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 perceptionHow AI decides where to lookSpatial attention compared with self-attentionWhy vision transformers connect distant image regionsThe limitations of saliency maps and AI heatmapsHow AI retinal scans can support medical specialistsWhy machine vision fails when devices or environments changeHow 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.”