AI can’t replace ethical judgement and accountability, The Tax Institute says

Tim Sandow, president of The Tax Institute, has reminded tax agents to treat AI outputs with a healthy degree of skepticism, and avoid putting confidential client information into open-source AI tools.

“You are still absolutely responsible for the work that AI produces, and also you need to be very careful around maintaining client confidentiality when you’re using AI tools,” Sandow told Accounting Times.

“Whether it’s AI that’s doing the work, or whether it’s another staff member that’s doing the work, under the TASA Code of Professional Conduct, it’s always the tax agent who’s responsible.”

Sandow said that AI tools could help practitioners automate some routine tasks and alleviate their workloads amid ongoing staffing shortages. However, he reminded tax agents that AI outputs must be evaluated with professional judgement and skepticism.

“In the same way that we have quality systems that allow us to supervise the work that’s been done by staff members, you have to have the same sort of quality control over what AI is doing,” he said.

He added that tax agents should not enter confidential client information into open-source AI environments. The Tax Institute previously warned practitioners against entering confidential information into ChatGPT, as it would count as a breach of their Tax Agent Services Act 2009 (TASA) privacy obligations.

“If you are using an open-source program like ChatGPT and you are using it to, say, summarise a letter of advice, and it’s got confidential information, then you actually don’t know where that information is going,” Sandow explained.

“If the AI is using itself to teach itself about a new topic, then the risk is that your confidential information could turn up in somebody else’s search result when they’re searching on similar topics.”

Given the complexity of the tax space, Sandow said that AI tools such as ChatGPT could easily give out-of-date information or return answers relevant to tax laws in overseas jurisdictions.

He added that AI lacked the capacity to make ethical judgements, an important role fulfilled by tax practitioners in the ever-evolving and sometimes grey area of tax.

“Just because you’re allowed to do something doesn’t mean you should do it,” he explained.

“Tax is full of complexity and full of grey areas, and sometimes you actually have to make a decision about what is the risk appetite of the client, and what do they want to do? So that’s where there’s still this overlay of decision-making, and standing back from an ethical perspective.”

He added that AI tools could not effectively navigate the emotional side of tax matters such as succession planning. Human judgement would have to remain central to the ethical and emotional complexities of tax matters.

“When we’re looking at succession planning, there’s tax elements, but there’s often bigger emotional elements as well. And so I think ChatGPT will give you answers, but you still have to overlay that with what you are actually trying to achieve.”