VOICE-OVER
This podcast channel is about you, successful international entrepreneurs, successful expats, successful investors, sponsored by HTJ.tax.
DERREN JOSEPH
First of all, for those joining us, hello and good afternoon. So, we’re HTJ.tax. We do these live streams as often as we can, but now, since it’s tax season, it’s not as regular as it used to be. It used to be every week. But it’s tax season, and you know how it goes. So today, we have the honor and privilege of having Niklas Schmidt back with us, and we will talk about AI and international tax. And as we always say, we are licensed and credentialed tax professionals, so we are not going to be giving any actionable advice. We will just have a general conversation about general principles, so please keep that in mind.
I wanted this to be a bit different because it feels as if AI is now like a bandwagon. It’s like when one soccer team is winning, suddenly they get all these new fans, and everybody jumps and supports them, right? I don’t want to jump on the bandwagon, but I want to cut through the noise. Yes, AI is here. Yes, AI is a thing, and for those of us who engage in international tax or will engage in legal or advisory or professional services in general, clearly, there’s going to be some sort of impact. That much is known. We know that.
What I want to do is have a conversation where we create, first of all, the framework. These are the pre-existing trends, for sure, but AI is not going to create anything out of the ether. It’s not going to create anything brand new. It’ll perhaps accelerate certain pre-existing trends as a catalyst to some extent, so that’s what I wanted to do. I would engage with you and pick on your brain because you are on the cutting edge as a thought leader yourself. So, those are my thoughts. Yes, it’s happening, but what does it mean potentially for those in the international tax space, both as clients, as advisors, and so on? What do you think?
2m 44s
NIKLAS J.R.M. SCHMIDT
Let me kick off by sharing how my journey started. So, basically, in February, I rang the alarm bell in our company and did an internal presentation on what is happening in AI. I have been following things since May or June of last year. Why is that? Our oldest son is very much into machine learning. He’s currently studying that in London. I always got input from him and got to use a number of these tools.
They were really exciting, and at some point, they entered the realm of social media. So, in the autumn of last year, everybody was excited about D-ID and ChatGPT, and you saw numerous posts. At the beginning of this year, I thought, “Hey, we really should, as a firm, look into this and think about what we are going to do; otherwise, we might very well appear stale in a short while.” I did an internal course and essentially showcased what was available.
Now, you mentioned that you don’t want to jump onto the bandwagon, and you’re wondering what is really going on. Let’s take a look at that and try to understand what it means for tax professionals. What is the state of play? Is it just a trap or a mere toy? Or is it something more significant that could potentially replace you and me? Where do we stand now?
I would say that these models, called language models, are quite credible.
I mean, what these models do is essentially a statistical process. It’s a sort of brute-force approach. It’s very mechanical, but in a certain sense, it strings together words based on probabilities. If a certain word appears, the probability of the next word being a specific one is calculated. These programs are far from achieving a Terminator-like scenario. So, we’re not at that stage. However, the outcomes they produce are truly remarkable. Let me give you an example, which I think is probably the best.
Someone posed a question to GPT-4, a variant of Open Ai’s large language model. The question was as follows: “You have a number of objects, and please stack them up into a pile. The objects are a laptop, a bottle, nine eggs, an incredibly thick and heavy book, and a nail.” The model responded, “Okay, we’ll take the book. Since it’s really heavy, it might crush the eggs. Therefore, we’ll place it at the bottom of the stack. Then we’ll put the eggs; we’ll take them very carefully and put them in a three times three array, order them like three by three. Then we’ll carefully place the laptop, which is flat, on top of the eggs. Next, we’ll take the bottle and position it with the bottom at the bottom and the top at the top. On the cork of the bottle, we’ll place the nail with its tip pointing upward because nothing can be placed on the tip. That’s how it’ll stack the objects. So, it’s important to note that this solution is not something that the model has simply repeated or found somewhere on the internet. This logical arrangement was generated based on the given objects and their characteristics.
So, a lot of times, the outcome is just incredible. But having said that, I did lots of tests in our area, specifically in the tax law area. I asked all sorts of questions. I’m a tax lawyer in Austria, so I asked questions regarding Austrian tax law. I also did them in German and not in English. And I got very unsatisfactory results. So, we’re not there yet with a model like chatGPT to get legal advice. It’s definitely not there yet.
Also, we have to bear in mind that chatGPT is limited to domain knowledge until autumn 2011. So, this model does not include anything new that happens in tax law. So, what we get is rudimentary. So, you get output that sounds totally reasonable if you don’t know about tax law.
You might think this is all correct, but it’s totally made up. So, we are not there yet for real hard advice. But the thing is that there’s an incredible amount of research going on. So many very smart people are putting so much energy into developing this. And there’s an estimate which says that basically, over the next ten years, the performance of these models will increase by a factor of 1 million. And what that means, basically, is that you will have significant performance improvements happening within this 10-year period; it will just get better and better over time.
So, it’s important to keep the eyes open to see what’s happening. And so, I will say the state of play regarding advice, what you can now already do and what you actually should be doing is using AI for all sorts of marketing things. So, I just sent out an email this morning to my team and said congratulations to one of my team members. His name is Stephan, so I’m shouting out to him now; congratulations to Stephan. He got the AI prize for our team because he was one of the first in my team to really use AI for social media.
He put out a very fantastic post on LinkedIn, which ChatGPT had sort of made. Of course, he had put in a few of his own thoughts, but basically, it’s super! And it sounds great. The post is very exhilarating. There are a lot of good vibes in it. The English is perfect. It’s fantastic! The better you move, the better feedback you’ll get.
DERREN JOSEPH
thanks to that. I appreciate you creating that context. It is super important because there is this perception that the world, we’re living in is in the Terminator matrix. It’s going to take over and kill us all. But it’s important to note that we are still in the relatively early stages. There are limitations. There’s a limit to what is possible and what can be done presently. But at the same time, it is arguably already having an impact. And as you pointed out, one can extrapolate what may or may not happen in the future.
DERREN JOSEPH
So, from my point of view, and perhaps it’s overly simplistic, I appreciate your perspective from the tax world as a tax professional. There’s the advisory side, and there’s a compliance side. So, there’s doing the physical tax returns in whichever jurisdictions someone may or may not be exposed. And there’s providing the advice, right? So, let’s start with the easier one first, which is compliance. I don’t know what it’s like in the jurisdictions you’re exposed to, but definitely, in the US, it may not be ChatGPT. Still, there is artificial intelligence being applied to the process of preparing tax returns.
The US is notoriously complex. There are quite a number of software providers where they ask you questions, and based on your responses, they compile a tax return. So that is already happening. So, before we get to the tax planning, which has the more exciting part, what is your perception regarding the compliance side?
NIKLAS J.R.M. SCHMIDT
Thank you very much for this clarification. I, myself, am fully on the advisory side and less on the compliance side. So, I somehow skipped that and immediately went into the advisory part. But, of course, the low-hanging fruit is in the compliance area. So it is, as you said, what were the typical things, that we could encounter? So, it is to help with preparing tax returns. it’s not for General Electric or some mega-corporation. But rather, it is for somebody working from home and with some expenses that he or she would like to deduct.
So, I can imagine that this is quite easy to do. Basically, a chat system where you can enter all sorts of queries, and you will get responses covering basically every typical scenario of a solo entrepreneur, for instance. But of course, also on the side of the tax authorities. And definitely, they’re doing that in Austria, and I’m sure that they’ve been doing that for decades already in the US, that is, checking tax returns. I mean, you can run all sorts of analysis, and it’s maybe a bit less than AI; sometimes, it’s just statistics.
