AI Tools Carry Accuracy and Privacy Risks for Financial Advice

Australian government site MoneySmart states that AI tools can produce inaccurate or biased information. It warns against relying on AI tools such as Copilot, ChatGPT, Claude and Gemini alone, especially for investing. A senior lecturer in the Department of Finance at the University of Melbourne says one reason these systems can provide inaccurate information is hallucination.

Hallucination occurs when the technology presents something inaccurate or completely fictitious as fact. A professor of business information systems at the University of Sydney says hallucinations and inaccuracies happen because the tools are optimised for plausibility rather than accuracy.

The systems generate responses that seem plausible to keep users engaged on the platform.

Bias and privacy concerns The senior lecturer says bias is another risk.

In the financial space, AI tools are likely to be influenced by traditional biases such as gender and race. Some research has suggested that AI might automate financial advice that suits the risk preferences of a young man. Other research found AI would stop recommending anything risky if it realised the user was a woman.

Simple detailed prompting can help counter this and remove a lot of those biases, the lecturer says. The professor says users should be concerned about privacy and the security of financial and personal information. It is difficult to keep track of the ever-shifting privacy policies of all these different companies, but users should assume that whatever they put in there will be used to train those models.

A financial educator and former financial planner recommends against sharing financial documents such as bank statements, pay slips and tax returns with AI.

Lower-risk uses and alternatives MoneySmart says publicly available, general-purpose AI can be a useful learning tool. This can include breaking down complex information, answering simple questions, suggesting topics to research next and highlighting general market trends.

The financial educator says using AI when researching a financial concept or approach is a lower risk way to use it. Users could ask questions such as what risks or considerations apply to a particular investment. The senior lecturer suggests using AI to help with tasks where users can check the work and provide timely feedback, such as building a household budget.

One problem with using AI for personal investing advice is that negative feedback which helps tailor the advice would be very delayed. The lecturer says some AI tools are better at particular tasks than others and the best one to use might come down to what is being asked.

The lecturer would avoid the free version for something as important as personal finance. Both the lecturer and the professor recommend cross-checking any information with other sources. Users can also use more than one AI tool and compare those responses.

Reliable and regulated financial advice can be very expensive to access, the financial educator says. Government websites such as MoneySmart have become more education focused and often include useful information and resources. The CEO of Financial Counselling Australia recommends heading to the MoneySmart website for information if experiencing financial difficulty.

For more tailored advice, people can seek the support of a financial counsellor through the National Debt Helpline website, which is free, independent and confidential. Banks and insurance companies also have an abundance of financial education information, the financial educator says.

There are also helpful podcasts and books, but users want to be able to vet that against an objective source.