Poetry and Artificial Intelligence

Occasionally we host guest posts on this blog. They are often topical, provocative or simply informative. This guest post by Dr. Iain Strachan is all three. As always, your comments and questions are welcome.

Can AI pass the Turing Test today? by Iain Strachan

In a Invisible Poets Road Show in Derby, Steve Wheeler asked me if I thought AI had passed the Turing test. I replied "Yes ... and No." 

I had just read a poem "A chatBot named Christopher" about Alan Turing, where I had claimed that the answer is "No".

However, AI-generated poetry continues to fool us again and again. I have been fooled by it. I once praised someone's Villanelle on Invisible Poets that on closer examination turned out to be AI generated.

Why does this happen? I think it's to do with the way we interact with pop songs. I asked a member of my family "When you listen to a pop song, do you think about the words?" He replied "Not really. If I know the words, I'll sing along to the tune, but I don't think about what they mean."

I expect most people are like that. Pop songs have to be singable, so the lyrics fit the tune, and so we are only engaging with the words on a superficial level. So they need to flow smoothly, have simple rhyme schemes etc.

Human poetry is different. It doesn't always have a smooth iambic pentameter rhythm; for example:

For thou'rt slave to fate, chance, kings and desperate men (John Donne), or
The soil/is bare now, nor can foot feel, being shod (Gerard Manley Hopkins).

See how the strong syllables pile up with no light syllables between. The Hopkins also has enjambment, where the sense carries on over the line break. Song lyrics don't do this - normally each line stands alone.

It's the same with AI generated poetry. It is polished, and flows nicely; each line is self-contained. But, whereas with a pop song, you can find depth and meaning in the lyrics: a story told, or a telling metaphor, if you examine an AI poem carefully, you won't find any depth; it falls apart as a sequence of poetic sounding phrases and clichés strung together with no clear overall message.

So if you find a poem that seems super smooth and polished, take a closer look before you enthuse about it. Don't give the AI fakers their serotonin boost! If it's AI, it will fall apart and you'll find the words of my chatBot poem to be still true:

Chatbots today can't pass the Turing Test
Their show of understanding's fake, at best.

Iain Strachan

Image used under a Creative Commons License

Why AI Can't be Creative

From time to time, we invite members to write guest blog posts on topical issues that impact upon poetry. Here is a fascinating article on AI and poetry/creative writing by our very own Iain Strachan, formerly a computer scientist and always a poet!

One of the keys to original creative work is that it should be.... well, original. It should contain phrases that haven't been seen before - that take us by surprise and delight us. However, a program like ChatGPT will struggle to produce anything like this, because it can only really produce output that is resembles the text and poems that have been used to train it. 

I asked ChatGPT to provide ten possible completions to this line, taken from Gerard Manley Hopkins's poem "God's Grandeur" 

    It will flame out like ... 

Here is what it came up with: 

a comet streaking through the night. 

a candle in the wind. 

a phoenix rising from the ashes. 

the last rays of a dying sun. 

fireworks bursting in the sky. 

a shooting star's brief flight. 

the embers of a forgotten fire. 

lightning splitting the heavens. 

a spark in the dark, then vanish. 

the final flare of a burning match. 

These phrases are all well-known, and not exactly unexpected; they are bland and literal. Compare that with the completion in Hopkins's poem. If you don't already know the poem this will be completely unexpected: It will flame out like shining from shook foil. 

Why is it that chatGPT can't produce stunningly original completions like that? The reason lies in statistics. In generating continuations, it uses probability, and a random number generator that is tuned (like loaded dice) to be heavily biased towards more likely (and hence less surprising) continuations. Also, if it chose an unlikely continuation, it probably won't make sense at all. There are vastly more nonsensical unexpected continuations than ones that make sense. 

What can we learn from this and apply to our own poems? A lot of our writing can be automatic, sometimes creative, but sometimes commonplace. So next time you write a poem, go through the first draft and examine each word or phrase and see if you can make a better one, instead of the expected word, make it unexpected. Make it a metaphor, rather than literal. 

Here's an example from one of my poems, describing the grid system of roads in Washington DC:

Geometric city. Anonymous streets 

Labelled with letters and numbers 

????? the shops into blocks. 

Here are chatGPT's 10 boring suggestions: 

Neatly dividing 

Methodically carving 

Carefully segmenting 

Precisely arranging 

Artfully separating 

Rigorously partitioning 

Strategically placing 

Systematically organizing 

Elegantly splitting 

Thoughtfully grouping 

Here's mine: Cheese-wire. 

I'll bet you didn't see that coming! It was based not on poems I'd seen before but lived experience; how as a kid I was fascinated to see cheese cut up with a cheese-wire.

Iain Strachan

Image from Wikimedia Commons