AI · Ethics
Can artificial intelligence make mistakes? How to check
It can, and it errs with great confidence: it invents sources, numbers and laws. See why this happens and a simple method to check before you trust it.
Yes, artificial intelligence makes mistakes, and its mistakes have a dangerous feature: they're written with the same confidence as the right answer. The best-known assistants, such as ChatGPT, Claude, Gemini and Copilot, can invent an article of law, cite a book that doesn't exist, get a calculation wrong, mix up dates or state something outdated. Experts call this "hallucination." It isn't a deliberate lie: the tool generates the most likely text based on what it has learned, and the likely isn't always the true.
That doesn't make AI useless. It makes checking mandatory, the same way you count your change.
Why AI is not a chatbot
If AI were a chatbot with fixed answers, it would be enough to trust the script. Because it generates new answers every time, it's up to the user to judge what they receive.
In Bairo's view, AI is an extension of you, and that's why the responsibility remains yours. He says that "artificial intelligence is an extension of the human, not the problem." In the architecture he proposes for the ecosystem he founded, truth is the validator: nothing should count just because a machine said it.
Step by step today
- Know where it errs most. Exact numbers, dates, people's names, quotations, laws and court decisions, prices, recent news, local matters in your city. In those cases, be doubly suspicious.
- Ask for the source and open the source:
For each important claim in your answer, indicate where
it comes from. If you don't have a source I can open, say
clearly "I have no source for this."
- Ask it to point out its own uncertainty:
Reread your answer and mark what you're certain about, what
is likely and what may be wrong or outdated.
- Check in two independent sources. For health, use official agencies and medical societies. For laws, the official legislation portal. For money, the Central Bank and the CVM. For news, recognized media outlets.
- Ask in a different way. Ask the same question in other words, or in another assistant. If the answers diverge, it's a sign you need to check.
- Redo the math. Use a calculator or a spreadsheet. AI has improved a lot at math, but it still makes mistakes in long calculations.
- Be wary of the answer that pleases you too much. Tools tend to agree with whoever is asking. Ask for the opposite:
Now play the part of someone who disagrees with this conclusion. What are
the best arguments against it and what might I be
failing to see?
- Watch out for images, audio and video. AI also creates fake content that looks real. Before sharing something shocking, look for its origin.
- Teach this at home. Children, teenagers and older people deserve to know the machine can be wrong. Checking is a habit that protects the family.
A common example: someone asks what the deadline is to dispute a charge, and the AI answers with a number of days and an article of law. The answer looks perfect. On opening the official text, the person discovers the article deals with another subject, or that the deadline has changed. Five minutes of checking prevented a wrong decision. That's the habit worth building: use AI to get to the right question faster, and the official source to get to the right answer.
How Bairo did it
Bairo Leandro Gonzalez Martinez learned to build with artificial intelligence without knowing how to code. He says he spent six months at home, in front of the computer, getting things wrong, redoing and checking, until he had raised the foundation of the platform and the AI of CEASA Bank, a financial platform for agribusiness.
Perhaps that's why he gives verification so much weight. BTZ Chain, the network he proposes, is conceived as a validator of truthfulness and proof of humanity, so that humans and machines can work together in a place where the truth can be checked. For him, no one needs to trust a machine blindly. It's enough to be able to verify.
Precautions
Checking is also protecting yourself: don't share AI information without verifying it; don't make health, legal or money decisions based only on what it said; don't paste personal data to "test" the tool; and remember that, besides getting facts wrong, it reproduces biases from the texts it learned from, and can be unfair to women, Black people, immigrants, older people and people from underrepresented regions.