Generative AI
Interpretable AI
Black box
Generative AI is like an eyeless alien that read all the internet text and decided that all visible colors like brown and all invisible light like UV and infrared can be represented by 2 dimensions, say frequency and temperature. But we don't know if these are the 2 dimensions it uses or something else. Nobody knows how the words we speak to it are represented internally in AI.
Transparent
Interpretable AI is like a human with eyes that sees a visible color like brown and interprets it as red, green and blue via three cone types before sending it to the brain for processing. So we represent brown as 3 dimensions, i.e. RGB, which is understandable to humans, and we know exactly how the words we speak to it and the output it produces are represented internally.
Unpredictable
When you give part of a sentence to generative AI, it predicts the most probable word that comes next. If it generates one word the first time and another word the second time, that makes it feel more like a human than a cookie cutter bot that repeats the same word every time. But using the same next word prediction for everything from financial calculation to code may not be reproducible and break regulations.
Predictable
Interpretable AI constructs SQL deterministically from the named dimensions that are human understandable. This SQL does the financial calculations required to answer your prompt. That makes the answer repeatable, and auditable on how it arrived at the answer.