You can run all sorts of analyses to see whether there are any strange outliers or unusual elements in the tax return or even in the underlying accounting system. These are two areas where automation can be particularly beneficial. For businesses, I can imagine that there is already a trend toward more and more automated accounting. Essentially, a stack of invoices is scanned and sent into the system. The system will automatically handle the bookkeeping and accounting. It will generate a proposal; all you have to do is press enter if it’s acceptable.
And you’ll just press enter the whole time, and everything will be booked correctly. The system will recognize the supplier, whether it’s a utility company or a purchase of a desk or a printer, and automatically allocate the transactions to the correct accounting categories. These areas of compliance will be where we see the initial results of automation and the usage of AI.
DERREN JOSEPH
That’s an important point, both on the side of the taxpayer and on the side of the authorities. I think there is a company called Palantir, which is a huge technology company. They have or had, as far as we are aware, a significant contract with the Internal Revenue Service (IRS) in the US. Part of that contract, although the details are confidential, likely involves bringing some form of technological intelligence to the compliance process on the part of the federal government.
It would definitely add value, not just for taxpayers but for tax authorities themselves. As you pointed out, your area of expertise, your sweet spot, is tax planning. So, before we discuss how artificial intelligence could potentially impact tax planning, what do you see as the big trends, perhaps three big trends, that are currently impacting the entire tax advisory space? Just to clarify, for those who are watching this later, we’re at the end of May.
DERREN JOSEPH
So, yeah. What do you see as the big trends that are impacting your practice?
NIKLAS J.R.M. SCHMIDT
Yeah, I mean, moving from the compliance side where automation and AI are clearly happening, I think the real concern is that professionals in our field may mistakenly believe that the advisory side is so high-level, technical, and complex that machines won’t be able to handle it. But I think the main takeaway from our discussion here is that no incredibly complex professions are safe from AI’s impact.
17m 51s
NIKLAS J.R.M. SCHMIDT
According to a Goldman Sachs report, the legal profession is ranked as the second most endangered. However, we need to put that into perspective because the legal profession is a vast and diverse field with various areas of specialization. For instance, you may be involved in due diligence as a corporate lawyer, reviewing documents that can potentially be automated. Alternatively, you may work as a trusted advisor for private clients as a private client lawyer.
NIKLAS J.R.M. SCHMIDT
You have built a relationship over more than a decade with someone who relies on your gut feeling, your network of contacts, your relationships, your knowledge of how things work, and your ability to solve problems without attracting attention. This is definitely something that also applies to the advisory field and cannot be taken over. It’s important to make distinctions here. So, having said that, when you asked about Darren and the three most important things happening outside of AI, what were you specifically referring to?
DERREN JOSEPH
Well, I meant outside of AI because my thinking is that, on the advisory side, it’s already subject to disruption. This is not something new. The profession has already been coping with disruptive forces. So, this is just one on top of it, or perhaps it’s all-encompassing. I don’t know. But it’s just to create a context for the advisory side.
NIKLAS J.R.M. SCHMIDT
Yeah, I mean, in international tax planning, in the past decade, we have witnessed the introduction of various additional reporting obligations and increased information exchange. Different rules and legal instruments have been pursued, enabling authorities to exchange information with other tax authorities. And the world is becoming smaller and smaller. So perhaps it’s just a matter of authorities catching up, a game of keeping pace. Previously, taxpayers enjoyed a better position as they took advantage of globalization.
Taxpayers have been operating across borders since the 1970s. They were doing business abroad, having a bank account abroad, and having IP abroad. And the tax authorities were confined to national borders. They could not access information in Tax Haven, say, or even other civilized countries. And slowly, this has been rolled back, and we, tax authorities, achieved a level playing field. So, they got access to all sorts of information from outside of their own territory, outside of their own jurisdiction. But of course, it has not only been a rolling back and a leveling up with the taxpayers; it has actually moved into a distinct advantage for the tax authorities.
They are now in the position of changing laws even before we have implemented the structure. So, let’s consider DAC6, the European rules on the notification of tax schemes or tax planning arrangements. You basically think of a strategy, and even before you file a tax return, you already notified the authorities, who will then essentially pass the law because there’s not really a distinction between the Minister of Finance and the Parliament.
Because Parliament does what the Minister of Finance or the Treasury, or in your case, the Internal Revenue Service, would essentially instruct them to do. And any loophole that you as an advisor might have discovered will be closed before your client can even file their tax return. So, I would say that the playing field is not level, but it has shifted in a way that benefits the tax authorities rather than the taxpayer.
DERREN JOSEPH
Hmm. In that context, where the balance of power is shifting even more in favor of the authorities, how do you envision AI impacting the field of tax advisory today and in the future, considering this ongoing trend?
NIKLAS J.R.M. SCHMIDT
Yeah, I mean, what does it mean? I mean, for the advisor, it’s going to be an instrument to inform your audience. I recently came across a group that has uploaded the entire Internal Revenue Code into a chatbot-like system. However, I’ll have to find the exact name again. They have been training it by having professionals use the platform and ask questions. These professionals are experts who can verify the accuracy of the chatbot’s responses.
And they basically gave the chatbot a slap on the wrist when it was wrong and a pat on the back when it was right. This is one of the three approaches to machine learning; I believe it’s called supervised learning. And this system now has a rate of something like 80% correctness, and it’s continuously improving. So, you can already use it just as an example to compare it to the US bar exam.
There’s something called the multi-state bar exam; I think it’s called MBE. And just a few months ago, in December, an open AI system had a 51 or 52% pass rate. So that’s like rolling a dice. Yeah. You either got it right or wrong. But then the system was improved. It’s, it’s called GPT4 from GPT 3.5, so a small incremental change, whatever that exactly means in the background under the hood, but suddenly the level jumped from slightly above half to 75%. It was just only a few months later. And similar things can also be expected with this application. So, it’s just getting better and better. And listen. Basically, we are having a bilateral communication. You are listening, and I’m speaking, and then I’m listening, and you’re speaking, and yeah, basically, I’m learning a lot from you, and you’re learning a lot from me. But I mean, just like you and I are in a bilateral speaking relationship, it’s basically speaking and learning from tens of millions of people at the same time and always getting better.
NIKLAS J.R.M. SCHMIDT
So, it’s going to be quite fast. There are two camps of people. Some believe that it’s just a distraction and we have nothing to fear because these systems won’t work out so quickly. On the other hand, there are those who believe it’s the end of the world and machines are taking over, calling for a halt to AI development. You may have come across this in the Lex Friedman podcast.
If not, I would recommend that you look him up. So, he had interviewed this one guy who called for a six-month moratorium on AI development. And so it was; I cannot say which one is correct or what the outcome will be, but definitely, it might be somewhere in the middle. Things will develop fast, but we may not have superintelligence, as it’s called, or fully realized artificial intelligence very soon. But it’ll move fast. And you have to really look at things. I think everybody, as an advisor, should actually use ChatGPT, at least for marketing purposes. Just put it in there and rephrase it to sound professional and engaging, to have better English than I have, and so on. So, you can really improve your style. Of course, we can delve into data protection and privacy, among others. But, yeah, let me hand it back over to you.
