From Information to Delegation: Mapping Human-AI Financial Decision Making
This paper introduces a behavioral framework to measure human-AI decision authority, revealing through an analysis of 1.5 million interactions that while consumers heavily utilize AI for financial information and judgment, they rarely delegate actual financial execution to it.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you're standing in a giant, bustling library where the shelves are made of light and the books are made of conversations. For a long time, people have been asking the librarians (who are actually super-smart computer programs called AI) simple questions like, "Where is the history section?" or "What's the weather like?" This is the world of Information Retrieval: getting facts from a machine. But recently, these librarians have started acting more like partners. They don't just point to a book; they start helping you write the story, plan the plot, or even decide which character to save. This shift is happening in the world of Human-AI Decision Making, a field that studies how we share the mental load of making choices with computers. The big question isn't just "What does the AI know?" but "How much of the steering wheel are we letting the AI hold?" Are we just using it as a map, or are we letting it drive the car? This matters because as AI gets better at making choices, we need to understand if we are still the captains of our own financial ships or if we're slowly handing over the keys.
This paper, titled "From Information to Delegation: Mapping Human-AI Financial Decision Making," dives deep into the wallet of the digital age to see exactly how people are using AI to handle their money. The authors, a team from Stripe Partners, treated over 1.5 million real conversations between users in the United States and India and two popular AI assistants, ChatGPT and Gemini. They didn't just count how many times people asked about money; they built a new kind of "behavioral microscope" to look at how people interact with the AI. They sorted every chat into three levels of trust, like a video game difficulty setting:
- Level 1 (Inform): The AI is a librarian. You ask, "What is the interest rate?" and it answers. You do the thinking.
- Level 2 (Shape): The AI is a coach. You ask, "Should I buy this stock?" or "How do I fix my credit score?" and the AI gives advice, compares options, or helps you plan. You still make the final call, but the AI is shaping your decision.
- Level 3 (Act): The AI is a driver. You say, "Move my money to the best account," and the AI actually goes and does it.
The researchers found that while financial services are already a huge part of how people use AI (about half of the users in their study talked about money at some point), we are mostly stuck in the "Librarian" and "Coach" zones. The data shows that consumers overwhelmingly use AI to inform their choices (about 63.5% of chats in the US and 72.1% in India) and shape their strategies (about 58.6% in the US and 49.7% in India). People love asking AI to compare credit cards, explain tax rules, or figure out the best way to pay off debt.
However, the paper explicitly rules out the idea that people are ready to let AI take the wheel. The "Driver" level, where the AI actually executes a financial transaction or makes a decision on its own, is incredibly rare. In the US, only 0.3% of financial chats were about delegation, and in India, it was a tiny 0.1%. Almost none of these were about the AI making big, autonomous choices; they were mostly simple, instruction-based tasks like setting up a budget alert or tracking spending. The authors suggest that despite the hype about "agentic AI" (AI that acts on its own), people are currently very cautious. They want a smart assistant to help them think, but they are not ready to let the robot sign the checks.
The study also highlights some fun differences between the two countries. In the US, people used AI mostly to solve problems with their bank accounts, fix payment issues, or navigate insurance and public aid. In India, the conversations were more focused on investments, stock market research, and learning about finance. But the main takeaway is the same for both: we are in a "Shape" era. We are using AI to build better financial maps and plan our routes, but we are still the ones holding the steering wheel, ready to take over if the road gets tricky. The authors conclude that this framework gives us a baseline to watch how things change in the future, suggesting that for now, AI is a powerful tool for judgment, not a replacement for it.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.