AI in the Small Tax Practice

SPEAKER_01 0:02

Welcome to Accounting in the Wild, where we'll discuss various subjects in the tax and accounting world. Also featuring tales from the field and interviews on various tax and accounting subjects. All this and much, much more on Accounting in the Wild. And now here's your host, Jason Dinison.

SPEAKER_00 0:26

Welcome to Accounting in the Wild. It's been a while since we've done this. Like almost a year. The podcast is not dead, though. It was just on hiatus as we figured out what we wanted this podcast to be, and that is by a small practitioner for small practitioners. So we're gonna have episodes every two weeks now. Sometimes it'll just be me talking like this time, and other times we will have guests. This time around on Accounting in the Wild, I wanted to talk about the IRS's initial thoughts on AI usage for practitioners. They released this back in June. And there's some things in that that I think we need to talk about as practitioners. So this was a news release from the IRS that laid out their thoughts on AI. And if you read this news release, we'll have a link in the show notes. A lot of it is stuff that you've probably seen elsewhere if you have looked at anything that's given AI cautions for practitioners. Now, it's always good to have reminders of those guidelines and those cautions. But keep in mind with some of this stuff, it's more than just common sense. There is stuff in the tax code that actually could lead to prison time with some of this stuff. So we do need to be thinking about proper AI usage. So what we're gonna do today is look at what the IRS said, and then we'll bring it back to what does the average practitioner need to do? And it's one of those things where you're probably using AI even if you don't know that you're using AI. I think that a lot of us who do use AI think that we know exactly what AI we're using. I use Chat GPT or I use Claude. But keep in mind that there could be AI that's embedded in things that you're using. You might not even be consciously aware that there's AI in some of the things that you're using. So the IRS draws from Circular 230. That's what a lot of their it's a news release slash like initial guidance. It's really initial thoughts on proper AI usage. A lot of it comes from Circular 230. Now Circular 230 has not been updated since 2014, and there's a lot of things in Circular 230 that are out of date and kinda quaint, actually. But the thing is the concepts relating to ethical AI usage really are timeless, timeless and universal, and I think Circular 230 actually holds up quite well at handling AI. The IRS starts their document by talking about AI hallucinations. So a hallucination is simply the AI makes something up that is inaccurate. It doesn't happen every time you use AI, of course, but it's always a risk that it's going to happen. And even big players get caught up in this. Deloitte Australia used AI to generate a report for the Australian government, and they charge the Australian government in US dollars like two hundred and ninety thousand dollars. And this report had at least twenty errors in it, ranging from false citations to made up quotations. And it was the AI that did this, but nobody was reviewing it apparently. Or if they did review the document, nobody caught it. Deloitte ended up having to refund some of the fee that it charged the government because of this. The other thing that happens seemingly all the time is in the United States attorneys get busted for submitting court briefs and other documents generated by AI where the AI has hallucinated citations and other items in the document. And apparently the attorney either didn't review the document, or if they did review it, they didn't catch those things, and the thing is the judge or the judge's clerks catch those things, and then the attorney gets called on the carpet for it. Now hopefully nothing that dramatic has ever happened to you. I hope that it hasn't. But anytime that you use AI in your practice, you need to be aware of your circular two hundred thirty expectations, which in a lot of cases revolve around due diligence, so human review. We just talked about AI hallucinations, and that's what the IRS does in their document. They started by talking about AI hallucinations, then they move into the regulatory framework of proper AI usage, starting with due diligence. So Circular 230 10.22 says that practitioners, and I'm quoting here, a practitioner must exercise due diligence in preparing or assisting in the preparation of, approving, and filing tax returns, documents, affidavits, and other papers relating to internal revenue service matters, and then it goes on and on talking about how you basically have to verify and review the things that you're putting together. So in other words, whatever you're asking AI to do, you need to review what it spits out. And that seems like a common sense thing, and I suppose it is, but obviously it's not always happening. Someone at Deloitte surely somebody at Deloitte reviewed that document, but somehow those hallucinations made it through into something that was published and given to the government, not given to the government, they charged the government for it. In addition to being a violation of Circular two hundred thirty, things like that are just embarrassing in general. Imagine you put something together for a client and there are AI generated inaccuracies in it that you didn't catch, but your client does. That would be a really embarrassing situation for you. The IRS then goes on to talk about 10.37 of Circular 230, standards for giving written advice, which gets into how when you're examining something like a tax situation or you're giving advice, you need to base it on reasonable facts and circumstances, consider all the relevant facts and circumstances, use reasonable efforts to identify and ascertain facts relevant to the advice that you're given. So in other words, your sources. Are you using reliable sources? Not rely on representation, statements, findings, or agreements of the taxpayer or other person if reliance on them would be unreasonable. Now that is really where AI comes into play. AI is going to say whatever it says from examining things, and you need to be able to look at it and determine if it's reasonable rather than just rolling with it. And this gets into one of the things that I like to talk about when we talk about AI and tax pros. I feel like, especially once you've done this for a while, we develop kind of a sixth sense on this sort of stuff. Think about when you're reviewing a tax return. You usually will be able to tell that something is not quite right. And you may not know exactly what ain't