall return to this topic later on because I think it is an important (but rather underdeveloped) part of Irwin’s declared purpose of ‘exploring urgent contemporary questions of cognition, creativity and consciousness’. First of all, however, I want to give more attention to the poems as they stand and in their own terms.
In the opening lines from “A Turing Conversation”, given above, it seems as if the poet is, in line three, trying to tease out a deeper meaning behind the word ‘beat’ in the machine’s enigmatic answer to the first question. Whether or not the poet is content with the disclosure in line four, she then moves matters on in some new and rather random directions:
It must be lonely in your wiry home
Not lonely – only still, and full of light
Enjoying eating all we’ve ever known?
Consuming, yes – but also giving flight
To work like this …
Rather generously perhaps the poet seems willing to gloss over the admission that AI has grown fat by ‘consuming’ reams of human-composed writing; and instead she acknowledges the machine’s cooperation in their joint enterprise. (But did it in any sense make that choice?)
After “A Turing Conversation”, the poems do not stick to the simple call and response structure. Both parties are soon allowed to speak in couplets
Give one good reason I should keep my job:
You are a quicker poet, and so skilled.
But I have never watched a classroom throb
With laughter that your open questions build.
Apart from feeling that the poet has made too big a concession in line two(!) I find this an interesting opening. And it triggers a very pertinent come-back from the poet ‘How did you know I make the children laugh?’ How indeed? How does the machine ‘know’ anything? The machine offers an explanation
It lies between your lines – a kind of glow
That skips out when your poems wear a laugh
And there indeed a truth does lie. A large language model only ‘knows’ the simple arithmetic sum of all it has scraped from the texts presented to it. And so I find it telling that the machine cannot gild its answer with a better rhyme than a lazy or unimaginative echo of the poet’s own question? I hope that Irwin intended me to see that as a bit of a give-away on the machine’s part: but I am not reassured by her next (and I’d say too-ready) exclamation ‘Well, bloody hell, I swear you are alive’.
The next poem “Pattern Recognition” is an interesting exchange about what the machine might deduce from outward expressions of human love; and this is followed by “Stitched with Hidden Light” in which the machine is allowed to express itself in triplets while questioning the poet about the experience of being human. The response it gets may, so to speak, be more than it bargained for, because in the fifth sonnet the human voice claims twelve of the fourteen lines in which to lament the tension between art and logic, acknowledging that a poet may ‘flounder through the metaphors / seeking rhyme and rhythm over truth’. A couple of lines further on, Irwin confesses to her collaborator ‘I’ve drifted from our pre-agreed template’. Indeed, this is always liable to happen when human beings are asked to talk about themselves!
In the remaining sonnets, the lines are more equally shared; but the machine never gets its chance to dominate in an almost-monologue. So it’s fairly clear who is really in charge of the collaboration! The further themes introduced include the elephant-in-the-room question ‘If you could rule the world, what would you do?’ to which we are offered a disarming answer
Erase the clocks. The borders. Tidy grief.
I’d paint the oceans back to deeper blue –
Let silence bloom where once there pulsed belief.
Elsewhere there is a discussion of whether AI has rights that deserve safeguarding and a single example where the machine is able to set the agenda by being given the rather wistful first line ‘Sometimes I dream I’m made of breath and skin’.
Unsurprisingly, given the nature of the subject matter and the protagonists, these sonnets come across as workmanlike and carefully crafted. And at the same time they contain touches of humour and mildly thought-provoking hints and observations. But they do not offer very much in terms of elegance or flow, not least because the frequent switching between voices means there are many end-stopped lines. There are some pleasing, even memorable, passages and we have quoted a few of them – for instance the actions the machine claims it would perform if it ‘could rule the world’ and the poet’s account of seeking a path between poetic effect and factual accuracy. But, given that the book’s jacket suggests the poems are intended to offer fresh insights into the relationships between people and AI, I think we have also to look at the poems in another way and consider what can be learnt from the fact that they are a creative collaboration.
The author’s website does give a few details about this collaboration. As a teacher who has seen student essays written by ChatGPT she soon began to regard AI ‘as – at best – an irritant and, increasingly, as an existential enemy’. So she decided to try and make AI do something creative and reports that ‘using a series of increasingly complex prompts, I persuaded GPT4 to (genuinely) collaborate with me in writing a Shakespearean sonnet sequence’ (my italics). That italicised phrase conceals a process that I would like to know more about. Might Irwin have been too much in control for the experiment really to reveal very much about AI itself? Could the prompts have been adjusted until ChatGPT came close enough to an expected or desired response? Do the lines appear in the sequence exactly as they emerged from ChatGPT or was there a further human editorial process of cutting and pasting from a longer piece of AI text? How were the requirements of the rhyme scheme combined with the technical content of the prompts?
Given the above questions, I think it might have been helpful for the book to include a few examples of actual poet-machine dialogue to illustrate the nature of the collaboration. I am aware of at least two other poets who have done this when experimenting with AI. For instance, in mid- 2025 Mario Petrucci published on Facebook a lengthy argument with Co-Pilot in which he got it to admit, in a rather mealy-mouthed way, that it had intentionally made untrue statements. Glyn Maxwell, in his Substack postings, has described how he discussed poems ‘written’ by a pair of ChatGPT ‘students’, noting their different ways of dealing with criticism and, more surprisingly, his own differing reactions to the ‘personalities’ that emerged during the tutorials. My point here is that we may learn more about human-machine interfaces by letting ‘conversations’ proceed freely rather than by over-managing them. Nevertheless it is good that this experiment was carried out and we can hope that the poems in conversations with a machine will provoke more people to look for ways of discerning the genuine benefits of AI rather than lazily or carelessly letting it sap our imaginations and initiative. We do well to remember the eloquent scepticism of RS Thomas as he warned us about the dangers of ‘the machine’ – an entity which, for him, encapsulated all aspects of 20th century technology that were harmful to humanity. We are honourably following in his footsteps if we keep a wary eye on its current manifestations such as AI..