Understanding Artificial Intelligence – what ChatGPT & Co. can (currently) really do

Patrick Straub, as appeared in Alpha1 Journal 2/2025.

Knowledge has never been so accessible. One click, one sentence – and an artificial intelligence formulates the perfect answer. But when it comes to one's own health, uncertainty, hope, or fear, things become more complicated.

People with rare diseases like alpha-1 antitrypsin deficiency know how arduous the search for reliable information can be. A machine that answers every question instantly seems appealing. But is it worth trusting?

What exactly is a "language model"?

When people talk about "artificial intelligence" today, they usually mean a specific type: so-called Large Language Models (LLMs). This sounds complicated, but it's essentially a computer program that has been trained on a vast amount of text. You can think of it as a huge library in which the program has learned how language works. It knows which words frequently occur together, how sentences are structured, and how to formulate a coherent response.

ChatGPT, Gemini, and other systems use this knowledge to write texts or answer questions. However, they don't think like humans – they don't truly "understand" the content. Instead, they calculate which word is most likely to follow next.

So, if you ask, "What is pulmonary fibrosis?", the program searches its internal "language world" for patterns and formulations that fit well. This creates the impression that it really knows what it's about – even though it's actually just imitating language.

How are the answers generated?

To enable such a system to respond, it was trained on billions of words from books, newspapers, and websites. This is how it learned how people write about things.

Systems such as ChatGPT or Gemini can now also access current information – for example, via an internet connection. This is called grounding. The program then specifically searches for reliable sources to verify or supplement its answer.
Nevertheless, caution is advised: Not every source on the internet is reliable, and even modern AI models can be wrong. They combine language patterns with found data – they don't "know" what is true, but only try to formulate the most appropriate answers.

Why many people find AI exciting

Those living with a rare disease know the feeling of finding hardly any helpful answers on search engines. Many affected individuals report that AI chatbots help them understand technical terms or decipher medical reports. Some use them to summarize longer texts or to formulate questions they want to ask at their next doctor's appointment. AI can also provide emotional relief: it responds instantly, it doesn't judge, and it's always available. This also demonstrates the strength of this technology: it makes knowledge more accessible. Those who have difficulty reading, seeing, or typing can communicate with it via voice input. Some even use it to translate foreign language texts or to put complicated scientific passages into simpler terms.

Where there is light, is there also shadow?

As impressive as the possibilities are, there are clear limitations. One of the biggest problems is so-called "hallucinations." This means that the AI invents things when it can't find a suitable answer. It sounds very convincing, even though what it's saying is simply wrong.

For example, if you ask about a rare side effect of a medication, the model might give a plausible but fabricated answer. This is because it doesn't "know" what is true – it generates texts that appear linguistically credible.

Data privacy is also important. Some AI systems store user input to improve their services. Depending on the provider, this can sometimes be deactivated. Therefore, anyone entering their medical history, lab results, or personal data risks this information ending up on third-party servers.

And one more thing: AI is not objective. It learns from human texts – and humans make mistakes or have biases. So if a language model has been trained on such texts, it can adopt these distortions.

How to use AI safely

You don't have to completely abandon ChatGPT and similar technologies. The key is to use them consciously and intelligently.

Here are a few simple rules:

1. Do not enter any personal data.
Do not upload names, diagnoses, or documents.

2. Always check health information.
Official sources are better: e.g., university hospitals, patient organizations, or the Robert Koch Institute.

3. AI as an aid, not as a doctor.
You can have things explained to you, but you should never base a diagnosis or treatment decision solely on that.

4. Keep your common sense.
If something sounds strange – it's better to read it again or ask someone who knows about it.

5. Ask where the information comes from.
Many AIs can now specify which sources they rely on. So, if you want to know how an answer is generated, feel free to ask: "What source are you using for this?" – this increases transparency and traceability.

Those who adhere to these principles can derive great benefit from technology without endangering themselves.

A future with a sense of proportion

Artificial intelligence will change medicine – that much is certain. But how much we trust it depends on how responsibly we use it. AI can help us be more informed, better prepared for doctor's appointments, and understand connections. But it can also mislead us if we use it uncritically.

That's why education is so important. Those who understand how AI works can interpret its responses. It is neither omniscient nor inherently dangerous – but rather a tool. And as with any tool, it's how you use it that matters.

With a clear head and an open mind

ChatGPT, Gemini, and similar programs can be valuable companions in everyday life – especially for people with rare diseases. They translate, explain, and help with understanding. But they do not replace medical expertise.

The best approach is to remain curious but critical. This way, the new technology can be used safely and effectively – and perhaps in the end, it will even help bring a bit more clarity to one's own life with a rare disease.