How AI Decides What to See and What to Ignore

👁️ How does artificial intelligence decide what to see?

Your eyes can look directly at something without your brain ever noticing it. AI faces a similar problem. A camera may capture every pixel, but the system must still decide which parts of an image matter and which parts it can safely ignore.

In this episode of A Beginner’s Guide to AI, we examine spatial attention in humans and visual attention in artificial intelligence. You will learn how the brain uses a mental spotlight, why seeing is not the same as noticing, and how attention mechanisms help computer vision systems process complex images.

We also investigate the limitations of AI attention. A model can identify the correct object for the wrong reason, use backgrounds as shortcuts, or create a convincing heatmap without truly understanding the scene.

🏥 Our central case study follows the collaboration between Google DeepMind and Moorfields Eye Hospital. Their medical AI system analysed three-dimensional OCT retinal scans, created detailed tissue maps, and recommended how urgently patients should be referred. It performed at a level comparable with leading specialists in a retrospective test. Then a different scanner caused its accuracy to fall dramatically.

The anatomy had not changed. The machine’s view of it had.

🔍 Key highlights:

• How spatial attention filters human perception
• How AI decides where to look
• Spatial attention compared with self-attention
• Why vision transformers connect distant image regions
• The limitations of saliency maps and AI heatmaps
• How AI retinal scans can support medical specialists
• Why machine vision fails when devices or environments change
• How humans and AI can compensate for each other’s blind spots

📧💌📧

Tune in to get my thoughts and all episodes, and don’t forget to subscribe to our newsletter: beginnersguideto.ai

📧💌📧

Quotes from the Episode

• “Spatial attention begins with a simple problem: there is too much world and not enough brain.”
• “The anatomy had not changed. The machine’s view of it had.”
• “Every spotlight reveals something. Every spotlight also leaves something in the dark.”

About Dietmar Fischer

Dietmar is a podcaster and digital marketer from Argo.berlin. If you want to get your AI or digital marketing moving, contact him at argoberlin.com

Hosted on Acast. See acast.com/privacy for more information.