Two tech-related things made me laugh this week. One was Donald Trump’s childlike exuberance at seeing the dash panel of a Tesla on the White House lawn, and his wondrous exclamation that “everything is computer”.
The other was equally hilarious, also tied to politics. Keir Starmer stood up yesterday in Hull and said waste would be thrown by the wayside and the civil service would lose its bloat … thanks to the transformative effects of AI.
What I knew, and no one else did until my story for New Scientist was published shortly afterwards, was that the prime minister was talking nonsense. I had seen the government’s AI-powered revolution first hand. And it was all hooey.
With what is believed to be a world first use of freedom of information (FoI) laws for this purpose, I had obtained the interactions Peter Kyle, the UK technology secretary, had with ChatGPT for his job. (The department Kyle heads had rejected an earlier FoI request because granting it ran the risk of unveiling his own personal conversations with ChatGPT.) If Kyle’s chats are any indication of what the AI revolution promises in government, then we’re in for a tough time.
My knowledge of the UK’s podcast scene is what you would expect as a middle-aged white man who spends too much time on the internet – so even I could answer the question Kyle first posed to the chatbot: which podcasts should he appear on to reach a wide audience.
What Kyle asked next – which of the four podcasts ChatGPT had recommended had the most listeners – might have seemed like a sensible use of the generative AI tool, but it was also something that a quick Google could have solved in much the same time; and with fewer concerns that it was making up numbers like an overconfident Oxbridge graduate. AI “hallucinations” – the submission of false, incorrect or misleading results – remain a real and present fear with generative AI tools, and are alarmingly common. So too is the idea of mixing things up a little: recent BBC research showed AI tools such as ChatGPT got major things wrong when summarising its reporting.
As to the tech secretary’s query about why AI adoption is so slow in British small- and medium-sized businesses (SMBs), that was equally unenlightening. It isn’t a novel or original take to say that data protection laws, such as general data protection regulations (GDPR), may stymie the development of a technology that is subject to multiple lawsuits about data breaches and copyright risks. Nor – you would hope – is it beyond the civil service’s ken to understand that “many SMBs are unaware of these programs or find them difficult to navigate”, as ChatGPT solemnly told Kyle in point seven of a plodding 10-point treatise. “Not reassuring that a minister with a department of experts is using AI for these sorts of questions,” as one critic put it.
ChatGPT was stronger at answering Kyle’s questions seeking simplified analogies for complicated concepts such as “quantum” or “antimatter” – though, as some social media commentators have pointed out, it’s a little worrying that Kyle felt the need to ask an AI chatbot for a definition of “digital inclusion”. Presumably he should know, as the person responsible for overseeing it?
I was generally surprised both by the comparative lack of interaction Kyle had had with ChatGPT, given he’s spent the past several months telling anyone with an audio recorder about his hankering for the technology, and by the sorts of questions he was asking. In fact, my response on seeing the interaction between AI and the minister at the sharp end of the government’s use and knowledge of it was: is this it? Is this the generative AI revolution we’re banking our future and our economy on?
There are ways to harness AI in government for the public good. Churning through complicated data to unpick patterns that the human eye misses, for example; AI-powered drug discovery; deep research of the type that can add insights long forgotten or never previously thought about – these are all attainable wins.
Quite how the government’s talk of AI’s promise and potential chimes with the reality of how it is used now is something the tech secretary will have to answer for himself. Perhaps he can ask ChatGPT for some help there, too.
