AI != LLMs.

The other day, I attended a lecture at the DevNetNoord mini-conference. The speaker was telling us about how they use AI in their products. Nothing spectacular so far, you’d say.  The funny thing? He more or less apologised for “not talking about LLMs” as if AI begins and ends with a chatbot. And yet the guy sitting next to me whispered, with the awe of someone watching HAL 9000 boot up live: “Wow… this is real AI.”

These two reactions perfectly capture the confusion people still have about AI. That led to this article.

People think of AI as LLMs and Chatbots. But that’s not reality. LLMs are one of the most exciting things ever to come out of the labs of the AI specialist, and indeed, the one that gets the most attention from the general public. In contrast, most of the interesting things happening in AI, such as cancer detection from MRI scans, self-driving cars, fraud detection, and so on, have nothing to do with LLMs.

However, AI is not new. The term AI was coined in 1956 at the Dartmouth College conference. There have been numerous attempts to make machines smarter ever since (I do have a rather nice keynote about the history of AI, if you’re interested, reach out to me!).

For instance, the guy on stage talked about reinforcement learning. I found that funny, since not a lot of people talk about that, yet here this guy was covering a topic I personally presented at the GlobalAI meetup two weeks earlier. And I also had that same response from the audience: “This is cool, we need more of these kinds of lectures!” Apparently, people think this is ‘real AI’, whatever that might mean.

But I get it. When building a Reinforcement Learning (or RL in short, we love abbreviations, don’t we?), you need to know about neural networks. Ok, to be honest: you can do without them if you are using a simpler form, known as Q-Tables, but they are pretty limited. As the speaker showed, you need some initial data that fits a particular form and is readily available. And you might want to use a digital twin of the real world to help get training data.

You need to build a network with the right number of input neurons, the correct number of hidden layers, and the appropriate number of output neurons. In other words, you need to be aware of how AI works. I guess that’s why the guy next to me said, “This is real AI”.

And of course, the fact that the speaker apologised for not talking about LLMs shows how most people perceive AI.

My suggestion to you, if you are interested in this sort of thing: look around at other uses of AI. Don’t be blindsided by all the chatbots and copilots out there. They are amazing and can deliver significant value, but AI is much broader than that. If you limit AI to chatbots, you’re missing 90% of the possibilities. So, go out and explore what you can do with AI!