Almedalen Seminar: "How does AI actually works?
On 27 June, the Department of Mathematics organised a talk during Almedalen Week on how AI works.
AI is discussed everywhere in society these days, but it is rarely explained how it works. There are even a few common misconceptions about it. The main ideas behind AI are not so difficult to understand. Deeper knowledge should be useful for decision makers and educators in particular.
Today's Artificial Intelligence (AI) is less of a thinking brain than a gigantic statistics machine based on what is called deep learning. It is not often well explained in any detail how AI works which is a missed opportunity since the ideas behind it are rather simple. Moreover, there are several misconceptions partly due to the usage of the words 'intelligence', 'deep' and 'neural networks'. In this lecture we will begin by defining the terms used and discuss the paradigm of machine learning for today's AI, listing the three main types of machine learning. Then we will explain how deep learning and generative AI work, which is on the level of high school mathematics. Finally it will be emphasized that nobody, not even the developers and researchers themselves, understands really *why* it all works so miraculously well. This uncomfortable circumstance is much related to the problems and risks of AI.
The lecture was given by Anders Karlsson.