Written by Daniel G. Goldstein, Senior Principal Research Manager at Microsoft Research in New York City.
Philosopher Derek Parfit once observed that we often neglect our future selves due to a failure of imagination or false beliefs. In the context of retirement planning, this tendency is particularly concerning. When we’re young, it’s challenging to believe that we will one day grow old and face the consequences of the financial decisions we make today. Additionally, envisioning life in the distant future can seem too daunting. Traditionally, financial advisors have helped bridge this gap by guiding individuals to consider their future selves. This process can range from simply asking people what they hope to achieve in retirement, to creating detailed projections of future wealth, to demonstrating how today’s decisions impact tomorrow’s finances.
In the past, researchers have explored ways to make the future easier to imagine, thereby strengthening the connection between the present and future selves. In 2008, colleagues and I developed the Distribution Builder tool, which allowed users to create a probability distribution of retirement wealth outcomes and simulate the experience of having a random draw from that distribution determine their fate. In the 2010s, we took a VR approach, using age-progression software to show individuals how they might look in retirement. This technology enabled us to assess how such visualizations could influence saving behavior in both real and hypothetical scenarios.
Today, with the advent of powerful generative AI, we can go beyond calculating future wealth levels or age-progressing faces. AI has the capability to create photorealistic images which can be amazingly persuasive. This technology has already been used to help people reimagine their cities, including envisioning more pedestrian-friendly environments, and leading to increased support for new policies. In the realm of retirement planning, AI could also be harnessed to create customized illustrations depicting life in retirement under various conditions, showing what a future home or vacation might look like. These AI-driven visualizers could become powerful tools for financial advisors, helping them engage clients more effectively. Researchers Shlomo Benartzi, Hal Hershfield, and I are designing interventions to test this hunch.
Beyond serving as a tool for financial advisors to counsel relatively wealthy individual clients, this technology is likely to be quickly democratized by platforms that service more ordinary investors in 401(k) plans, individual retirement accounts, and other broad-based investment vehicles. As a result, a rapidly growing proportion of the population is likely to increasingly use AI to perform some of the tasks traditionally offered by individual financial advisors to their clients.
The idea of an AI as a broadly available financial advisor presents both promise and peril. What could go wrong? For one, AI might provide incorrect information or exhibit biases against certain groups. While an unspecialized large language model might do this out of the box, I believe that future interactions will not rely on general-purpose AIs, but rather on domain-specific applications that use AI as a component. An AI designed to be a financial advisor could be trained on carefully vetted materials, avoiding the pitfalls of misinformation found on internet forums. It would perform complex calculations by leveraging external calculators, and all the financial forecasting tools and data sources developed over the last century would be integrated into these applications as part of the AI interface. Moreover, AI applications can be equipped with guidelines and guardrails to help avoid certain biases.
No doubt, some errors and biases will inevitably slip through—especially when the technology is new, but these issues can be addressed over time. In engineering, logged problems tend to get fixed, and because AI interactions can be logged and audited, such errors and biases can be identified and corrected. This iterative improvement process mirrors what we’ve seen with other AI innovations, such as self-driving cars, which, while still far from perfect, has improved dramatically in a very short period and generally has produced a better safety record than human drivers.
While concerns about accuracy and bias are critical, AI also holds tremendous promise. Ironically, AI systems can exhibit empathy—one study even found that AI was perceived as more empathetic than human medical professionals. Additionally, AI can be more accessible than a human advisor, available at any time of day, and infinitely patient in answering questions. One of AI’s strengths is its ability to translate complex, technical language into plain language that people can easily understand.
The question is not whether people will turn to AI financial advisors—financial questions are surely among the millions of queries submitted to consumer-facing tools like ChatGPT, Microsoft Copilot, and Claude every day. The real question is how we can design AI-based financial and retirement planning applications to provide the best possible advice.
Views of our Guest Bloggers are theirs alone, and not of the Pension Research Council, the Wharton School, or the University of Pennsylvania.
