CTA Address 2024: How will AI affect tax professionals and the tax system?

The event on Wednesday, June 5, was chaired by CIOT President Charlotte Barbour, with speaker Conrad Young, former Chief Digital Officer at Deloitte' and Chair of the Advisory Board at the Oxford Internet Institute. The other panellists, and respondents to the address, were:

• Bivek Sharma, Chief Technology Officer for PwC UK
• Shan Sun, Tax Technology lead at Deliveroo

Conrad opened by reassuring the audience that none of his slides had been produced by ChatGPT. He said AI is a “difficult topic” for tax professionals, and the sheer volume of AI programs can be “overwhelming” for many.

He began his remarks with a brief guide to the history of AI. It began back in the 1950s with research at Dartmouth College and has progressed to more recent developments including the launch of programmes like ChatGPT, he explained.

Conrad used as an example how the latest iteration of AI, machine learning, can be taught to distinguish between cats and dogs. However, he warned that AI is still not perfect, with concerns about bias, privacy, transparency or simply “making things up”, though solutions to these issues continue to develop.

In the world of tax, Conrad said revenue authorities are looking to:

• Acquire more data from taxpayers and intermediaries
• Use AI to analyse and scrutinise that data
• Use digital channels to transform interactions with taxpayers

As an example, he said French authorities used AI to uncover 140,000 undeclared swimming pools in the country in 2023, amounting to €40 million in tax. On the other hand, the Dutch government resigned after 20,000 families were wrongly accused of child benefit fraud, in part due to flawed investigations aided by AI.

Many tax and finance firms, including the Big Four, are already utilising AI, while taxpayers sometimes also rely on AI programs themselves. Conrad said the three main types of AI - rules based, discriminative machine learning and generative machine learning - are well-suited to different elements of work in tax.

Looking to the future, Conrad said AI can offer automation, augmentation of human processes or transformation, where the change is so dramatic that it allows a completely new approach. Referring to the Forum of Tax Administration’s “Tax Administration 3.0”, he said there are areas where AI can help employ tax rules underpinned by “natural systems”, such as online marketplaces or payment providers. In these cases, all the data is recorded and stored, and therefore the ecosystem is in place for tax considerations including creating VAT returns. This can also help cut down on tax returns with “pre-filled returns”, or even accelerate tax payments via “split payments”, where levies such as VAT are withheld in the payment channel.

Conrad said that AI can improve two main areas of the tax system:

• Policy: AI could increase the tax take through reducing error and fraud and delivering a cash flow benefit. It can also allow better overall management of the economy and make targeted policy measures more effective.
• Administration: In this area, AI can improve efficiency and effectiveness, as well as governance and accountability. Conrad suggested AI can be used to convert “rules as code”, where tax guidelines are more “machine readable”.

Conrad finally explored the impact of AI on the work of tax professionals. He reiterated that automation and augmentation can help with workloads, especially large data sets. However, he warned that these large data sets often need “very granular” analysis, and raised the question of whether AI could affect human jobs. Furthermore, the rise of large data sets could actually increase workload rather than reduce it.

Within the tax services market, an influx of AI providers could affect the status-quo, with taxpayers moving away from established bodies to new services, while more widely within the tax profession, AI could be used for training or skills. Conrad warned that AI could “be the end” of billing “by the hour” and other commercial models.

Moving on to the panel discussion, Bivek Sharma said “AI is coming” but it is moving so quickly that businesses need to take a look at where it will end up in the future, rather than how it is now. He said Generative Pre-trained Transformer 4 (GPT-4), was released last year and its successor, GPT-5, is coming soon, despite us not yet utilising the full capabilities of 4.

Bivek said he knows one tax professional who initially thought AI “couldn’t possible do what I do” but who is now an ardent user. For example, he asked GPT for the five strongest arguments to use with HMRC in respect of a particular case, then asked what HMRC would say, and finally what the chances of winning the arguments were. We are becoming “augmented advisers”, he concluded.

Shan Sun said ChatGPT had to be created as a chatbot to increase greater public interest, so that users could feel they were “having a chat” with AI.

Shan said taxpayers would remain “ultimately responsible” for their own tax returns, adding: “I can’t really imagine a tax system coming in to play with no human intervention whatsoever.”

She said machine learning will always have errors and we must allow for that. In fact, human users at some point in the process need to know the correct answers to judge how accurate an AI system is.

Shan said AI is still a “product that is not mature enough”, concluding that while many people are worried about its effect on the job market, there is the opportunity of “creating something new” and generating more jobs.

Addressing a question from the audience on how AI could change the way we design taxes, Conrad said it will have a “massive impact”. However, he added that other elements of tax design should also be considered, such as incentivising certain behaviour or managing the economy.

From the floor CIOT council member John Barnett asked how AI can “feed back” into a ‘broken’ and ‘badly drafted’ tax system, likening it to the use of VAR in football where “it gets a bad press but a lot of the things people criticise about VAR are not VAR” -  it’s that the rules are wrong. Shan said AI can help find loopholes when designing taxes but warned about teaching AI rules which some feel “don’t make sense”. “There’s the phrase ‘garbage in, garbage out’,” she added.

On safeguards for AI, Bivek warned about “hallucinations”, which is when AI makes up incorrect answers. He said “some form of human in the loop” is needed to ensure accuracy, using the example of one site which he challenged and found it had used Reddit and Facebook posts as sources. “What these generative AIs give you is a very sensible-sounding answer which on the face of it seems correct,” he added, but this doesn’t mean it is necessarily accurate.

Shan said AI will need to be monitored, but it is “inevitable” it will be monitored by other AI itself, due to the large data sets.

Asked about the impact by jurisdiction, Conrad said “different countries are at different places”, with countries with unique personal identifiers for citizens at an advantage.

Concluding, Conrad said the more AI is adopted, the more it will affect legislation. He added that AI “is going to transform the tax world”, adding: “We should all take it pretty seriously, albeit with a degree of scepticism.”

Watch the full address.