ML = AI?

At Machine Learning Week Europe, I surprisingly had several conversations about whether “machine learning” was still the right name for the conference, because of the many sessions about “AI”. One participant even raised the objection that AI was not ML at all. 🤯

His argument: ML is about numbers, but AI is about language. The discussion was quickly over. First, I had to leave to moderate the next AI lecture, sorry, ML lecture. Second, I had a convincing counterargument: Large Language Models do not process language, but embeddings translate text into vectors and train a predictive model on those numbers using deep learning, which is a subtype of ML.

Nevertheless, in many presentations and discussions, there was a noticeable desire to distinguish AI from ML. The term “classic ML” was very popular, i.e., the “old school algorithms” vs. the “new kids on the block”. The “old hands” labeled their tried-and-tested tools “analytical AI” to make it clear that there are other forms of AI besides “generative AI.” However, generative AI is analytical AI: transformers analyze (training) data.

My opinion on this: “AI” has always been and still is a marketing term that says everything and nothing. Above all, it describes what we hope for. “ML” describes what we aim for. As of today, all AI uses ML. AI is data-driven and analytics-powered. That’s it. Or, as mathematicians call it: “applied statistics.”

When asked whether AI = ML or ML = AI, moderator Sebastian Wernicke had the pragmatic answer in the closing discussion: “Who cares?”