HMRC will soon begin artificial intelligence (AI) trials in some of its processes, starting with customer contact and casework, according to the man responsible for the Revenue’s digital transformation
Speaking at the Public Sector Show in London, HMRC’s interim chief digital and transformation officer Mike Potter stated the plans were being conducted with the ultimate aim to “take the graft out of people’s jobs” so they could “focus on higher value work”.
The trials will be launched “in the near future” as part of a move towards automation, with the Revenue initially looking at using AI to manage routine processes.
Potter’s presentation provided little information about the tests, stating that there was “a need to talk with staff” before making more details public.
However, he stated that the tax authority is exploring three areas it believes are viable to explore AI options, and a slide in his presentation included the following early use cases:
- contact handling: AI to direct people to the right places or information without human intervention;
- casework, where AI would be used to enhance decision-making;
- and AI prompts to assist taxpayers with effective self-service.
Such trials are part of a government-wide push to use data more intelligently and to supplement its workforce and processes with machine learning.
According to UKAuthority, HMRC is the latest department to trial AI, with the DWP currently doing proof of concept work in automating processes and the Department for Culture, Media and Sport drawing up plans for research and development into AI as part of the national digital strategy.
‘No such thing as AI – yet’
Nichola Ross Martin, managing director of Ross Martin Tax, has been studying the use of machine learning and AI in tax, and while the benefits of adoption for the Revenue are obvious in terms of streamlining processes at a basic level, she believes there are fundamental problems which need to be addressed before the approach is rolled out further afield.
“At the moment HMRC is just trying to automate their helplines – they’re just creating help bots,” Ross Martin told AccountingWEB. “If you can do a flowchart of something you can make a bot for it, but you’ve got to know where you’re starting – what the question is.
“The main problem with AI is making sure it understands the question, and the question can be posed in so many different ways. In tax cases, we say every case is determined by its individual facts and circumstances, and unfortunately that means you’ll have to program a line on each bit of your code to deal with each circumstance. You’re left wondering what possible questions can be asked, and that’s the problem the Revenue has got.”
While the mainstream media write excitedly about computers beating chess grandmasters, Ross Martin believes that this is not AI as we understand it.
“There’s no such thing as Artificial Intelligence yet,” she said. “It’s called that, but it’s not intelligence as we understand it. It’s currently about how we use bots. All they’ve [chess-playing computers] done is memorised vast numbers of moves – there are only so many. In chess you never get someone asking ‘can I describe to you the shape of my queen, and will it have an impact on my move?’ That doesn’t matter in chess because a queen is a queen, but it might affect your result in tax.”
While computer science is undergoing massive change, and in five years’ time we could have something resembling artificial intelligence available, Ross Martin believes there are also systematic issues around how this new technology is applied to tax if and when it arrives.
“Our legal system makes things very difficult”, she said. “We have judge-made law, and that does not translate very well into coding.”
Ross Martin believes that if the tax authority is serious about AI adoption, the tax system needs to be simplified in a very linear way to adapt to it.
“The way I would approach tax law is to say ‘this is what we can do really easily, how do we make rules that fit into this?’” said she said. “It might be that a lot of our rules are not actually fit for purpose.
“We’ve got some really basic rules where this has happened. A very good example is what happened with the statutory residence test, where we had a collection of hideous rules and the Revenue then managed to change them into quite a decent set of rules. There are still massive problems with it, but it’s a lot better than it was and we could actually do that to a lot of taxes.”