right, but you know something's not right here. And you dig in and you find what's not right on that 1040. And it's all just kind of a sixth sense, like you know something is not right here. I feel like you can develop that with AI as well. It's a sixth sense that just like you look at it at whatever the AI is saying on something, and it's like, I'm not sure that that's right. Something doesn't sound right here. I think that with AI, AI, there are a lot of criticisms of AI that are valid. And I think that maybe if you're just starting out in your career, maybe it's not so good to rely on AI. But once you've been doing this for a while, AI can be very helpful because we know how things should look. We know how things should be done, and we have that sense of things where we can look at what the AI has given us, and we know. We may not know exactly what's wrong, but we know this isn't quite right. Now that feeds into the next thing, which is some people have said that AI will get you 80% of the way. Now that's just that's not necessarily a literal thing, it's just some people have said that, but I actually have found that that's true in my own anecdotal experience. AI will get you about 80% of the way on whatever it is that you're trying to do. And the thing is, people will say that the whole oh it only gets you eighty percent of the way. They say it as a criticism. I actually look at it as a good thing because it could do 80% of the heavy lifting on something, so that you're left with review and cleanup work, but it's better than having to do everything yourself. I think you look at AI as junior staff, very knowledgeable, a powerful ability to look things up, but they don't always know all the nuance and how all the pieces should fit together. And so you always need to go through and verify whatever it is that the junior staff is doing, and that really feeds into the next part of the IRS guidance, talking about procedures to ensure compliance. And the IRS talks about 10.35 and 10.36. 10.35 of Circular 230 relates to competence, and 10.36 relates to procedures to ensure compliance. So basically 10.35 says you need to be competent with how AI works. Understand whatever tool it is that you're using, and know how it works and what it's doing and what it's giving you. And if you have staff, make sure that they're trained. If your staff is using that tool, make sure they're trained on it. 10.36 gets into procedures on that. So 10.35 is competence, which would get into training. 10.36 gets into actual procedures. Plus on 10.36 with procedures, it's not just knowing how to use the AI, it's also having internal policies on what is given to AI. Now, all of those things probably seem like somewhat basic common sense ethical things, like yeah, yeah, yeah. Use common sense when using AI. But there's more that the IRS gets into where it gets a lot heavier, like go to jail type of stuff. So that 6713 and 7216, those are code sections. They get into usage of client data, usage and disclosure. So an AI tool is in fact a third party. And if you're giving client data to that AI tool, you're disclosing data to a third party. Which means you need your client's consent in order to do that. And you also need to be aware of like what is this tool doing with the data that you're feeding it? Now, Section 7216, if you think about it at all, you probably think of it on things like offshoring work to India, for example, and that is true. But it can also come into play in many other ways, even if your data never leaves the United States. Usage of AI is one of those ways that Section 7216 can come into play. And Section 7216 comes with not just monetary penalties, but up to a year in jail as well for violations. Section 7216 and 6713, they combine to create monetary penalties as well. And I won't read all the monetary penalties to you, just know that there are monetary penalties. Up to a hundred thousand dollars per violation if there's a crime involving identity theft because of your actions. So be careful with what you're giving to our robot overlords. And then the IRS ends their document by talking about fees. And this is something that I'm still pondering myself is what they really mean by fees. So the IRS in their news release, they mentioned Circular 230 and the prohibition against charging what are called unconscionable fees. And the IRS talks about how AI can reduce the time involved in researching and drafting advice, and that cost savings should be passed on openly, and that practitioners should fairly credit to the client's account any cost reductions. Now, if you're online, especially if you're on LinkedIn, my goodness that caused a stir on LinkedIn. I would say on that, what I would say on that is that we should think about what we're doing and just think about this some more before we make any hasty changes to how we bill. Some people have pointed out that the IRS's words seem like they would be true if you're billing based on time. So billing by the hour, if you are billing a research task that would have taken you two hours, but now it takes you one hour because of AI, how can you ethically bill your client for two hours now? So I think some of this of how what the IRS's words mean here, it depends. If you're billing by the hour, it would seem like the IRS would frown upon billing by the hour anymore if you're using AI tools or billing at the same rates anyway if you're using AI tools. But what if you don't bill by the hour? I don't know. I mean the IRS doesn't specifically say in this document Let's start landing the plane. We've been going on for a while with just me talking. So how do we navigate all of this? This is what I would recommend for small practitioners. Write down start. Just start by writing down the AI you know you use. It might just be Claude or Chat GPT. And then think about and write down the AI that might exist in other places, like your tax prep software or your email. And then think about like if you're using ChatGPT for tax research, but you never give client-specific information to ChatGPT and it doesn't have access to any client files, then you really don't have anything to disclose. But if it's something where client data is going into the machine, you need to make sure that that gets added to your client disclosures. And then think about and start documenting your AI procedures. And especially if you bill by the hour, think about how you are doing your billing for things like tax research. Well, that'll do it for this episode of Accounting in the Wild. We'll do this every two weeks going forward, and like I said at the start, we'll have guests every now and then as well. Thanks for joining us. I'm Jason Dinason, saying so long until next time.