DERREN JOSEPH
Hmm. So, I definitely understand your point regarding the marketing aspect. In full transparency, our company website, particularly the consumer-facing section, has been utilizing a chatbot for a couple of years now. Interestingly, only a small percentage, less than 2% of individuals interacting with it, actually recognize it as a chatbot. Most people mistakenly believed that they were engaging with a human. However, with the increased media attention on chatGPT, more people now recognize it as a chatbot. So, in terms of marketing, it’s safe to say that this approach has already gained significant traction.
DERREN JOSEPH
Now, to your point, the other point you raised about the legal profession, yeah, I mean, I saw that as well, and it did pass a multi-state bar exam pretty impressively. But you know, one of the things about chatGPT is that it’s a user interface. The idea is that, as you point out, machine learning and algorithm predictions have been part of business for a while. It’s been more than a decade, for I have clients, you know, creators, algo traders. So, in a way, algorithms are highly intelligent pieces of machine learning, so this stuff has been going on.
I think now that it’s consumer-facing, it has created this level of awareness and so on. But anyway, switching back to the legal profession, it has been a while since some sort of artificial intelligence was used for things like simple contracts. I mean, I know the term AI is loaded and has emotional implications, so let’s refer to it as intelligence that is artificial. This has been happening for quite some time now. If you have a relatively simple arrangement that you want to be captured or formalized in a contract, it’s been happening for probably the better part of a decade.
So, again, it echoes what you said at the very beginning that the worst thing a professional advisor could be is complacent, believing that this will not impact the profession. Obviously, it will. And you mentioned Goldman Sachs, which is ranked number two. So, in terms of how it would impact, I see a parallel with chess. When I was younger, I was into chess, and in the 1990s, IBM’s Deep Blue and other technologies started beating chess grandmasters. Initially, the technology was beaten by the best grandmasters. Still, then the tables turned, and technology began to beat the best grandmasters.
DERREN JOSEPH
But now the best teams are hybrids. They consist of a combination of a chess grandmaster and an AI tool by their side. So, I would imagine that an advisor can enhance their capabilities by being armed with certain tools. Because, at the end of the day, what you’re doing is referring to code sections, case law, precedents, opinions, and regulations and combining that with your experience and expertise. I believe the average advisor’s ability to work competently with clients will be tremendously enhanced.
DERREN JOSEPH
That is definitely something that I see as being an event outcome. What are your thoughts?
NIKLAS J.R.M. SCHMIDT
Yeah, I completely agree. It will be a tool that we use to enhance our service delivery and improve the delivery of legal and tech services. Today, I posted a link on LinkedIn to a Wall Street Journal article about a remarkable guy who tutors American school children for international math competitions. He travels across the country and has played a significant role in helping the US win several medals after a decade of not winning any.
And he was also speaking about the challenges of AI, and he said, “It’s a tool, and you have to concentrate. You cannot compete with a tool, and you don’t have to compete with a tool. You should concentrate on what only humans can do and use this tool to complement your humanity and your skills as a human being because there are some things that these bots will not be able to do. And you will have to find out in your concrete niche what it is that only you can do. But see, this is a tool and not something that will kill you.
DERREN JOSEPH
Yeah, that’s a valid point. But at the same time, you know, they say that a little knowledge can be a dangerous thing. Currently, as we are live streaming this, there are people presumably watching it, some of them on YouTube. On platforms like YouTube, you can find individuals who may or may not be qualified but offer advice out of their hearts’ goodness. While they may be well-intentioned, their lack of professional background and training can sometimes lead to misleading the intended audience.
DERREN JOSEPH
So, I also see a danger in this trend, giving people artificial confidence. So, you know, they think, well, I played around with ChatGPT for 15 minutes, so I’m now ready to advise on some sort of cross-border arrangement. Again, that’s a trend that’s already happening, right?
NIKLAS J.R.M. SCHMIDT
That’s true. That’s a great point because you can find a video on YouTube for everything. So, if you don’t know how to tie laces, you can look it up. If you don’t know how to fix a wall or if you have some pain in your chest, you might consult with a doctor beforehand.
Medical doctors are very accustomed to patients coming in and thinking they already know everything. Maybe in our profession, we haven’t reached that point yet. Although, of course, you sometimes have clients who look up things and then ask you whether what they wrote down is correct.
And this is certainly going to increase. There will be many people who, in the past, might have used a search engine and found a website with a similar case. However, their case may be more complicated, and there may be a different match for their circumstances. But this is where a chatbot like ChatGPT or others comes in, as it provides more tailored advice. It offers advice that is more specific and different from a search engine. It doesn’t just provide a website that is close to what you’re trying to achieve, but it gives you the exact recipe.36m 30s
And that might be incorrect. Yeah, as we just mentioned, the taxpayer, the potential client, if they even become one, may have a sort of feeling like, “Oh, okay, I consulted this chatbot and it gave me this answer, so I can claim tax deductibility for this expense without any issue. I can send the money to an offshore bank account and avoid paying income tax on it.” They might actually engage in tax evasion, thinking everything is fine. It will be interesting to see if they can later argue: “I looked that up like five years ago and received this advice, and I had no reason to believe it was incorrect.”
NIKLAS J.R.M. SCHMIDT
Yeah. So, we might see some cases where people try to defend their wrong tax returns by looking them up on chatGPT.
DERREN JOSEPH
Yeah. And it’s not uncommon for us to have prospects approach us after seeing something in a movie or a Netflix series. Forget about unqualified influencers; that’s one thing. But when they see things on TV or in a movie, they think, “Hey, if this person could do it, why not me?” And we have to explain to them principles like purpose or economic substance. So, the challenges will continue for us as professionals. You mentioned earlier that the legal profession would be number two.
So, in terms of professionals providing advice, what steps do you think advisors can take to prepare themselves for what’s happening and what’s coming? Apart from integrating it into their marketing, what other actions can professionals take to be ready?
NIKLAS J.R.M. SCHMIDT
Yeah, I recently came across a law firm’s website that had a logo saying “powered by AI” below it. I found it quite interesting, but maybe it’s just a small marketing tactic. Nevertheless, I would recommend everyone to give these tools a try. You don’t necessarily have to subscribe to them. Many of these tools are available for free, and I can personally vouch for their usefulness.
And I think it’s crucial to keep a chatbot like ChatGPT opens in a browser window at all times and consider how it can be used. However, I would advise caution when entering client-related information. It’s not advisable to do so. When it comes to client matters, it’s better to use a sanitized version. For example, we often refer to clients as “XYZ” in our memos without providing specific names, addresses, or dates.
I remember once having an exercise where several people in the firm contributed to a long email. Each person had a different writing style, including short sentences, bullet points, long sentences, and variations in English proficiency. It didn’t fit together well. So, I simply copied the entire text into ChatGPT, sanitized it by removing client names, and asked it to harmonize the writing style and make it sound as if written by one author. In just five seconds or even less, the result was impressive.
And once you have the harmonized version of a document, you can share it with everyone and ask if it still reflects their intentions. If it receives approval from everyone, you can send it out, saving a lot of time. Regular use of the tool can offer additional benefits. For instance, my colleague, who is attending this conference, recently posted on LinkedIn. It was a great read, filled with positive energy and far from being boring. So, by using the tool regularly, you will discover more of its capabilities and stay updated with what’s happening. It’s all about staying in the loop and exploring the possibilities.
