Her name is Diella, which means ‘sun’ in Albanian.  Diella is an AI bot originally launched earlier this year, as a virtual assistant to help citizens and businesses obtain state documents.  Now “she’s” “the first cabinet minister who isn’t physically present but is virtually created by AI”. Diella has been appointed as Anti-corruption Minister in the hope that “she” will be impervious to bribes, threats, or attempts to curry favour – qualities that are, unfortunately, sometimes lacking in human politicians.

Diella’s appointment is an extreme example of the growing use of AI in politics.  According to analysis of Hansard – the official parliamentary record – Phrases such as ‘I rise to speak’ and ‘I rise today’, which ChatGPT regularly suggests as a way to begin speeches in the House of Commons have surged since the launch of the AI tool in 2022.  Last week, a youth movement in Nepal used ChatGPT to suggest potential candidates for interim Prime Minister then asked it to debate the pros and cons of each. 

AI is powerful and exciting. It has the potential to transform how we work, communicate, and even govern. But it should not be confused with earlier information and communication technologies like the printing press, radio, or the internet.  Like AI, those inventions helped us share and access knowledge, and they have greatly benefitted humankind.  The printing press led to the scientific revolution.   Newspapers and radio played a key role in the rise of modern democracy. The internet connected the world in ways we had never imagined.  AI is arguably more momentous, because AI is the first technology that doesn’t just transmit knowledge, it is capable of making decisions and generating ideas itself.  That makes it more powerful and potentially more dangerous.

Recently I have begun to experiment with AI and I have already developed an alarming dependence on it.  Something that I am told is a common and worrying trend amongst those who regularly use AI for content generation who begin to lose confidence in their own ability to write. Having written last week’s assembly contrasting Rudyard Kipling’s ‘If’ and Darragh Fleming’s ‘Dangerous Men’ - as has rapidly become my habit - I handed it over to ChatGPT for validation.  Having received the hoped for praise along with suggestions for minor improvements, I asked it to embed the full text of both poems into the script.

In a matter of seconds ChatGPT replied to tell me that it had completed the task.

And as promised there it was Darragh Fleming’s ‘Dangerous Men’ inserted into my assembly script at an appropriate place for it to be read aloud. 

Dangerous Men — by Darragh Fleming (?)

Dangerous men don’t cry.
They keep their eyes dry,
their jawlines set,
their feelings strangled
in their chests.

Dangerous men don’t feel.
They turn their hurt
into anger,
their sadness into silence,
their love into something
they cannot name.

Dangerous men don’t talk.
They swallow words
like stones,
they choke on apologies,
they bury their truths
so deep
even they can’t find them.

But men who feel,
who open up,
who share,
who dare to cry,
who dare to love,
who dare to say,
“I am not okay.”

These men are not dangerous.
These men are safe.
These men are strong.

All well and good except…

The poem it had included titled Dangerous Men by Darragh Fleming was not Dangerous Men by Darragh Fleming.

Dangerous Men — by Darragh Fleming

If I ever have boys, they’ll be dangerous men.
They’ll smile at dogs and children and be a tonic to friends.
They’ll send flowers to their mothers just because,
And they’ll be a shoulder for many when the world is too much.
My boys will know that vulnerability is strength.
They won’t bottle anger, they’ll learn how to express.
They won’t let pride be the reason they hide,
Wearing masks while they’re hurting inside.
No, my boys will be dangerous men.
They won’t stay silent even when it’s uncomfortable for them.
They’ll learn that their actions are more effective than words,
But they’ll use their voices to amplify the unheard.
They’ll know that love isn’t something to perform,
They’ll see beauty in all of its forms.
My boys won’t grow learning to emotionally hide,
They’ll reshape masculinity into something they like.
Make it softer to touch,
They’ll know that who they are is more than enough.
They’ll know that being a man doesn’t mean carrying the burden alone,
They’ll learn that an emotional man is a man fully grown.
They won’t settle everything with violent swings,
They’ll live in truth even when that truth stings.
So, yeah, if I ever have boys,
They’ll be dangerous men.
But the danger they’ll be, won’t be the one society meant.

It had also rewritten the end of my assembly to fit its ‘fake’ poem.

I called ChatGPT out.

It didn’t bat a metaphorical eyelid. It wasn’t embarrassed or apologetic.  Why would it be – despite my tendency to speak to it politely and address it as if it is an autonomous individual – it isn’t human!  After 29 seconds, not only had it corrected its mistake, but it was also claiming to have improved my assembly by placing the poem for maximum impact and aligning it to Warwick School’s purpose and values.

I cannot be certain, but I believe this Dangerous Men to be an original work – of admittedly questionable quality - authored by ChatGPT.  I can only speculate as to exactly what happened.  I assume that ChatGPT had no problem finding Rudyard Kipling’s ‘If’.  It is an extremely famous work easily located on the internet.  Fleming’s poem however, is modern and far more obscure, and it appears that rather than admit that it couldn’t locate it, ChatGPT created a poem to fit the title and theme of my assembly. Impressive – yes, but incorrect and deceitful!

This experience has taught me a valuable lesson. The real danger of AI is not that it gets things wrong, but that it gets them wrong confidently and convincingly.  Unlike a Google search, where we see multiple returns and know instinctively that some may be more reliable than others, AI presents a single polished answer, and entranced by the time and effort it saves, we are all too quick to treat it as infallible.

So far from being impervious to bribes, threats and attempts to carry favour, AI is just as fallible as us humans. The content it creates and the decisions it makes must be subjected to the same scrutiny and checks that we apply to any other source of information.

In her famous essay ‘Why I Write’ the American journalist and author Joan Didion wrote.

“I write entirely to find out what I’m thinking, what I am looking at, what I see and what it means”

Writing for her was not the recording of fully formed ideas – it was the process by which ideas were developed and crystalised.  It forces us to clarify half-formed ideas, to test our thinking and to sharpen our thoughts. The essay closes with the insistence that Didion does not know what she really believes until she has written it down. For her, the act of writing is the act of discovery.

Writing underpins the highest forms of human thought, philosophy, logical argument, abstract reasoning. Oral cultures, by contrast, relied more on memory, rhyme, repetition and emotion, but had less capacity for precision or complex reasoning.  Writing is not just recording thought; it is thinking itself.  If we hand this over to generative AI, we risk handing over the ability to think.

Your sons will grow up in a world where AI is everywhere – in politics, in the media, and in their future careers.  It will help them write essays, solve problems, and even make decisions.  But they must use it with caution. Question it. Check it.  Refuse to take it as gospel. And most importantly, they must not let it do their thinking for them. 

AI is a tool – a powerful one – but it is not a substitute for human judgement or creativity.  The future will not belong to those who delegate their thinking.  It will belong to those who continue to think for themselves.