Show Notes
About This Episode
Computer vision is usually discussed through two lenses: the spectacular, like self-driving cars, and the alarming, like surveillance. Neither quite captures where the technology actually is.
In this episode, Mikkel Svold sits down with Andreas Møgelmose, Associate Professor of AI at Aalborg University's Visual Analysis & Perception Lab, to map out where computer vision is genuinely working right now. In Danish hospitals, an AI system is reading x-rays and cutting average emergency room wait times by an hour. On factory floors, cameras guide robotic arms through welding, painting, and packing. The most common application of computer vision, Andreas argues, is not face recognition or surveillance. It's industrial production.
The episode also clears up a few persistent myths. The trolley problem, the idea that self-driving cars must consciously decide between harming one person or another, turns out to have almost no relationship with how these systems actually work. The privacy question is more substantive, but the nature of the problem is different from what most people assume.
The most striking moment comes near the end: Andreas explains that telling an AI system not to recognise you doesn't erase your data from it. It adds more. The right to be forgotten, enshrined in European law, runs into a technical wall that the field hasn't yet found a way around.
In This Episode
- The use cases most people don't think about: industrial vision, x-ray diagnosis, and robotic guidance
- Why the most common application of computer vision is in factories, not surveillance
- How a Danish hospital cut emergency room wait times by an hour using AI
- The trolley problem myth: why self-driving cars don't decide who to harm
- What cameras can and can't do for humanoid robots
- Why AI has enabled surveillance at a scale that was previously impossible, even with existing cameras
- Machine unlearning: why asking a system to forget your face actually adds more data about you
- The difference between broad category recognition and the fine-grained decisions AI still struggles with
Chapters
- 00:26 Introduction and what computer vision actually is
- 01:11 Why vision may be the most important sense for a computer
- 02:04 Can computers rely on vision alone?
- 03:39 Use cases in healthcare, industry, and robotics
- 07:20 Debunking the trolley problem in self-driving AI
- 13:36 Stereo vision and 3D understanding in machines
- 18:03 Humanoid robots and what vision makes possible
- 21:08 Privacy, surveillance, and what AI actually changes
- 27:37 Fine-grained recognition and open problems
- 29:32 Machine unlearning and the right to be forgotten
Key Quotes
"Not many AI scientists want to replace humans. What we really want to do is augment whatever the humans are doing already."
"AI can enable surveillance and mass surveillance on a scale that has never before been seen."
"You can put a layer on top of the system saying, the guy that walks like this, you're not allowed to recognise. But that doesn't erase your information from the system. That actually adds more information about you to the system."
About Andreas Møgelmose
Andreas Møgelmose is Associate Professor of AI at Aalborg University, where he works in the Visual Analysis & Perception Lab. His research spans applied computer vision, from pre-operative screening automation for anesthesiologists to colour segmentation in video. He brings both theoretical grounding and a practical orientation to questions about where computer vision is genuinely ready for deployment, and where it still falls short.
Contact & Follow
Questions, topic ideas, or guest suggestions: podcast@bigideasonly.com
Find more episodes at [podcast website]