FF/Etcetera

Brain Teaser

The Trust Calibration Challenge

Jan 02, 2026

An AI system’s confidence correlates with accuracy as follows:

  • 90-100% confidence: 95% accurate
  • 80-89% confidence: 85% accurate
  • 70-79% confidence: 75% accurate
  • Below 70% confidence: 60% accurate

If the system processes 1000 decisions with the confidence distribution: 200 high (90-100%), 300 medium-high (80-89%), 300 medium (70-79%), and 200 low (below 70%), how many decisions should you expect to be correct?

Answer:

Total decisions: 1000

Calculate the number of correct decisions for each confidence level:

  1. High confidence: Number of decisions: 200 Accuracy: 95% Expected correct = 200 * 0.95 = 190
  2. Medium-high confidence: Number of decisions: 300 Accuracy: 85% Expected correct = 300 * 0.85 = 255
  3. Medium confidence: Number of decisions: 300 Accuracy: 75% Expected correct = 300 * 0.75 = 225
  4. Low confidence: Number of decisions: 200 Accuracy: 60% Expected correct = 200 * 0.60 = 120

Now, sum up the expected correct decisions from all categories: Total correct = 190 + 255 + 225 + 120

Let’s add them: 190 + 255 = 445 445 + 225 = 670 670 + 120 = 790

Therefore, you should expect 790 decisions to be correct.

Final Answer: 790