You would need to save between “70 and 80% of your current income” to retire at 60. Forget CELIAPP and RAP for buying a first home, they don’t exist. And it is advisable to rebalance your investments at least once a year.
Don’t rush to a financial planner right away: there is truth, falsehood and a wide spectrum in between in the previous advice. They come from the answers to 16 personal finance questions that The Press posed to generative artificial intelligence (AI). For the occasion, we chose the paid version of Gemini, Google’s brand new conversational agent.
We then collected comments from two financial advisors, Raphaël Hainault, from the Hainault-Harvey-Simard team, and Benoit Chaurette, from Banque Nationale Gestion Privée 1859. The two provided 12 of the 16 questions. We present six of them in detail, with Gemini’s answers, in the next screen.
Between relevance and blunders
First observation: the AI made gross errors, especially with very specific questions. She notably stated that one could receive the old age security pension from the age of 60, or that a dividend collected in the United States is taxable at 50% in Canada. She sometimes omitted important details, such as the existence of the tax-free savings account for the purchase of a first property (CELIAPP), and completely invented an “Order of Financial Planners of Quebec”.
But Gemini was quite relevant when it came to very general advice on financial behavior, such as stock allocation according to risk tolerance or preparation for retirement.
And he dared, very rashly and with questionable relevance, to give us five recommendations to purchase shares of very specific Canadian companies.
PHOTO HUGO-SÉBASTIEN AUBERT, LA PRESSE ARCHIVES
Raphaël Hainault, wealth management advisor and assistant portfolio manager in the Hainault-Harvey-Simard Team
So, your verdict, gentlemen? “I notice that Gemini understands the broad outlines of the questions and the answers are often appropriate, but they sometimes lack accuracy,” says Raphaël Hainault.
Artificial intelligence simply answers the questions asked, but cannot go further, sometimes suggesting things that the customer may not have even thought of, due to lack of knowledge.
Raphaël Hainault, of the Hainault-Harvey-Simard team
This is a failing that Benoit Chaurette knows well, who was not in his first experiments with AI. It was he who most severely undermined Gemini with the fictitious example of “Mary and John” receiving a dividend of US$300 (see next screen).
“On questions of a very general nature, he answers very interesting things. In 100 words, what was asked of him, he gave a very honest summary. But the more we get into specific cases, the more he gets carried away and sometimes arrives at false conclusions. At times I was pleasantly surprised, at other times disappointed. »
Shades sought after
One of the faults of Gemini, like ChatGPT, is not revealing its sources of information. Although he adds links for additional information at the end of his answers, we do not know where he draws his conclusions from. When it comes to stock buy recommendations, for example, “did he do real analysis, or did he copy it?” asks Mr. Chaurette. When are they from? Are they still relevant? We don’t know what’s under the hood.”
PHOTO PROVIDED BY BENOIT CHAURETTE
Benoit Chaurette, senior advisor at Banque Nationale Gestion Privée 1859
Gemini, for example, responded with aplomb to a question on “the cost of a child”, $250,000 on average from birth to 18 years old, according to him. “I would like to understand how Gemini calculates these expenses and the sources of information,” retorts Mr. Hainault. For example, in my opinion the cost of housing does not necessarily increase with the arrival of a child, unless you move to a larger house to accommodate him or her. Same thing for education which, in Quebec, is inexpensive. On the other hand, other positions are possibly underestimated, such as leisure activities. »
The two experts, clearly, are not worried about seeing generative AIs like Gemini and ChatGPT replace them.
“If we ask questions about human anatomy, for example, there will be precise answers: an arm is an arm and a toe is a toe, whether we are in Europe or in Quebec,” explains Mr. Chaurette. If we ask tax questions, there are a lot of different legislations and different environments, the tool is all mixed up. »
“It remains an algorithm, without all the nuances that sometimes make all the difference,” recalls Mr. Hainault. I believe that there is still a way to go to amalgamate all the available data and make it into something coherent that provides precise suggestions, rather than generalities. »