Generative AI
Interpretable AI
Black box
Generative AI is like an alien that read all the internet text and chose its own internal representation to "see" color. These representations are called dimensions and we don't know what they mean. The frequency of light could be its dimension or something we don't understand. Neither do we know how they interpret the words we speak to it.
Understandable
Humans see brown as red, green and blue through three cone types in our eyes. Prereasoner is trained the same way. Its dimensions are named after schema.org properties such as country and currency. So, the question and the answer can be traced back to those properties in our model.
Unpredictable
Generative AI predicts the most probable word that comes next. It can pick one word the first time and another word the second time. That makes them human-like in chatbots. But in a financial calculation, it means the same question can return a different number.
Predictable
Prereasoner converts a question into named dimensions such as "SUM" of "SALES" in "FRANCE". This is converted into an SQL query or Python code, deterministically. When it runs, the database or code computes the number. With the same data, you get the same query and the same answer.
