It’s just the way you do metrics for AI. You assign probabilities to the successful detection for a single doctor based on real world data e.g. one doctor will identify breast cancer correctly 90% of the time. You then assign some probability for when two doctors would consult together and successfully identify the breast cancer e.g. 95%. Then a group of doctors etc. etc.

If your AI is 96% effective then you would say it is better than two doctors.

Yes the title is a little strangely worded but in the context of the above it makes sense.