Artificial intelligence laboratory OpenAI launched ChatGPT in November 2022. In the months since, the internet has been abuzz with discussions of how advanced AI could reshape society — including the financial services industry.
Some people see AIs like ChatGPT as tools that could increase productivity, while others see them as potential competitors in the job market.
Here’s what a financial advisor and a machine learning engineer think about the future of AI in financial services — and what ChatGPT has to say for itself.
What is ChatGPT?
ChatGPT is an AI program whose name stands for “chat generative pretrained transformer.” It’s an example of a large language model or LLM. (AI engineers really like acronyms.)
In simple terms, an LLM is a program that can respond to messages using patterns learned from training materials. ChatGPT’s training materials include hundreds of gigabytes of data and billions of words of text from books, websites and other sources.
Could an AI like ChatGPT take peoples’ jobs?
AI programs such as ChatGPT can master human languages and coding languages — and they can’t accidentally forget information the way humans can. With that in mind, some people worry that AI could automate away a lot of jobs in the coming decades, including “knowledge jobs,” which were previously considered safe from conventional, factory-robot-style automation.
In November 2017 — five years before the launch of ChatGPT — consulting firm McKinsey published a report estimating that between 400 million and 800 million workers globally could be displaced by automation by 2030.
LLMs have already encroached on some knowledge-based professions, such as financial journalism. For example, the technology website CNET announced in January 2023 that it had published several dozen AI-written personal finance articles.
To see whether financial advisors could be next, the author of this article asked ChatGPT a series of personal finance questions.
Does ChatGPT give good financial advice?
ChatGPT, for its part, is modest about its financial advisory capabilities. “As a language model AI, I can provide information and insights on personal finance, but I cannot provide personalized financial advice,” it said in a written interview.
(The author asked ChatGPT for consent to quote it in this article. “I do not have personal opinions or control over my use,” it said, although it did ask for its comments to be attributed to ChatGPT and OpenAI.)
Below are several quotes pulled from a conversation between the author and ChatGPT, with comments from Rick Nott, a certified financial planner with California-based registered investment advisor LourdMurray.
Question: “What are some of the best investments for a young person who wants to start saving for retirement?”
ChatGPT started its answer by recommending employer-sponsored retirement plans like 401(k)s and a traditional or Roth IRA.
“A traditional IRA offers tax-deductible contributions and tax-deferred growth, while a Roth IRA offers tax-free withdrawals in retirement,” it said.
It also recommended index funds, suggested individual stocks and exchange-traded funds “if you’re comfortable with a higher level of risk,” and urged caution when selecting individual investments.
“Be sure to do your research and consider working with a financial advisor to develop a diversified investment portfolio,” ChatGPT said.
Nott pointed out that although the question asked for “some of the best investments,” ChatGPT’s first two answers — 401(k)s and individual retirement accounts — are investment methods, while only the final two — individual stocks and ETFs — are actual investments.
“Technically, it answered the prompt correctly. But for someone who doesn’t have the requisite knowledge, that might be a little confusing,” Nott said.
Question: “What are some good ways to increase my income tax refund?”
ChatGPT suggested keeping records and receipts and making sure to “claim all eligible deductions, such as charitable donations, mortgage interest, state and local taxes, and work-related expenses,” and noted that medical expenses “which exceed a certain percentage of your income” could also be deducted.
The AI also recommended contributing to a 401(k) or IRA, taking advantage of tax credits such as the earned income tax credit and child tax credit, and adjusting your withholding.
“If you received a large refund this year, you may be able to adjust your withholding so that you receive more in your paycheck throughout the year instead of waiting for a big refund at tax time,” ChatGPT said.
Nott said in an email interview that these answers are “not complete.”
“For most people the standard deduction will be higher than itemizing. So by taking all deductions you might inadvertently get less of a tax refund. Similarly, medical expenses usually only matter if you itemize deductions,” he said.
He said that ChatGPT’s point about withholding is “probably the best and most widely-applicable answer,” but it’s missing some critical caveats.
“This doesn’t address 1099/independent contractors who have no taxes withheld and may have to make estimated tax payments,” Nott said.
On the whole, Nott described ChatGPT’s answers as “70% or 80% accurate.”
What do financial advisors think about ChatGPT?
With that in mind, Nott said he isn’t worried about losing his job to an AI such as ChatGPT.
Well-rounded financial advisors, he said, need to gather information on the “needs, values, goals and important relationships” of clients.
“That is very much a therapist-style line of questioning. There’s a vulnerability — which is where you get the real, true answers to things — that you have to elicit through that initial conversation,” he said.
Nott isn’t sure that humans will ever trust an AI enough to show that vulnerability.
“My view is that unless humans are comfortable enough with an AI… it’s gonna be a poor driver of the things we’re really doing as wealth advisors,” he said.
What do AI engineers think about AI financial advisors?
Matthew Alhonte is a machine learning engineer for health care technology company Actium Health and has worked on AI model design. He said in a written interview that trust might not actually be a problem for a hypothetical AI financial advisor.
“There are studies showing that people are often a lot more willing to be honest with a machine than a person,” Alhonte said.
A 2014 study by researchers at the Institute for Creative Technologies and Bard College backs up that claim. It found that patients in health screening interviews reported a “lower fear of self-disclosure” when they were told that the interviewer was an automated program.
Alhonte noted that an LLM could be trained with human-curated data to ask sensitive financial planning questions like, “Do you have any disabled dependents who’d need to retire on your savings with you?”
He also disagreed with the pop-culture perception that AI comes across as robotic. He said that LLMs such as ChatGPT are “actually pretty good at emulating the tone and style of writing” of a human, which could help an AI financial advisor build trust with clients.
However, Alhonte cautioned that LLMs “have very low reliability.”
He said that these systems are “good most of the time but catastrophically wrong one time out of 100,” which “probably wouldn’t be acceptable for something where it matters to be catastrophically wrong.”
What does ChatGPT think?
“It’s possible that advanced AI technology could play a role in the financial advisory industry in the future. However, it’s unlikely that AI will completely replace human financial advisors in the near future,” ChatGPT said.
But given the level of activity in the AI industry, that “advanced AI technology” may not be far off.
ChatGPT was only released a few months ago. In an interview with StrictlyVC, OpenAI CEO Sam Altman didn’t confirm or deny that OpenAI could release GPT-4 — the successor to the GPT-3.5 LLM, which powers ChatGPT — this year. Google is also testing an LLM-based ChatGPT competitor called Bard. It plans to let developers start integrating Bard into products next month and open it to the public “in the coming weeks.”
For now, ChatGPT seems to agree with Nott that some aspects of the financial advisor job need a human touch.
It said financial advisors need “emotional intelligence, empathy, and the ability to build trust and rapport,” qualities that “cannot be easily replicated by AI.”