The chief secretary to the Treasury has called the Sky News-Chat GPT spending review projection “pretty good” and scored it 70%.
Darren Jones compared the real spending review, delivered by Rachel Reeves on Wednesday, and the Sky News AI (artificial intelligence) projection last week.
Sky News took the Treasury’s spring statement, past spending reviews, the ‘main estimates’ from the Treasury website, and the Institute for Fiscal Studies’ projections, and put them into ChatGPT, asking it to calculate the winners and losers in the spending review.
Politics Hub: Tap here for live updates
This was done 10 days ahead of the review – before several departments had agreed their budgets with the Treasury – on the basis of projections based on those public documents. It also comes amid a big debate kicked off by Sky News about the level of error of AI.
The Sky News-AI projection correctly put defence and health as the biggest winners, the Foreign Office as the biggest loser, and identified many departments would lose out in real terms overall.
It suggested the education budget would be smaller than it turned out, but correctly highlighted the challenges for departments like the Home Office and environment.
Watch what happened with Sky’s AI-generated spending review
Please use Chrome browser for a more accessible video player
AI writes the spending review
Reviewing the exercise, the author of the real spending review told Sky News that this pioneering use of AI was “pretty, pretty good”.
He added: “I could be out of a job next time in 2027, which to be honest, it’s not a bad idea given the process I’ve just had to go through.”
The Treasury made a number of accounting changes to so-called “mega projects” which AI could not have anticipated, and changed some of the numbers.
Please use Chrome browser for a more accessible video player
Sky’s economics editor Ed Conway takes a look at the key takeaways from chancellor Rachel Reeves’ spending review.
Asked to give it a score, Mr Jones replied: “I’m going to give it 70%.”
The spending review includes AI as a tool to save money in various government processes.
Asked if 70% accuracy is good enough for government, he replied: “Well we’re not using your AI. We’ve got our own AI, which is called HMT GPT, and it helps us pull together all the information across government to be able to make better, evidence-informed decisions.”
