Four good reasons to not blindly trust AI

Why should we think twice before using AI as a therapist, romantic partner, or to help with homework? AI expert Elisabeth Wetzer explains.

By Petter Bjørklund, Communications Officer at SFI Visual Intelligence

We are using artificial intelligence (AI) more than ever. Everything from doing schoolwork to having a chat about things we are curious about.

In fact, Norway uses the most AI in all of Europe. A whopping 56% of Norwegians use AI for different things.

“Smart to think twice”

Even though AI can help us with many different things, we shouldn’t always trust what it tells us.

This says Elisabeth Wetzer, who is an associate professor at Visual Intelligence.

Wetzer researches AI at Visual Intelligence, a research centre at UiT.

She believes we should think carefully before using AI for different purposes

“There are several reasons why,” says Wetzer.

1. AI can make up things that are not true

AI is designed to answer confidently, even when what it says isn’t always correct or true.

AI can make all sorts of strange claims. For instance, ChatGPT and Grok have fabricated sources that don’t exist and events that never happened

“Many people assume that an answer must be correct simply because it comes from a machine. That isn’t always the case,” says Wetzer.

She encourages us to always double-check the information AI gives us. This could mean checking in a book or other trustworthy sources online.

Wetzer specifically warns against taking health advice from AI. Research shows that AI is not something you should rely on when it comes to medical advice.

“ChatGPT is not a doctor. At the worst, this advice could be directly harmful to your health,” she says.

2. AI learns things about you – without you even knowing it

AI learns by looking at many examples of the same thing. An AI that has learned to recognize cats, for example, is trained on thousands of pictures of cats.

But AI also learns when we talk to it and share things with it. It can therefore learn very personal things about us, for example if we use it as psychologist or a romantic partner.

What we tell AI doesn’t simply disappear. Open AI, the developer of ChatGPT, uses the chatlogs to further develop the chatbot.
“This data is extremely valuable to AI developers. You never know what the data might be used for,” says Wetzer.

3. AI can discriminate against people

AI imitates humans remarkably well, Wetzer says.

So well that it has become good at learning how we behave against one another – both for better and for worse.

Many people experience hate and harassment based on their skin color, sexual orientation or gender. Just like humans, AI can also learn to become racist, homophobic or to favor men over women.

For instance, ChatGPT has spoken disparagingly about people who speak African American English.

This happens because AI is often trained on text, images and videos from the internet and social media, such as Reddit, YouTube or 4chan. This means it can inherit prejudiced attitudes towards different groups of people.

4. AI rarely disagrees with us

Have you noticed that AI chatbots often agree with you? Maybe sometimes a bit too much?

AI is designed to validate what we think and believe about things.

“It will always try to see things from your point of view – no matter what,” says Wetzer.

This can happen if you tell AI about an argument with a friend.

“AI will often tell you that it’s completely understandable that you were furious with them,” she says.

This can affect how we behave towards one another.

“In the worst case, you may become less willing to listen to people who have different opinions than you,” explains Wetzer.