The rapid advance of generative AI technology is seeing chatbots become increasingly competent at formulating complex tax advice and research, accounting market participants have said.
Jeff Saviano, global tax innovation leader at EY, argues that while generative AI technology is frequently seen as little more than an administrative instrument, it actually bears the potential to offer bespoke advice and make nuanced judgements.
“We’re seeing these tools already being used in a multitude of ways,” he says, specifically citing TaxGPT, the self-professed world’s first AI-powered tax assistant, which launched in May. “I wouldn’t cut and paste [its outputs] and send it to a client, but it’s so powerful that it’s a great place to start.”
Advisory expertise
Saviano explains that he has experimented with TaxGPT, challenging it by presenting mock scenarios and asking it to cross-reference client information with the relevant tax law and produce a judgement. He endorses its ability to take various factors into account and offer both qualitative and quantitative conclusions on tax liability, saying that the tool “got it right”.
He also praises the tool’s ability to conduct “persona-based research”, explaining that it can analyse the characteristics of multiple parties and advise on how to defend a position in support of party A versus party B.
“That’s something that I find to be the most powerful in terms of its distinction from how more traditional search engines work,” he says, adding that EY is “moving really fast” to embed this technology into its service delivery.
Rob Hackney, tax manager at DSG Chartered Accountants, offers a similar perspective despite practicing in the industry’s mid-market. While its emergence has been slower in tax than other areas, AI could have a “transformative impact on the role of a tax professional”, he says.
“Although the adoption of AI in a tax environment has been slower than in other business areas, we’re seeing it in increasing use, particularly given the extremely rapid speed at which large volumes of data can be processed.”
Like Saviano, Hackney endorses AI’s proficiency in a wide range of tax concerns, including identifying misanalysis, contentious items, disallowable costs or expenditure qualifying for specific tax reliefs.
Citing a more specific use case, he also explains that AI-based classifiers can automate the capital allowances classification of a large dataset, tagging each datapoint with a confidence score that allows human review quickly identify areas of concern. This, he says, demonstrates the capabilities of AI in an advisory context.
“While earlier applications may focus on compliance-based or numerical tasks, AI has clear potential to assist in an advisory context. And inevitably, more complex businesses will experience a wider range of specific areas in which AI can be applied in a tax context, especially in multinational enterprises.”
Mitigating the risks
But while Hackney hails the potential benefits of AI for tax accountants, arguing that it will enable advisers to dedicate their time to higher-value work, he is also quick to warn of the potential dangers. Chiefly among these is “unrealistic expectations”, he says, adding that the value of nuanced human judgement must not be underestimated.
“Care must be taken to avoid unrealistic expectations – if data quality is poor, AI can only do so much to generate useful results. There will always be areas where professional judgement is required which is significantly beyond what current AI solutions can provide.”
Hackney goes on to add that, while regulators have a role to play in terms of adapting data protection regulations, the key consideration for market participants is the provision of appropriate training. This is “crucial” to the successful deployment of AI, he says.
Concerns around training are strongly echoed by EY’s Saviano, who warns of chatbots’ susceptibility to hallucinations (generating false information). The art of “prompt engineering” – correctly formulating questions – is critical in this regard, he says.
“These generative AI tools are so powerful that there’s both an art and a science to asking it questions. There’s a whole world of knowing how to prompt it for research.”
Saviano has also been actively participating in the drive towards deepening education and regulation around AI technology, having launched the EY Advanced Technology Tax Lab network in collaboration with MIT in 2018.
Most recently, the network’s focus has been on extending the American Bar Association’s Canon of Professional Ethics to include the use of artificial intelligence. The onus on “duty of competence” has been among the key tenets emerging from this process, Saviano explains.
“We’re just about at the point now of saying that in order to be a competent lawyer, you need to have considered how artificial intelligence can improve how you are serving your clients,” he says.
“The foundation of this is something a bit like the Hippocratic Oath – first do no harm.”