“In my field of tax, the answers that AI comes back with are not always clear and require further research.” Chartered accountant GAIL FREEMAN sounds a cautionary note about relying on artificial intelligence for tax research.

I have recently been using an artificial intelligence program for tax research. I realise that the answers depend on the quality of the questions.

I remember at university being told there were open and closed questions. Subsequent research indicates that there are more types of questions than simply open and closed.

I see open questions referred to as the 5Ws – namely who, what, where, when and why. I also ask many questions starting with how.

An example of an open-ended question I asked recently is: “What records do I need to keep to claim working-from-home deductions.”

The AI program I use is tax-specialist designed and it came back with a large amount of information. It also had five references and some potential questions to ask.

I did not accept the answers as being accurate without following them through and looking at all the other references that were there. This is so that I do not provide wrong information.

Sometimes the answers are not clear, so I need to ask another question for clarity. I am mindful that at the bottom of the research it says “AI can make mistakes”.

In contrast, when I ask a closed question, I expect an answer of yes or no. For example: “Can I claim working-from-home expenses?”

The possible answers are yes, no or maybe and maybe requires more information to come up with a definitive yes or no answer.

When asked that question, AI came back with a much shorter answer with only one reference and stated that I had to satisfy certain criteria. The quality of the answers depends on the quality of the questions asked.

I might ask a reflective question so that I can satisfy myself that the AI is correct. It is basically a form of doublechecking.

In my field of tax, the answers that come back are not always clear and require further research. But if the starting point isn’t clear and accurate then the subsequent questions are not going to provide the answers I am looking for.

Albert Einstein once said: “If I had an hour to solve a problem and my life depended on the solution, I would spend the first 55 minutes determining the proper question to ask. Once I know the proper question I could solve the problem in less than five minutes”. This comment is very pertinent when framing questions for AI.

Depending on the complexity of the question I’m trying to research, I often break it down into smaller chunks.

I read a news article a few days ago indicating that if a property is sold within 10 years it can be subject to tax in NZ. It is just not called capital gains tax. However, I asked AI, is there a capital gains tax in NZ? AI provided an answer that says NZ does not have a capital gains tax and then gave me a lot more information. So AI does not always give correct answers.

I don’t want to be the person named in the media who provided wrong answers because they took AI shortcuts. AI is powerful and useful, but you must check the output and ask the right questions.

If you need help with any tax-related matter, including using AI in your business, ask the experts at Gail Freeman & Co Pty Ltd on 02 6295 2844, email [email protected] or visit gailfreeman.com.au

Disclaimer
This column contains general advice, please do not rely on it. If you require specific advice on this topic please contact Gail Freeman or your professional adviser. Authorised Representative of Lifespan Financial Planning Pty Ltd AFS Lic No. 229892.

News all day, every day at CityNews.com.au.