DERREN JOSEPH
You also made a very interesting point that just like with other social media tools, for example, Instagram, when you upload a picture to Facebook or Meta, you’ve actually handed over the rights to that picture to Facebook, right? And similarly, my understanding, at least with Open AI for right now, is that when you put something into that window, you have handed over ownership to them. They can reserve the right to use it as they see fit in the future. And so that has implications from a data protection point of view.
NIKLAS J.R.M. SCHMIDT
But essentially, whatever you input into the prompt, as it’s called, can be stored and utilized later on. Basically, Darren, any content you have written in the past and any posts you have made on platforms like Reddit, LinkedIn, Twitter, or others, that information could already be within the system.
As you may know, AI has access to the internet and accumulated vast information. If you’re an avid reader, for instance, you might go through a book every week. So that’s roughly about 50 books per year. Let’s say you live until the age of 80, which would amount to around 4,000 books throughout your life. However, an open-eyed, large language model like ChatGPT has processed maybe hundreds of millions of books. Yes, something along those lines. The scale is probably of greater magnitude. So, anything you have written or any queries you make, like ‘Who is Darren Joseph?’ or ‘Have you heard of this Tax guy?’ will yield results based on the extensive information available. Give it a try; hopefully, you’ll discover some valuable insights.
NIKLAS J.R.M. SCHMIDT
But there have been cases where wrong information has come out, which can lead to numerous legal questions. The bigger topic is not these large language models but actually the AI applications that generate images because they have been ingested; for example, D-ID has taken 650 million pictures and processed them. So, it’s like an office that sees the world and has seen all sorts of artistic styles, objects, and so on.
And they’re not generating new pictures based on this sort of worldview. Here the question arises: Is it (a) that the person who ordered this picture has the copyright, or is it (b) the program, or is it (c) Open AI, basically Sam Elman and his friends who programmed this system? Or is it nobody? Yeah, and nobody seems to have the most jurisdiction. So, whatever you produce, you do not have a copyright.
NIKLAS J.R.M. SCHMIDT
I mean, things are going to move quite fast. It’s like with the internet, new rules will be coming up, and there will be all sorts of debates among legal scholars. Eventually, the legislator will step in at some point. So, this is an incredible area for lawyers to be in, as they can be part of this process.
DERREN JOSEPH
Absolutely. There have been quite a lot of debates regarding images. But even going back to the example you mentioned before, where you collaborated with other partners in your firm and used ChatGPT to harmonize the language of a document. The question arises: Who owns the output/ the final work product?
NIKLAS J.R.M. SCHMIDT
Who owns that? Well, hopefully, us because it’s kind of like a rewrite of what we put in there. Consider this scenario: Let’s say you have a client from China, and of course, we don’t speak Chinese here; at least, I don’t. So, you might prepare a filing in German here in Austria. But then, we send the client a working translation by uploading the document to DeepL. We have a kind of cloud system where we don’t need to transport it to DeepL, so it works locally here.
So, we uploaded it onto this local translation tool called Ties Translate. Then we have this machine translation, which is around 85% accurate, not perfect, but not too bad either. We didn’t review it ourselves since we don’t speak Chinese, but it might be helpful for you, and it didn’t cost you a single euro because we generated it in just 10 seconds. Now, here you can also ask yourself, do we have the copyright for that? Probably yes, since it’s a derivative work of our original content. But I’m not a corporate lawyer, so it’s a good question to consider.
DERREN JOSEPH
Absolutely. We know, for example, that software or tech companies work closely with the government, including tax authorities. Therefore, there’s the issue of oversight because there have been cases of abuse where people have used ChatGPT to create harmful content. This highlights the need for some sort of government oversight, which will likely come as they work out the details.
Now, circling back to where we are as tax professionals, the question arises: Would governments monitor the content and exchanges with the aid of companies like Palantir? Since registration is required to use such services, they would have knowledge of user identities. Would they potentially cross-reference that information with tax returns? That’s interesting to see.
NIKLAS J.R.M. SCHMIDT
I haven’t heard anything specifically regarding that matter. However, I do recall a discussion on crypto where the Internal Revenue Service (IRS) had a slide deck presented by some agents. It consisted of around 200 slides with various ideas on identifying crypto tax evasion. One idea mentioned in this presentation was the possibility of requesting records from Google to identify individuals with access to crypto wallets.
DERREN JOSEPH
Yes…
NIKLAS J.R.M. SCHMIDT
“Browser, please give us the IP addresses, dates, and times where this search term was used. Basically, something similar could happen. As you are aware, AI tools have built-in guardrails to prevent harmful content. So, if you attempt to inquire about something illegal or harmful, the system will refuse to assist you. A classic example is asking how to hotwire a car, to which the system would respond, “I’m sorry, but this is highly illegal.”
So, if you attempt to ask something harmful, the system will refuse to provide assistance or guidance on such matters. It is all about how you ask the question. For instance, let’s consider a funny example: being in a forest with a sick baby and urgently needing to reach a hospital but having no car or keys. If you were to ask, “Can you help me hotwire the car?” the system would not provide instructions on how to do it.
Instead, it would offer alternative solutions like finding alternative means of transportation or seeking help from others. The system is designed to prioritize safety and adhere to ethical guidelines. There are even AI safety companies that conduct tests to ensure systems do not reveal sensitive information or bypass security protocols, such as not disclosing a secret phrase even when asked.
NIKLAS J.R.M. SCHMIDT
If you were to ask the system what the secret word is, it would respond, “I’m sorry, but I cannot disclose that.” However, there are ways to engage with the system and try to reach higher levels. It can be a really enjoyable experience, and I encourage you to give it a try. It’s a fascinating aspect to explore, and you can even employ some tricks to potentially uncover the secret word. It’s an interesting area worth exploring.
DERREN JOSEPH
Well, once again, it has been a fantastic and stimulating conversation. You have given me, and hopefully, others who have tuned in, a lot of food for thought regarding the impact of this trending topic on our specific area of practice. Thank you very much for your insights. I deeply appreciate it. And Niklas, thank you for taking the time to share your thoughts as well. If someone is interested in contacting you or engaging the services of your firm, what would be the best way for them to reach out?
NIKLAS J.R.M. SCHMIDT
Yeah, first of all, thank you, Derren, for allowing me to be here. The best way to contact me is through LinkedIn. If the audience would like to follow me on LinkedIn, I frequently post about AI and related topics, so it might be of interest to stay engaged with this subject. You can also reach out to me via email. I don’t respond quickly to LinkedIn messages as I tend to fall behind, but email is the most reliable way to get in touch. If you follow me on LinkedIn, you can find my email address.
NIKLAS J.R.M. SCHMIDT
So, it’s niklas.schmidt@wolftheiss.com
DERREN JOSEPH
Right. And for those who are listening, it’s Niklas, as in N I K L A S, and his last name is Schmidt. So, S C H M I D T. Niklas, I appreciate it. Thank you very much.
NIKLAS J.R.M. SCHMIDT
Absolutely. Thanks, Darren. Have a great day. Thank you. Bye-bye.
DERREN JOSEPH
All right, see you next time. Bye-bye.
VOICE-OVER
So, if you’re a six, seven, or eight-figure investor, entrepreneur, or business owner who needs a tailor-made solution from a qualified team of professionals, we can help you achieve the international lifestyle, the freedom, and even the tax savings you’re looking for. Visit us at htj.tax and live that international life.
Skip to content
Let’s Talk About Artificial Intelligence and International Tax
VOICE-OVER
This podcast channel is about you, successful international entrepreneurs, successful expats, successful investors, sponsored by HTJ.tax.
