I recently teased that we might want to take a look at other forms of AI as well. Because it's important to understand that AI is much broader than ChatGPT or DeepSeek, to name just two examples.
That kind of AI tries to hold conversations. It was designed to blow people away with a wow moment from the comfort of their homes. But the real wow moment of AI often lies miles beyond that.
And the funny thing is this: the most oversized form of AI is actually the LLM, such as ChatGPT, Claude, and Mistral. Other forms of AI can be smaller and more focused. Anyway, an LLM is a type of AI that can sometimes have unpredictable consequences. For example, the U.S. state of Florida recently filed a lawsuit against OpenAI over concerns about how ChatGPT behaves in certain situations. Think of strange or potentially harmful advice given in response to unusual questions. Those are exactly the kinds of risks critics of these systems point to.
For a similar reason, OpenAI also ended up in conflict with Elon Musk. Among other things, his criticism focuses on the fact that OpenAI has become increasingly commercial, while it originally also had a scientific and open mission.
We're not going to zoom in on all that news. My point is much simpler: AI is far more than that. So before we dismiss AI as a concept altogether, it might be worth taking a brief look at where AI is being used responsibly.
AI as a Tool
Take science, for example. There are countless models developed by researchers who use AI simply to speed up processes. Think of analyzing photographs of clay tablets containing texts or symbols that we don't yet fully understand, or sometimes don't understand at all. Those are legitimate applications of AI.
The same applies to AI systems that analyze medical images. For example, MRI scans, where AI can help doctors detect abnormalities more quickly. There too, it serves as a tool that adds value.
What Is XAI?
And that brings us to XAI. That stands for "Explainable Artificial Intelligence." And what exactly is that? I'm glad you asked. Otherwise this blog would have had to end right here.
Anyway, since you brought it up: XAI is AI without an opaque "black box." While some companies focus mainly on impressing people with spectacular results, XAI is all about transparency and accountability. The AI helps perform a task faster, but every step and every decision can be examined and understood.
If a scientist asks an AI to help decipher a three-thousand-year-old clay tablet, that scientist probably isn't looking for an authoritative answer with no supporting evidence. They want to know why the AI reached a particular conclusion.
And that's really how AI should be used. With a traceable and verifiable chain of reasoning that can be shared with other scientists who will evaluate the research.
That last part is essential in science. It's called peer review, and it's basically what happens when another scientist says, "Wow, that's interesting. Let me check that." And then they actually do check it. That is only possible when all the information needed to reproduce the work is available. That's how it's supposed to work. That's how science works.
The computer as a well-behaved dog
This is why it's perfectly fine to use AI. At the end of the day, it's still a computer. It speeds things up. The computer doesn't sleep. The computer can process enormous amounts of information in a very short time.
The only thing we as humans should insist on is that the computer explains how it arrived at its conclusion. That way, everyone can verify whether that conclusion is actually correct.
And to paraphrase myself from the first AI article: that's how you turn an old rust bucket into a well-behaved dog.
AI is not the problem. Charlatans trying to make a quick profit with the help of AI are the ones we need to keep somewhat in check. That's all.
There is still room for improvement
Of course, this does not mean that AI is a finished technology. Quite the opposite. We need to continue exploring how AI can become more efficient and more mature. Even ethically responsible AI currently has a significant footprint. Training and running large models simply requires a considerable amount of energy.
This is not unique to AI, by the way. The first computers occupied entire buildings. Today, we carry smartphones in our pockets that are many times more powerful than those systems ever were. As technology evolves, it generally becomes smaller, faster, and more efficient.
AI is not going away. That is why it is important to make it energy-efficient as soon as we can. Just as we have done with virtually every other technology that we now take for granted.