Intelligence Cost Index
A proposed basket of professional tasks for tracking the total cost of acceptable machine-assisted work over time.
The Intelligence Cost Index would track the price of completing a standardized basket of cognitive tasks to an acceptable quality threshold.
The goal is not to measure model intelligence in the abstract. It is to measure the cost of getting useful work done.
Initial basket
Candidate tasks:
- produce a first-pass company credit memo;
- review a commercial contract for material risks;
- reconcile two structured datasets;
- classify customer-service cases;
- create a market map from source documents;
- generate and test a small software change.
For each task, the index should capture inference cost, tool cost, human setup, human review, failures, rework, latency, and quality.
Output
The public version could show a simple time series. The paid or private version could contain model comparisons, task-level cuts, methodology changes, and downloadable observations.