DERREN JOSEPH
First of all, for those joining us, hello and good afternoon. So, we’re HTJ.tax. We do these live streams as often as we can, but now, since it’s tax season, it’s not as regular as it used to be. It used to be every week. But it’s tax season, and you know how it goes. So today, we have the honor and privilege of having Niklas Schmidt back with us, and we will talk about AI and international tax. And as we always say, we are licensed and credentialed tax professionals, so we are not going to be giving any actionable advice. We will just have a general conversation about general principles, so please keep that in mind.
I wanted this to be a bit different because it feels as if AI is now like a bandwagon. It’s like when one soccer team is winning, suddenly they get all these new fans, and everybody jumps and supports them, right? I don’t want to jump on the bandwagon, but I want to cut through the noise. Yes, AI is here. Yes, AI is a thing, and for those of us who engage in international tax or will engage in legal or advisory or professional services in general, clearly, there’s going to be some sort of impact. That much is known. We know that.
What I want to do is have a conversation where we create, first of all, the framework. These are the pre-existing trends, for sure, but AI is not going to create anything out of the ether. It’s not going to create anything brand new. It’ll perhaps accelerate certain pre-existing trends as a catalyst to some extent, so that’s what I wanted to do. I would engage with you and pick on your brain because you are on the cutting edge as a thought leader yourself. So, those are my thoughts. Yes, it’s happening, but what does it mean potentially for those in the international tax space, both as clients, as advisors, and so on? What do you think?
2m 44s
NIKLAS J.R.M. SCHMIDT
Let me kick off by sharing how my journey started. So, basically, in February, I rang the alarm bell in our company and did an internal presentation on what is happening in AI. I have been following things since May or June of last year. Why is that? Our oldest son is very much into machine learning. He’s currently studying that in London. I always got input from him and got to use a number of these tools.
They were really exciting, and at some point, they entered the realm of social media. So, in the autumn of last year, everybody was excited about D-ID and ChatGPT, and you saw numerous posts. At the beginning of this year, I thought, “Hey, we really should, as a firm, look into this and think about what we are going to do; otherwise, we might very well appear stale in a short while.” I did an internal course and essentially showcased what was available.
Now, you mentioned that you don’t want to jump onto the bandwagon, and you’re wondering what is really going on. Let’s take a look at that and try to understand what it means for tax professionals. What is the state of play? Is it just a trap or a mere toy? Or is it something more significant that could potentially replace you and me? Where do we stand now?
I would say that these models, called language models, are quite credible.
I mean, what these models do is essentially a statistical process. It’s a sort of brute-force approach. It’s very mechanical, but in a certain sense, it strings together words based on probabilities. If a certain word appears, the probability of the next word being a specific one is calculated. These programs are far from achieving a Terminator-like scenario. So, we’re not at that stage. However, the outcomes they produce are truly remarkable. Let me give you an example, which I think is probably the best.
Someone posed a question to GPT-4, a variant of Open Ai’s large language model. The question was as follows: “You have a number of objects, and please stack them up into a pile. The objects are a laptop, a bottle, nine eggs, an incredibly thick and heavy book, and a nail.” The model responded, “Okay, we’ll take the book. Since it’s really heavy, it might crush the eggs. Therefore, we’ll place it at the bottom of the stack. Then we’ll put the eggs; we’ll take them very carefully and put them in a three times three array, order them like three by three. Then we’ll carefully place the laptop, which is flat, on top of the eggs. Next, we’ll take the bottle and position it with the bottom at the bottom and the top at the top. On the cork of the bottle, we’ll place the nail with its tip pointing upward because nothing can be placed on the tip. That’s how it’ll stack the objects. So, it’s important to note that this solution is not something that the model has simply repeated or found somewhere on the internet. This logical arrangement was generated based on the given objects and their characteristics.
So, a lot of times, the outcome is just incredible. But having said that, I did lots of tests in our area, specifically in the tax law area. I asked all sorts of questions. I’m a tax lawyer in Austria, so I asked questions regarding Austrian tax law. I also did them in German and not in English. And I got very unsatisfactory results. So, we’re not there yet with a model like chatGPT to get legal advice. It’s definitely not there yet.
Also, we have to bear in mind that chatGPT is limited to domain knowledge until autumn 2011. So, this model does not include anything new that happens in tax law. So, what we get is rudimentary. So, you get output that sounds totally reasonable if you don’t know about tax law.
You might think this is all correct, but it’s totally made up. So, we are not there yet for real hard advice. But the thing is that there’s an incredible amount of research going on. So many very smart people are putting so much energy into developing this. And there’s an estimate which says that basically, over the next ten years, the performance of these models will increase by a factor of 1 million. And what that means, basically, is that you will have significant performance improvements happening within this 10-year period; it will just get better and better over time.
So, it’s important to keep the eyes open to see what’s happening. And so, I will say the state of play regarding advice, what you can now already do and what you actually should be doing is using AI for all sorts of marketing things. So, I just sent out an email this morning to my team and said congratulations to one of my team members. His name is Stephan, so I’m shouting out to him now; congratulations to Stephan. He got the AI prize for our team because he was one of the first in my team to really use AI for social media.
He put out a very fantastic post on LinkedIn, which ChatGPT had sort of made. Of course, he had put in a few of his own thoughts, but basically, it’s super! And it sounds great. The post is very exhilarating. There are a lot of good vibes in it. The English is perfect. It’s fantastic! The better you move, the better feedback you’ll get.
DERREN JOSEPH
thanks to that. I appreciate you creating that context. It is super important because there is this perception that the world, we’re living in is in the Terminator matrix. It’s going to take over and kill us all. But it’s important to note that we are still in the relatively early stages. There are limitations. There’s a limit to what is possible and what can be done presently. But at the same time, it is arguably already having an impact. And as you pointed out, one can extrapolate what may or may not happen in the future.
DERREN JOSEPH
So, from my point of view, and perhaps it’s overly simplistic, I appreciate your perspective from the tax world as a tax professional. There’s the advisory side, and there’s a compliance side. So, there’s doing the physical tax returns in whichever jurisdictions someone may or may not be exposed. And there’s providing the advice, right? So, let’s start with the easier one first, which is compliance. I don’t know what it’s like in the jurisdictions you’re exposed to, but definitely, in the US, it may not be ChatGPT. Still, there is artificial intelligence being applied to the process of preparing tax returns.
The US is notoriously complex. There are quite a number of software providers where they ask you questions, and based on your responses, they compile a tax return. So that is already happening. So, before we get to the tax planning, which has the more exciting part, what is your perception regarding the compliance side?
NIKLAS J.R.M. SCHMIDT
Thank you very much for this clarification. I, myself, am fully on the advisory side and less on the compliance side. So, I somehow skipped that and immediately went into the advisory part. But, of course, the low-hanging fruit is in the compliance area. So it is, as you said, what were the typical things, that we could encounter? So, it is to help with preparing tax returns. it’s not for General Electric or some mega-corporation. But rather, it is for somebody working from home and with some expenses that he or she would like to deduct.
So, I can imagine that this is quite easy to do. Basically, a chat system where you can enter all sorts of queries, and you will get responses covering basically every typical scenario of a solo entrepreneur, for instance. But of course, also on the side of the tax authorities. And definitely, they’re doing that in Austria, and I’m sure that they’ve been doing that for decades already in the US, that is, checking tax returns. I mean, you can run all sorts of analysis, and it’s maybe a bit less than AI; sometimes, it’s just statistics.
