– AI (Minus AI)

A few years ago, a man went on a trip, faithfully following the navigation system’s instructions. After a few curves and recalculations, the device happily announced: “You have reached your destination.” The problem? He stared at a huge expanse of water – as if the navigation system wanted to say: “Come on, dive in!”

The actual destination was miles away. In the end, he dispensed with technology and looked for the way with the good old paper map.

AI behaves similarly when it makes a mistake. IT does not hesitate. IT doesn’t say, “Hmm, I’m not sure about that.” IT gives you an answer with absolute certainty, as if IT were absolutely sure, even if that’s not true.

That’s the hard part, the tone, the formulation, the polish, everything seems trustworthy, but just because something sounds right doesn’t mean it’s right. Trust is not the same as rightness—and with AI, we have to keep that in mind.

So why does AI still make mistakes despite its computing power?

Usually, it boils down to one in three culprits. Think of them as the “usual suspects” in any AI failure.

1. Data bias

AI is trained on large amounts of data. This helps to identify patterns, make predictions, and provide answers based on what it has learned. The quality, variety, and balance of this data shape the lens through which AI sees the world. It’s like an old saying from my family: “Until a child visits the farmland of his friend’s father, he will always think his father has the largest farmland in the city.”

The situation is similar with AI if it is not trained with diverse and balanced data. IT sees the world only in a limited or even distorted way. IT may sound intelligent and confident, but IT might miss the bigger picture, which can lead to unfair or inaccurate answers.

Imagine that an AI trained to screen applicants sees predominantly male applicants in the past. There is a danger that IT will unconsciously prefer male candidates. (Oh, goodness, that’s like in real life, my bad.)

2. Poor or incomplete data

AI is only as intelligent as the data it is fed with. If the data is outdated, incorrect or incomplete, the results are also inaccurate. It’s the classic “garbage in, garbage out” problem.

A few months ago, someone developed an AI-powered global alert system that sends updates to a Telegram channel. Originally, the model referred to Donald Trump as a “former US president”. That would have been true some time ago, but not today. The model was either trained on data that was once correct but is now outdated, or the knowledge base used to improve the model response was outdated. When I switched to another model trained with updated data or an updated knowledge base, the problem was resolved.

The same is happening in other areas. A travel chatbot might suggest a great little café to you—only to find out that it closed three years ago. Again, the AI is not to blame. It simply works with old information.

3. Misinterpretation (or “hallucination”)

If the AI doesn’t have the right information, it doesn’t say “I don’t know” – it just makes something up and says it confidently. (A lot like social media?)

If you are preparing for a speech and ask ChatGPT for some inspiring, well-known quotes on the subject, you most likely will get a handful… and they sound great. The only problem? Some of them will be fictitious.

Fortunately, I have learned to always double-check everything. But imagine if someone stood on stage and quoted someone who never said those words. The AI does not warn you of such slip-ups. The same goes for the internet in all its glory. Especially, quotes are so often mistakenly (or purposely) credited to the wrong person.

In 2023, a lawyer used ChatGPT for legal research. The app invented fake case quotes, and I imagine they were convincing enough to convince the lawyer in court. I’m sure you already know how the whole thing ended.

I used AI to modify pictures I had taken, mostly of furniture projects I had completed for customers and felt proud of. I have a small display area in my workroom, even have a backdrop to take the perfect picture, but it’s all too time-consuming to set it all up. When I took a photo of a chair I had restored and hoped to just ‘clean up’ the messy background, AI did just that. Voila, the result was the perfect photography for my portfolio. It was my work, my workroom, but a different background.

When I wanted to share a picture of myself, but didn’t want to show the certificate with my full name in the background, I also altered the wall. In both instances, I asked a software program if AI had been used, and both times I got the answer that this was an AI picture. See, that’s where it’s wrong. It doesn’t know what has been altered, so it assumes it’s all artificial and not real. According to AI, I am not real, nor is the chair.

Then I got more curious and gave my book manuscript to AI. I wanted to know the verdict. Goodness, it answered all my questions. AI gave me two pages of answers about my writing, the tone, the people, the feel, and the emotions. It was an interesting read. Funny acording to AI, there were no mistakes in the manuscript, but that’s where it’s wrong. Two punctuation errors have been overlooked and can be found, guess where? In my book! 🙂 AI could not predict the outcome, nor could it explain a few paragraphs, when asked about the relations.

When you type a few pages of famous literature classics, for example, “The Tale of Two Cities,” especially the beginning of the book, which includes the famous line “It was the best of times, it was the worst of times,” you will learn that according to artitificial intelligence, the whole writing of the first paragraphs is not correct. There are simply too many mistakes in it—grammar, punctuation, and the length of the sentences. AI would butcher this famous book. AI can’t feel. AI can’t tell you how much is ‘altered’ or what part, so it just assumes it’s all wrong, or all … right.

Why do people need to understand that AI makes mistakes?

AI is not magic, but mathematics. If we don’t understand how and why AI makes mistakes, we give it more authority than it deserves and stop questioning it. This is risky. The way they shovel AI down our throat is dangerous. It’s downright frightening how many people DO NOT question AI.

Imagine how people use AI in research and are left with false facts. Think about the impact if an applicant is rejected by an algorithm because the training data was wrong or incomplete.

Imagine the consequences if patients ask a chatbot for medical insights and get wrong advice.

AI doesn’t know if it’s telling the truth. When we trust it blindly, it tacitly influences our decisions, beliefs, and even our rights. But if we understand why it makes mistakes, we remain in control—not the algorithm.

How to protect yourself:

• Ask for sources
  If you’re using AI to perform specific tasks, especially research-based tasks, ask for sources.
• Cross-check result 
  Compare with a trusted, human-verified source, or ask for a second-person review and confirmation.
• Beware of overestimating yourself
  Polished is not the same as correct.
• Use AI to design, not to make decisions
  Let it suggest, you confirm. You are the boss!

So, what should we think when we increasingly use AI? Well, it’s simple: “AI isn’t there to replace human intelligence; it is there to complement it.” The responsibility for oversight and review remains with us, and I hope that it will remain so.

AI will put websites like Wikipedia out of business if we let it happen. Why would I go to the Mayo Clinic or other trusted sources if AI gives me the same answers and I don’t have to scroll down? Well, that’s for you to decide. I know what I need to do. I keep on scrolling.

To be honest, I search online and add -AI (Minus AI) behind all my search phrases, because I still want to see the original sources, and it works like a charm.

For the moment, I am cured of AI but will continue using it to change a background in a picture, and get logical, analytical answers to questions.

AI will not take my job, nor can it compete with my crazy design choices. It’s a tool! It can help me find what I want, but it can’t see my imagination because sometimes even I don’t see the result before it’s actually finished. Do I make sense?

Yes! But don’t ask AI.