Bohang Sun at the University of Cambridge and Professor Sushma Grellscheid at the University of Bergen, and I just finished an academic paper on the emerging landscape of AIQ frameworks. You can find it at myaiq.org/paper.html while we find the right scientific journal to publish it.
In our research we found ten independently developed AIQ frameworks, spread across cognitive psychology, computer science, education, enterprise consulting, and marketing, with almost no cross-referencing between them. We decided to analyze, review, and compare them systematically.
We discovered two key axes on which to compare them.
The first is whether a framework treats AIQ as a stable trait, something you essentially have similar to IQ, or as a perishable skill that decays over time and requires active maintenance as the AI frontier shifts.
The second is the object of measurement. Most frameworks center on knowledge or performance, while leaving calibration implicit. Calibration is the ability to accurately read when AI can and cannot be trusted. For example you can perform well on routine AI-assisted tasks while remaining miscalibrated at the edges where AI fails in unexpected ways.
We did our best to place all ten frameworks along these two axes, and admit this was partially subjective. Clearly much more research is needed, not just on the frameworks called AIQ, but on the broader landscape of AI literacy and proficiency frameworks that use different names.
A more personal take
The paper is scientific and we tried to be as neutral as possible. But on this blog I can be more biased: I personally believe that a framework centered on human skill and calibration ability will be critically needed as AI becomes more deeply embedded in society. A credential that does not expire has limited credibility. And performance without calibration is the same as fluency without judgment.
Personally I plan to further work on the AIQ framework in the top right of the matrix. This is why I set up myaiq.org. The site is not officially launched yet, but so far hosts the framework, the paper, and a simple multiple choice question test on AIQ. I know full well this is not the real test. I am busy building a true agent that can measure your AIQ in an interactive question and answer format based on tasks. If you want to help test it, drop me a line or DM me and I will send you a private link.
What do you think? Do you agree that calibration is the missing piece? Do you have another framework and where would it fit? Join the conversation in the comments.