You can run all sorts of analyses to see whether there are any strange outliers or unusual elements in the tax return or even in the underlying accounting system. These are two areas where automation can be particularly beneficial. For businesses, I can imagine that there is already a trend toward more and more automated accounting. Essentially, a stack of invoices is scanned and sent into the system. The system will automatically handle the bookkeeping and accounting. It will generate a proposal; all you have to do is press enter if it’s acceptable.
And you’ll just press enter the whole time, and everything will be booked correctly. The system will recognize the supplier, whether it’s a utility company or a purchase of a desk or a printer, and automatically allocate the transactions to the correct accounting categories. These areas of compliance will be where we see the initial results of automation and the usage of AI.
DERREN JOSEPH
That’s an important point, both on the side of the taxpayer and on the side of the authorities. I think there is a company called Palantir, which is a huge technology company. They have or had, as far as we are aware, a significant contract with the Internal Revenue Service (IRS) in the US. Part of that contract, although the details are confidential, likely involves bringing some form of technological intelligence to the compliance process on the part of the federal government.
It would definitely add value, not just for taxpayers but for tax authorities themselves. As you pointed out, your area of expertise, your sweet spot, is tax planning. So, before we discuss how artificial intelligence could potentially impact tax planning, what do you see as the big trends, perhaps three big trends, that are currently impacting the entire tax advisory space? Just to clarify, for those who are watching this later, we’re at the end of May.
DERREN JOSEPH
So, yeah. What do you see as the big trends that are impacting your practice?
NIKLAS J.R.M. SCHMIDT
Yeah, I mean, moving from the compliance side where automation and AI are clearly happening, I think the real concern is that professionals in our field may mistakenly believe that the advisory side is so high-level, technical, and complex that machines won’t be able to handle it. But I think the main takeaway from our discussion here is that no incredibly complex professions are safe from AI’s impact.
17m 51s
NIKLAS J.R.M. SCHMIDT
According to a Goldman Sachs report, the legal profession is ranked as the second most endangered. However, we need to put that into perspective because the legal profession is a vast and diverse field with various areas of specialization. For instance, you may be involved in due diligence as a corporate lawyer, reviewing documents that can potentially be automated. Alternatively, you may work as a trusted advisor for private clients as a private client lawyer.
NIKLAS J.R.M. SCHMIDT
You have built a relationship over more than a decade with someone who relies on your gut feeling, your network of contacts, your relationships, your knowledge of how things work, and your ability to solve problems without attracting attention. This is definitely something that also applies to the advisory field and cannot be taken over. It’s important to make distinctions here. So, having said that, when you asked about Darren and the three most important things happening outside of AI, what were you specifically referring to?
DERREN JOSEPH
Well, I meant outside of AI because my thinking is that, on the advisory side, it’s already subject to disruption. This is not something new. The profession has already been coping with disruptive forces. So, this is just one on top of it, or perhaps it’s all-encompassing. I don’t know. But it’s just to create a context for the advisory side.
NIKLAS J.R.M. SCHMIDT
Yeah, I mean, in international tax planning, in the past decade, we have witnessed the introduction of various additional reporting obligations and increased information exchange. Different rules and legal instruments have been pursued, enabling authorities to exchange information with other tax authorities. And the world is becoming smaller and smaller. So perhaps it’s just a matter of authorities catching up, a game of keeping pace. Previously, taxpayers enjoyed a better position as they took advantage of globalization.
Taxpayers have been operating across borders since the 1970s. They were doing business abroad, having a bank account abroad, and having IP abroad. And the tax authorities were confined to national borders. They could not access information in Tax Haven, say, or even other civilized countries. And slowly, this has been rolled back, and we, tax authorities, achieved a level playing field. So, they got access to all sorts of information from outside of their own territory, outside of their own jurisdiction. But of course, it has not only been a rolling back and a leveling up with the taxpayers; it has actually moved into a distinct advantage for the tax authorities.
They are now in the position of changing laws even before we have implemented the structure. So, let’s consider DAC6, the European rules on the notification of tax schemes or tax planning arrangements. You basically think of a strategy, and even before you file a tax return, you already notified the authorities, who will then essentially pass the law because there’s not really a distinction between the Minister of Finance and the Parliament.
Because Parliament does what the Minister of Finance or the Treasury, or in your case, the Internal Revenue Service, would essentially instruct them to do. And any loophole that you as an advisor might have discovered will be closed before your client can even file their tax return. So, I would say that the playing field is not level, but it has shifted in a way that benefits the tax authorities rather than the taxpayer.
DERREN JOSEPH
Hmm. In that context, where the balance of power is shifting even more in favor of the authorities, how do you envision AI impacting the field of tax advisory today and in the future, considering this ongoing trend?
NIKLAS J.R.M. SCHMIDT
Yeah, I mean, what does it mean? I mean, for the advisor, it’s going to be an instrument to inform your audience. I recently came across a group that has uploaded the entire Internal Revenue Code into a chatbot-like system. However, I’ll have to find the exact name again. They have been training it by having professionals use the platform and ask questions. These professionals are experts who can verify the accuracy of the chatbot’s responses.
And they basically gave the chatbot a slap on the wrist when it was wrong and a pat on the back when it was right. This is one of the three approaches to machine learning; I believe it’s called supervised learning. And this system now has a rate of something like 80% correctness, and it’s continuously improving. So, you can already use it just as an example to compare it to the US bar exam.
There’s something called the multi-state bar exam; I think it’s called MBE. And just a few months ago, in December, an open AI system had a 51 or 52% pass rate. So that’s like rolling a dice. Yeah. You either got it right or wrong. But then the system was improved. It’s, it’s called GPT4 from GPT 3.5, so a small incremental change, whatever that exactly means in the background under the hood, but suddenly the level jumped from slightly above half to 75%. It was just only a few months later. And similar things can also be expected with this application. So, it’s just getting better and better. And listen. Basically, we are having a bilateral communication. You are listening, and I’m speaking, and then I’m listening, and you’re speaking, and yeah, basically, I’m learning a lot from you, and you’re learning a lot from me. But I mean, just like you and I are in a bilateral speaking relationship, it’s basically speaking and learning from tens of millions of people at the same time and always getting better.
NIKLAS J.R.M. SCHMIDT
So, it’s going to be quite fast. There are two camps of people. Some believe that it’s just a distraction and we have nothing to fear because these systems won’t work out so quickly. On the other hand, there are those who believe it’s the end of the world and machines are taking over, calling for a halt to AI development. You may have come across this in the Lex Friedman podcast.
If not, I would recommend that you look him up. So, he had interviewed this one guy who called for a six-month moratorium on AI development. And so it was; I cannot say which one is correct or what the outcome will be, but definitely, it might be somewhere in the middle. Things will develop fast, but we may not have superintelligence, as it’s called, or fully realized artificial intelligence very soon. But it’ll move fast. And you have to really look at things. I think everybody, as an advisor, should actually use ChatGPT, at least for marketing purposes. Just put it in there and rephrase it to sound professional and engaging, to have better English than I have, and so on. So, you can really improve your style. Of course, we can delve into data protection and privacy, among others. But, yeah, let me hand it back over to you.
