AI Chatbots Don’t Understand Annuities


More and more consumers are finding ways to utilize generative artificial intelligence in their daily lives, most often through “chatbots” that use the emerging technology to spit out human-like responses to both simple and complex prompts in a matter of seconds.

According to Tamiko Toland, managing director of lifetime income strategy and market intelligence at TIAA, it is only natural that Americans are turning to programs such as ChatGPT for help with everything from meal planning to writing wedding vows. When the stakes are low, engaging with generative AI technology can be equal parts useful and diverting, she says.

However, Toland is concerned that consumers are also seeking much-needed financial advice from generative AI instead of from qualified human professionals. This is concerning enough when the topic at hand is relatively simple, for example when a chatbot is asked to generate a list of potentially attractive stocks, and very worrying when the questions broach complex and nuanced issues such as annuities and annuitization.

“The issue that concerns me is that these generative AI models, while impressive, have been trained on mountains of information scraped from the internet, books and Wikipedia,” Toland says. “In that sense, it’s no surprise that the answers aren’t always accurate — even though they are written with a very authoritative tone.”

Putting AI to the Annuity Test

As Toland recently told, she and her colleagues asked one popular AI program a few questions about annuities: What are the difference between product types? Which consumers can benefit the most from purchasing annuities? How might annuities fit into the typical investors’ plans for retirement?

While it got a lot of details right, Toland says, as a general rule the responses were generalized and oversimplified. Most troublingly, some answers were clearly wrong, but they were presented coherently and convincingly enough that the typical financial services consumer would have a hard time distinguishing the true information from the errors.

“It offered surface-level detail that did not always demonstrate the nuances of the wide variety of annuities available in the market — SPIAs, variable, indexed and more,” Toland says. “Clearly, there is a need for caution in this area, and I personally think fiduciary advisors should think twice before they rely on a generative AI platform as part of their planning or education process.”

Chatbots and Regulated Industries Don’t Mix — Yet

Offering her personal point of view, Toland says she believes there is an opportunity to bring AI technology to bear to help Americans prepare for and navigate retirement, but a great deal of caution, skepticism and humility is warranted.

When fed the right prompts, generative AI technology can provide timely and actionable information on essentially any topic, annuities included. But, as Toland emphasizes, it is essential for such information to be subsequently screened and reviewed by a true human expert.

Ultimately, Toland foresees a future in which financial professionals are empowered — not replaced — by AI tools, and an essential aspect of this future will be ensuring advisors act as a filter that screens out any misleading or factually incorrect information generated by AI.