Why AI Makes Mistakes — and How to Avoid Them

By Oluwole Akinwale, Director, Professional Services

A few years ago, a good friend of mine set off on a trip, faithfully following his SatNav’s every instruction. After a series of turns and recalculations, the device cheerfully announced, “You have reached your destination.” The problem? He was staring at a vast expanse of water — as if the SatNav was saying, “Go on, plunge into the water!” His real destination was miles away. In the end, he abandoned the tech and found his way with the good old paper map.

AI is a lot like that when it’s wrong. It doesn’t hesitate. It doesn’t say, “Hmm, not sure about this.” It gives you an answer with total confidence, like it’s absolutely certain, even when it’s not.

That’s the tricky part: the tone, the wording, the polish … it all feels trustworthy. But just because something sounds right doesn’t mean it is right. Confidence isn’t the same as correctness — and with AI, we need to remember that.

So why does AI, with all its computing power, still get things wrong?

It usually comes down to one of three culprits. Think of them as the ‘usual suspects’ in any AI blunder.

1. Bias in the Data

AI is trained on large amounts of data, which helps it to detect patterns, make predictions, and deliver answers based on what it has “learned.” The quality, diversity and balance of that data shape the lens through which the AI sees the world. It’s like an old saying from my tribe — “Until a child visits his friend’s father’s farmland, he’ll always think his father has the biggest farmland in town.”

Similarly, if AI isn’t trained on diverse and balanced data, it ends up seeing the world in a limited or even distorted way. It might sound smart and confident, but it could miss the bigger picture, and that can lead to unfair or inaccurate answers.

Imagine an AI trained to screen job applicants mostly sees past hires who were men. There is a risk that it will unknowingly favor male candidates.

Got the idea? Let’s move to another point.

2. Bad or Incomplete Data

AI is only as smart as the data it’s been fed. If the data is outdated, incorrect, or incomplete, the results will also be inaccurate. It’s the classic ‘garbage in, garbage out’ problem.

A few months ago, I built an AI-powered global alert system that pushes updates to a Telegram channel. Initially, the model referred to Donald Trump as the “former U.S. President.” That would’ve been accurate a while ago, but not now. The model was either trained on data that was once accurate but has since become outdated over time, or the knowledge base used to enhance the model’s response was outdated. When I switched to another model trained on updated data or with an updated knowledge base, the issue was resolved.

The same thing happens in other areas. A travel chatbot might suggest a great little café — only for you to find out it closed down three years ago. Again, it’s not the AI’s fault. It’s just working with old info.

3. Misinterpretation (or “Hallucination”)

When AI doesn’t have the correct information, it doesn’t say “I don’t know” — it just makes something up and says it confidently.

I was once preparing for a speech and asked ChatGPT for some inspiring, well-known quotes on the topic. It gave me a handful… and they sounded great. The only problem? A couple of them were entirely made up.

Luckily, I’ve learned to always double-check. But imagine if I hadn’t — I would’ve stood on stage quoting someone who never said those words. That’s the kind of slip-up AI won’t warn you about.

In 2023, a lawyer used ChatGPT for legal research. It invented fake case citations, and I imagine they were confidently and sufficiently convincing for the lawyer to include them in court. I am sure you already know how it ended.

Why It Matters That People Understand AI Makes Mistakes:

AI isn’t magic, it’s math. When we don’t understand how and why AI gets things wrong, we give it more authority than it deserves and stop questioning it. That’s risky.

Imagine how people can leverage AI in research and walk away with wrong facts.

Think about the impact when a job applicant gets rejected by an algorithm because the training data was biased.

Imagine the consequences when patients ask a chatbot for medical insights and walk away with fake advice.

AI has no idea if it’s telling you the truth. When we trust it blindly, it starts to quietly shape our choices, beliefs, and even our rights. But when you do understand why it gets things wrong, you stay in control — not the algorithm.

How to Protect Yourself:

• Ask for sources:
  When you leverage AI to perform certain tasks, especially research-based ones, ask for sources.
• Cross-verify result:
  Compare with a trusted, human-verified source or ask for a second eye review and validation.
• Beware of overconfidence:
  Polished doesn’t equate to correct.
• Use AI for drafting, not decision-making:
  Let it suggest; you confirm.

What, then, should be our mindset as we increasingly adopt AI? Well, it’s simple: “AI isn’t here to replace human intelligence; it’s here to complement it.” The responsibility for oversight and verification still rests with us, and I hope it remains that way.