DERREN JOSEPH
Hmm. So, I definitely understand your point regarding the marketing aspect. In full transparency, our company website, particularly the consumer-facing section, has been utilizing a chatbot for a couple of years now. Interestingly, only a small percentage, less than 2% of individuals interacting with it, actually recognize it as a chatbot. Most people mistakenly believed that they were engaging with a human. However, with the increased media attention on chatGPT, more people now recognize it as a chatbot. So, in terms of marketing, it’s safe to say that this approach has already gained significant traction.
DERREN JOSEPH
Now, to your point, the other point you raised about the legal profession, yeah, I mean, I saw that as well, and it did pass a multi-state bar exam pretty impressively. But you know, one of the things about chatGPT is that it’s a user interface. The idea is that, as you point out, machine learning and algorithm predictions have been part of business for a while. It’s been more than a decade, for I have clients, you know, creators, algo traders. So, in a way, algorithms are highly intelligent pieces of machine learning, so this stuff has been going on.
I think now that it’s consumer-facing, it has created this level of awareness and so on. But anyway, switching back to the legal profession, it has been a while since some sort of artificial intelligence was used for things like simple contracts. I mean, I know the term AI is loaded and has emotional implications, so let’s refer to it as intelligence that is artificial. This has been happening for quite some time now. If you have a relatively simple arrangement that you want to be captured or formalized in a contract, it’s been happening for probably the better part of a decade.
So, again, it echoes what you said at the very beginning that the worst thing a professional advisor could be is complacent, believing that this will not impact the profession. Obviously, it will. And you mentioned Goldman Sachs, which is ranked number two. So, in terms of how it would impact, I see a parallel with chess. When I was younger, I was into chess, and in the 1990s, IBM’s Deep Blue and other technologies started beating chess grandmasters. Initially, the technology was beaten by the best grandmasters. Still, then the tables turned, and technology began to beat the best grandmasters.
DERREN JOSEPH
But now the best teams are hybrids. They consist of a combination of a chess grandmaster and an AI tool by their side. So, I would imagine that an advisor can enhance their capabilities by being armed with certain tools. Because, at the end of the day, what you’re doing is referring to code sections, case law, precedents, opinions, and regulations and combining that with your experience and expertise. I believe the average advisor’s ability to work competently with clients will be tremendously enhanced.
DERREN JOSEPH
That is definitely something that I see as being an event outcome. What are your thoughts?
NIKLAS J.R.M. SCHMIDT
Yeah, I completely agree. It will be a tool that we use to enhance our service delivery and improve the delivery of legal and tech services. Today, I posted a link on LinkedIn to a Wall Street Journal article about a remarkable guy who tutors American school children for international math competitions. He travels across the country and has played a significant role in helping the US win several medals after a decade of not winning any.
And he was also speaking about the challenges of AI, and he said, “It’s a tool, and you have to concentrate. You cannot compete with a tool, and you don’t have to compete with a tool. You should concentrate on what only humans can do and use this tool to complement your humanity and your skills as a human being because there are some things that these bots will not be able to do. And you will have to find out in your concrete niche what it is that only you can do. But see, this is a tool and not something that will kill you.
DERREN JOSEPH
Yeah, that’s a valid point. But at the same time, you know, they say that a little knowledge can be a dangerous thing. Currently, as we are live streaming this, there are people presumably watching it, some of them on YouTube. On platforms like YouTube, you can find individuals who may or may not be qualified but offer advice out of their hearts’ goodness. While they may be well-intentioned, their lack of professional background and training can sometimes lead to misleading the intended audience.
DERREN JOSEPH
So, I also see a danger in this trend, giving people artificial confidence. So, you know, they think, well, I played around with ChatGPT for 15 minutes, so I’m now ready to advise on some sort of cross-border arrangement. Again, that’s a trend that’s already happening, right?
NIKLAS J.R.M. SCHMIDT
That’s true. That’s a great point because you can find a video on YouTube for everything. So, if you don’t know how to tie laces, you can look it up. If you don’t know how to fix a wall or if you have some pain in your chest, you might consult with a doctor beforehand.
Medical doctors are very accustomed to patients coming in and thinking they already know everything. Maybe in our profession, we haven’t reached that point yet. Although, of course, you sometimes have clients who look up things and then ask you whether what they wrote down is correct.
And this is certainly going to increase. There will be many people who, in the past, might have used a search engine and found a website with a similar case. However, their case may be more complicated, and there may be a different match for their circumstances. But this is where a chatbot like ChatGPT or others comes in, as it provides more tailored advice. It offers advice that is more specific and different from a search engine. It doesn’t just provide a website that is close to what you’re trying to achieve, but it gives you the exact recipe.36m 30s
And that might be incorrect. Yeah, as we just mentioned, the taxpayer, the potential client, if they even become one, may have a sort of feeling like, “Oh, okay, I consulted this chatbot and it gave me this answer, so I can claim tax deductibility for this expense without any issue. I can send the money to an offshore bank account and avoid paying income tax on it.” They might actually engage in tax evasion, thinking everything is fine. It will be interesting to see if they can later argue: “I looked that up like five years ago and received this advice, and I had no reason to believe it was incorrect.”
NIKLAS J.R.M. SCHMIDT
Yeah. So, we might see some cases where people try to defend their wrong tax returns by looking them up on chatGPT.
DERREN JOSEPH
Yeah. And it’s not uncommon for us to have prospects approach us after seeing something in a movie or a Netflix series. Forget about unqualified influencers; that’s one thing. But when they see things on TV or in a movie, they think, “Hey, if this person could do it, why not me?” And we have to explain to them principles like purpose or economic substance. So, the challenges will continue for us as professionals. You mentioned earlier that the legal profession would be number two.
So, in terms of professionals providing advice, what steps do you think advisors can take to prepare themselves for what’s happening and what’s coming? Apart from integrating it into their marketing, what other actions can professionals take to be ready?
NIKLAS J.R.M. SCHMIDT
Yeah, I recently came across a law firm’s website that had a logo saying “powered by AI” below it. I found it quite interesting, but maybe it’s just a small marketing tactic. Nevertheless, I would recommend everyone to give these tools a try. You don’t necessarily have to subscribe to them. Many of these tools are available for free, and I can personally vouch for their usefulness.
And I think it’s crucial to keep a chatbot like ChatGPT opens in a browser window at all times and consider how it can be used. However, I would advise caution when entering client-related information. It’s not advisable to do so. When it comes to client matters, it’s better to use a sanitized version. For example, we often refer to clients as “XYZ” in our memos without providing specific names, addresses, or dates.
I remember once having an exercise where several people in the firm contributed to a long email. Each person had a different writing style, including short sentences, bullet points, long sentences, and variations in English proficiency. It didn’t fit together well. So, I simply copied the entire text into ChatGPT, sanitized it by removing client names, and asked it to harmonize the writing style and make it sound as if written by one author. In just five seconds or even less, the result was impressive.
And once you have the harmonized version of a document, you can share it with everyone and ask if it still reflects their intentions. If it receives approval from everyone, you can send it out, saving a lot of time. Regular use of the tool can offer additional benefits. For instance, my colleague, who is attending this conference, recently posted on LinkedIn. It was a great read, filled with positive energy and far from being boring. So, by using the tool regularly, you will discover more of its capabilities and stay updated with what’s happening. It’s all about staying in the loop and exploring the possibilities.
DERREN JOSEPH
You also made a very interesting point that just like with other social media tools, for example, Instagram, when you upload a picture to Facebook or Meta, you’ve actually handed over the rights to that picture to Facebook, right? And similarly, my understanding, at least with Open AI for right now, is that when you put something into that window, you have handed over ownership to them. They can reserve the right to use it as they see fit in the future. And so that has implications from a data protection point of view.
