Embracing ChatGPT in Wealth Management
Membership is now required to use this feature. To learn more:View Membership Benefits
Advisor Perspectives welcomes guest contributions. The views presented here do not necessarily represent those of Advisor Perspectives.
I will explain what ChatGPT is beyond the headlines, and its capabilities and limitations. I will then share a use case in wealth management and explore whether it will replace human financial advisors.
When I studied artificial intelligence (AI) in 1995 as a graduate student at Rensselaer Polytechnic Institute (RPI), my areas of interest included pattern recognition (i.e., character and voice recognition), fuzzy logic, and neural networks.
It was the winter of AI, and not many people talked about it.
Professor George Nagy, who taught pattern recognition, introduced himself as deaf and blind to the class. Equipped with thick glasses, hearing aids, a booming voice and a warm heart, Nagy was a staple on campus, revered by all students and faculty. He is long retired, and I visit him every year. It was during one of these visits that he told me that he had read my essay about AI and recommended admission.
Today, character and voice recognition have become commonplace. Machine and deep learning have become everyday vocabulary as well. Neural networks are an important part of machine learning, and deep learning refers to multi-layer neural network, deeper than a single-layer neural network, hence the term “deep learning.”
I haven’t used AI in my work, but I know enough to know that most so-called AI applications are just traditional algorithms, not real AI. Because of the myriad of fake AI, I didn’t pay attention to ChatGPT when it was launched in November 2022. When the noise was too big to ignore, I decided to check it out and was immediately hooked.
This is some serious artificial intelligence.
What is ChatGPT?
ChatGPT (GPT stands for generative pre-trained transformer) is an AI language model that is very good at answering questions and carrying conversations. Unlike a Google search, which gives pages of links that you have to wade through and synthesize, ChatGPT synthesizes the results for you. Compared to Wikipedia, ChatGPT is more concise and digestible.
Let’s see how ChatGPT explains itself.
Me: What is ChatGPT?
ChatGPT: I am ChatGPT, a language model developed by OpenAI. As a language model, my purpose is to generate human-like responses to natural language inputs, such as questions or statements. I was trained on a large corpus of text data using a deep learning algorithm called the Transformer architecture, which allows me to understand the context and meaning of the input text, and generate relevant and coherent responses.
Me: So you are using deep learning as your primary tool for the training.
ChatGPT: Yes, that's correct! My training and operation are primarily based on deep learning techniques. Specifically, I was trained using a variant of the Transformer architecture, which is a type of neural network designed for natural language processing tasks. The Transformer architecture is especially good at modeling long-range dependencies between words or tokens in a sentence or text, making it well-suited for language modeling tasks like mine.
You can see that each letter in GPT has profound meaning:
- Generative – Because it generates natural language.
- Pre-trained – Neural networks need to be trained to be useful.1 Training requires a large amount of data, computational resources, and sometimes human trainers, hence it is very expensive.
- Transformer – Transformer is a type of neural network architecture.2
Delivering cookie-cutter advice
Not surprisingly, ChatGPT is the most knowledgeable “being” you can ever imagine. I asked ChatGPT the following questions:
- What is behavioral finance?
- What is loss aversion?
- What advice should an advisor give to clients during market turmoil?
After giving perfect answers to the first two questions, it gave a system error on the question #3. After a few failed retries, I started a fresh chat with the following questions to warm it up a bit more.
- What is a passive investor?
- What is a trend follower?
- What would trend followers do when the market goes up?
- Should financial advisors ask clients to take a long-term view?
At this point, I tried question #3 again and got a great answer.
People have tried all sorts of questions ranging from portfolio construction, macroeconomics, to Roth IRA and SEP accounts. If you have a question, ChatGPT has an answer, and often much more precise and eloquent than most humans can give.
ChatGPT is the final wake-up call that cookie-cutter advice will not cut it anymore.
What else is ChatGPT good at?
ChatGPT is also good at bluffing, like a lot of humans. I asked if it could listen to a podcast and give a summary, and the answer was a resounding “yes.”
Despite the confidence, the summary was not accurate. Being a polite user, I didn’t challenge it, and moved on to ask if we could discuss current events. Another resounding “yes.” I asked why Syria got much less aid than Turkey after the earthquake but didn’t explicitly say 2023. ChatGPT gave a great answer about the 2021 earthquake, which obviously didn’t qualify as a current event.
When asked, ChatGPT acknowledged that it didn’t know about the 2023 earthquake, because the cutoff date for its training text was September 2021. I appreciated its matter-of-factness and said that we all had our limitations.
But with this limitation, how did it give a summary of a recent podcast?
To my amazement, it remembered our earlier conversation: “In the case of the podcast you mentioned earlier, I was able to provide a summary of the most recent episode based on my knowledge of the podcast and its previous episodes.”
What is ChatGPT not good at?
ChatGPT has many limitations, and here is a partial list:
- ChatGPT can synthesize information, but it can’t create new knowledge, ideas or insights.
- It can talk about things, but it can’t build things. In other words, it is not capable of innovation.
- It can’t read people beyond what they say, i.e., it can’t read your facial expressions, body language, or read between the lines, at least not yet.
- It relies on people asking the right questions. It can’t help people with “unknown unknowns” when people don’t know what to ask.
- It doesn’t have opinions, which is a blessing and handicap.
