Despite the name, artificial intelligence (AI) is not intelligent at all - but it can behave in a way that appears intelligent. But how does AI work? How much are today's methods able to understand about us and our world, and why is it so difficult for us humans to understand what the systems are learning?
In this talk, we will delve into the world of AI - or, to be more precise, machine learning - which is software that can progressively improve its performance on a specific task as it gains experience. The most advanced machine learning software in use today builds on so-called neural networks. Inspired by how our brain works, these networks are formed of layers of "neurons", with each layer able to learn an increasingly complex representation of the world - a method commonly referred to as "deep learning". In the talk, we will see just how complex information these neural networks are able to learn today, and we will discuss how they are constructed and gain some intuition about how they work. Importantly, we will see how it can be that despite the neural networks being built by us, we have very little understanding of what they actually learn and how to train them. Lastly, we will discuss some of the limitations of neural networks, and machine learning in general, illustrated by some problematic real-life use cases.
About the speaker
Kristoffer Stensbo-Smidt is a postdoc at University of Cambridge, where he works on probabilistic machine learning methods, which are methods able to learn complex representations of the world from very little data. He holds a BSc and MSc in astrophysics from the Niels Bohr Institute and a PhD in machine learning from the Department of Computer Science, University of Copenhagen. Besides his research, he has a keen interest in communicating the pros and cons of machine learning and is particularly interested in the ethical aspects of the technology. He has given numerous talks on the subject and is a columnist at Altinget: digital.
In case of questions feel free to contact student representative Jakob Forsberg Larsen, firstname.lastname@example.org.
* ANYONE CAN JOIN - Most of IDA's events are open to everyone, but you must have an IDA user profile in order to participate. It's free and non-binding - create a user profile now. As a member of IDA, you get a discount on many of our events, so please sign in with your member login when you sign up for an event. Read more about the benefits of an IDA membership here.