AI's Investment Illusion: Sophisticated Answers Hide Fragile Assumptions

There is something almost magnetic about the promise of algorithmic investing. A few keystrokes, and within seconds, a screen fills with precise percentages, economic justifications layered with sophistication, charts arranged with care, explanations of interest rates and inflation and diversification. It looks intelligent. It sounds intelligent. The name itself—artificial intelligence—seems to carry an implicit guarantee: surely it should make fewer mistakes than we do.

So more investors have begun handing their financial decisions to AI tools. The appeal is understandable. But here is what troubles me: many people ask for a portfolio, yet few have actually tested whether it works. How do you criticize something that sounds intelligent even in its name?

I decided to test it myself. Using ChatGPT, I asked it to build three investment portfolios for Brazil as if we were standing on December 31, 2019—one conservative, one moderate, one aggressive. The only constraint: use only information available up to that date. The response read like a polished strategy report. The allocations made sense. For the aggressive investor, the AI loaded up on stocks, small-cap funds, real estate investment trusts, and long-duration inflation-indexed bonds. The logic held: Brazil seemed poised for a structural shift toward lower interest rates, investors were rotating from fixed income into riskier assets, and the domestic market looked ready for years of gains. The conservative portfolio stayed concentrated in floating-rate bonds and short-term inflation protection. The moderate portfolio split the difference. Everything was coherent.

Then came the five years from 2020 to 2025. None of the three portfolios beat the CDI—the basic Brazilian interbank rate that serves as a benchmark. Not the conservative one. Not the moderate one. Not the aggressive one. The conservative portfolio returned 57.8 percent over the period. The CDI returned 69.4 percent. The aggressive portfolio, which accepted far more volatility and risk, returned only 37.7 percent. Even more stinging: all three fell short of the IHFA, an index tracking the average performance of Brazilian multimercado funds—the very thing investors often try to beat by going elsewhere.

The AI had not failed because it was incompetent. It had done exactly what it was trained to do: reorganize the past brilliantly. But the financial market has a cruel talent for destroying narratives that seem perfect.

I ran the test again, this time starting from the end of 2023, thinking perhaps the AI would learn from the changed landscape. It did shift its language. It now favored inflation-indexed bonds more heavily, argued that real interest rates might stay elevated longer, and shortened time horizons for aggressive investors. The logic sounded excellent again. Yet the new allocations barely changed from the first set. Compare them side by side and you see the same basic structure, which suggests either that the AI is simply scanning what banks recommend online, or that it is applying imported models that do not fit Brazil. Those who thought they were escaping traditional bank advice by asking an algorithm may have walked straight into the same trap.

The results for 2024 and 2025 told the same story: all three portfolios lagged the CDI and the multimercado index. This is the real problem. Investing is not about connecting past information or retrieving data from the internet. Investing is about dealing with what has not yet happened. It requires building scenarios, understanding what an investor actually wants, and developing sensitivity to risk. The genuine danger of AI in finance is not that it fails—it is that it creates the seductive illusion that uncertainty has finally become manageable. A quick answer delivered with confidence does not make the answer more intelligent or accurate. AI can be genuinely useful for organizing data, comparing strategies, building dashboards, spotting inconsistencies, and helping investors think more clearly. The trouble begins when we confuse sophistication with the ability to predict. Many investors are searching in artificial intelligence for something it may not yet be capable of delivering: a way to transform uncertainty into comfort. That search may end in disappointment.