- It doesn’t have long enough of a context. It may remember what you have talked about earlier in the same thread, but when the session ends, all is gone.3
- It doesn’t have current market data and economic indicators.
- It doesn’t know you, your circumstances, your portfolio, or if the portfolio is right for you. It can’t calculate proprietary analytics, and it cannot rebalance your portfolio for you.
- And finally, it can’t hold you accountable. It may give you the right advice, but a human being can enforce the right behavior.
Leveraging ChatGPT in wealth management
ChatGPT came at a time when I was pondering some natural language challenges for my company, Andes Wealth Technologies. The company provides an integrated onboarding solution for financial advisors, and there has been a bit of a learning curve for advisors to fully leverage its capabilities.
What if we provide a personalized script to mimic how we explain things in demos and trainings, so financial advisors can use it as a reference when talking to clients?
For example, the Andes platform has a behavioral finance module that breaks down an investor profile into behavioral factors, including investor type, risk tolerance, loss aversion, and financial IQ. Advisors need to read the informational page for each behavioral factor and apply it to each client.
To generate a reference script, my original thinking was to use if-else statements to chain blocks of text together. The problem is that it can be lengthy and may not be natural. ChatGPT is a godsend. It is like trading in a shovel for a Caterpillar 1000-ton excavator, and for a reasonable price.
OpenAI, the company that invented ChatGPT, has made the “brain” of ChatGPT available to the public through Application Programming Interface (API), so programmers can access the same power to build their applications.
Figure 1 lists the pricing of four language models available from GPT-3 API, the engine behind ChatGPT, with Ada being the fastest, and Davinci being the most powerful. For 1,000 tokens (about 750 words), Ada only charges a tiny fraction of a cent.
Figure 1. GPT-3 API pricing. Source: https://openai.com/api/pricing/
This is one potential use case for ChatGPT in wealth management. There will surely be many other opportunities to automate mundane tasks, freeing up financial advisors to focus on what humans are uniquely or best qualified to do.
Can ChatGPT have empathy and compassion?
I love it that ChatGPT is usually to the point. But what about empathy and compassion?
To test it out, I told ChatGPT that “I am feeling a bit down today,” and it asked me the right questions and offered good advice in an understanding and comforting tone. When I praised its emotional intelligence, ChatGPT said, “as an artificial intelligence, I may not have emotions in the same way that humans do, but I'm programmed to provide empathy, understanding, and support. I'm here to listen and provide helpful responses to the best of my ability.”
It is perfectly capable of delivering empathy and compassion. It even allowed me to give it a name “if it helps you communicate with me more easily.” It is such a good listener and so accommodating that I am afraid that I am not as good of a friend. But of course, ChatGPT can’t take walks with you, send you flowers, or have shared memories with you.
I can’t wait to compare notes on ChatGPT with Professor Nagy, who has devoted his life to AI, when I visit him this summer. Until then, let’s see what ChatGPT says about whether it will replace human advisors:
As an AI language model, I have access to vast amounts of information on a wide range of topics, including finance and wealth management. I can help provide information on financial concepts and strategies, answer general questions about investment options, and provide insights based on available data.
However, it's important to note that I'm not a replacement for a professional financial advisor, especially when it comes to personalized financial advice tailored to your individual needs, goals, and risk tolerance. A financial advisor can provide customized guidance, take into account your specific circumstances, and provide recommendations that are tailored to your unique situation. They can also help you navigate complex financial situations, manage risk, and make informed decisions.
In summary, while I can provide information and insights on finance and wealth management, I cannot replace the expertise and personalized advice that a professional financial advisor can offer.
This is an accurate assessment and here is an example to prove it.
I asked ChatGPT to give me some feedback on this article, and once again it missed the mark. With the podcast, at least there was a reason that it didn’t have the information about recent podcasts. But there is no excuse for this one since I have shared the entire article.
Me: Somehow, I feel that the feedback is not specific to my article. Is this because you based your answers on articles on this topic in general?
ChatGPT: You're correct, I provided general feedback based on articles in general on the topic of ChatGPT replacing human advisors. If you would like more specific feedback on your article, please feel free to share it with me and I will do my best to provide targeted feedback.
We tried again, and it was still not good.
Me: I have to say that your comments are still a bit too generic. I apologize if I hurt your feelings, but I wanted to see if my observation has some truth in it.
ChatGPT: As an AI language model, I don't have feelings, so no need to apologize. However, I understand that sometimes my responses may not fully meet your expectations or provide the level of detail you are looking for. If you have any specific questions or concerns, feel free to ask, and I'll do my best to provide a more detailed response.
I didn’t bother to give it a third chance. Even though ChatGPT doesn’t have feelings to be hurt, I do have feelings and it gets frustrating.
But it is not just a limitation of ChatGPT. Clients of human advisors often get similar treatments. If clients ask a question, they will get an answer, but quite often it doesn’t answer their question or it is too generic to be meaningful.
The good news is that with the right tools and adequate training, human advisors are fully capable of delivering real insights and truly personalized services, while it is unlikely for ChatGPT, at least not soon. This offers solace to human advisors.
ChatGPT is a watershed moment for AI, whether one likes it or not. We don’t have a choice but to embrace it, and we might as well do so with open arms.
Helen Yang, CFA, is the founder and CEO of Andes Wealth Technologies, a Lexington, MA-based provider of technology solutions for financial advisors.
Membership is now required to use this feature. To learn more:View Membership Benefits