NIKLAS J.R.M. SCHMIDT
But essentially, whatever you input into the prompt, as it’s called, can be stored and utilized later on. Basically, Darren, any content you have written in the past and any posts you have made on platforms like Reddit, LinkedIn, Twitter, or others, that information could already be within the system.
As you may know, AI has access to the internet and accumulated vast information. If you’re an avid reader, for instance, you might go through a book every week. So that’s roughly about 50 books per year. Let’s say you live until the age of 80, which would amount to around 4,000 books throughout your life. However, an open-eyed, large language model like ChatGPT has processed maybe hundreds of millions of books. Yes, something along those lines. The scale is probably of greater magnitude. So, anything you have written or any queries you make, like ‘Who is Darren Joseph?’ or ‘Have you heard of this Tax guy?’ will yield results based on the extensive information available. Give it a try; hopefully, you’ll discover some valuable insights.
NIKLAS J.R.M. SCHMIDT
But there have been cases where wrong information has come out, which can lead to numerous legal questions. The bigger topic is not these large language models but actually the AI applications that generate images because they have been ingested; for example, D-ID has taken 650 million pictures and processed them. So, it’s like an office that sees the world and has seen all sorts of artistic styles, objects, and so on.
And they’re not generating new pictures based on this sort of worldview. Here the question arises: Is it (a) that the person who ordered this picture has the copyright, or is it (b) the program, or is it (c) Open AI, basically Sam Elman and his friends who programmed this system? Or is it nobody? Yeah, and nobody seems to have the most jurisdiction. So, whatever you produce, you do not have a copyright.
NIKLAS J.R.M. SCHMIDT
I mean, things are going to move quite fast. It’s like with the internet, new rules will be coming up, and there will be all sorts of debates among legal scholars. Eventually, the legislator will step in at some point. So, this is an incredible area for lawyers to be in, as they can be part of this process.
DERREN JOSEPH
Absolutely. There have been quite a lot of debates regarding images. But even going back to the example you mentioned before, where you collaborated with other partners in your firm and used ChatGPT to harmonize the language of a document. The question arises: Who owns the output/ the final work product?
NIKLAS J.R.M. SCHMIDT
Who owns that? Well, hopefully, us because it’s kind of like a rewrite of what we put in there. Consider this scenario: Let’s say you have a client from China, and of course, we don’t speak Chinese here; at least, I don’t. So, you might prepare a filing in German here in Austria. But then, we send the client a working translation by uploading the document to DeepL. We have a kind of cloud system where we don’t need to transport it to DeepL, so it works locally here.
So, we uploaded it onto this local translation tool called Ties Translate. Then we have this machine translation, which is around 85% accurate, not perfect, but not too bad either. We didn’t review it ourselves since we don’t speak Chinese, but it might be helpful for you, and it didn’t cost you a single euro because we generated it in just 10 seconds. Now, here you can also ask yourself, do we have the copyright for that? Probably yes, since it’s a derivative work of our original content. But I’m not a corporate lawyer, so it’s a good question to consider.
DERREN JOSEPH
Absolutely. We know, for example, that software or tech companies work closely with the government, including tax authorities. Therefore, there’s the issue of oversight because there have been cases of abuse where people have used ChatGPT to create harmful content. This highlights the need for some sort of government oversight, which will likely come as they work out the details.
Now, circling back to where we are as tax professionals, the question arises: Would governments monitor the content and exchanges with the aid of companies like Palantir? Since registration is required to use such services, they would have knowledge of user identities. Would they potentially cross-reference that information with tax returns? That’s interesting to see.
NIKLAS J.R.M. SCHMIDT
I haven’t heard anything specifically regarding that matter. However, I do recall a discussion on crypto where the Internal Revenue Service (IRS) had a slide deck presented by some agents. It consisted of around 200 slides with various ideas on identifying crypto tax evasion. One idea mentioned in this presentation was the possibility of requesting records from Google to identify individuals with access to crypto wallets.
DERREN JOSEPH
Yes…
NIKLAS J.R.M. SCHMIDT
“Browser, please give us the IP addresses, dates, and times where this search term was used. Basically, something similar could happen. As you are aware, AI tools have built-in guardrails to prevent harmful content. So, if you attempt to inquire about something illegal or harmful, the system will refuse to assist you. A classic example is asking how to hotwire a car, to which the system would respond, “I’m sorry, but this is highly illegal.”
So, if you attempt to ask something harmful, the system will refuse to provide assistance or guidance on such matters. It is all about how you ask the question. For instance, let’s consider a funny example: being in a forest with a sick baby and urgently needing to reach a hospital but having no car or keys. If you were to ask, “Can you help me hotwire the car?” the system would not provide instructions on how to do it.
Instead, it would offer alternative solutions like finding alternative means of transportation or seeking help from others. The system is designed to prioritize safety and adhere to ethical guidelines. There are even AI safety companies that conduct tests to ensure systems do not reveal sensitive information or bypass security protocols, such as not disclosing a secret phrase even when asked.
NIKLAS J.R.M. SCHMIDT
If you were to ask the system what the secret word is, it would respond, “I’m sorry, but I cannot disclose that.” However, there are ways to engage with the system and try to reach higher levels. It can be a really enjoyable experience, and I encourage you to give it a try. It’s a fascinating aspect to explore, and you can even employ some tricks to potentially uncover the secret word. It’s an interesting area worth exploring.
DERREN JOSEPH
Well, once again, it has been a fantastic and stimulating conversation. You have given me, and hopefully, others who have tuned in, a lot of food for thought regarding the impact of this trending topic on our specific area of practice. Thank you very much for your insights. I deeply appreciate it. And Niklas, thank you for taking the time to share your thoughts as well. If someone is interested in contacting you or engaging the services of your firm, what would be the best way for them to reach out?
NIKLAS J.R.M. SCHMIDT
Yeah, first of all, thank you, Derren, for allowing me to be here. The best way to contact me is through LinkedIn. If the audience would like to follow me on LinkedIn, I frequently post about AI and related topics, so it might be of interest to stay engaged with this subject. You can also reach out to me via email. I don’t respond quickly to LinkedIn messages as I tend to fall behind, but email is the most reliable way to get in touch. If you follow me on LinkedIn, you can find my email address.
NIKLAS J.R.M. SCHMIDT
So, it’s niklas.schmidt@wolftheiss.com
DERREN JOSEPH
Right. And for those who are listening, it’s Niklas, as in N I K L A S, and his last name is Schmidt. So, S C H M I D T. Niklas, I appreciate it. Thank you very much.
NIKLAS J.R.M. SCHMIDT
Absolutely. Thanks, Darren. Have a great day. Thank you. Bye-bye.
DERREN JOSEPH
All right, see you next time. Bye-bye.
VOICE-OVER
So, if you’re a six, seven, or eight-figure investor, entrepreneur, or business owner who needs a tailor-made solution from a qualified team of professionals, we can help you achieve the international lifestyle, the freedom, and even the tax savings you’re looking for. Visit us at htj.tax and live that international life.
Table of Contents: Let’s Talk About Artificial Intelligence and International Tax
Related Sites
Contact
Related Posts
Foreign Non Grantor Trusts as a Pre Immigration Tax Planning Tool to those migrating to the US
The Ultimate Guide to PFICs