1. AI at least has value for business given a way to educate users
2. Before ML algorithms are better understood mathematically, industrial knowledges will still be the key
3. Dataset population's changeability and identifiability determine ML applications among industries
While AI explores a way for the predictive relationship, C3.ai
benefits only from its efforts to help business save labor costs related to algorithms and data.
Lacking mathematic explanation determines that, beyond uselessly trimming algorithms, we don't have direct ways to solve critical AI issues such as overfitting. Multi-layer perceptron can be explained by Universal Approximation. But CNN can still not be explained as either a sub-set of Universal Approximation or by other theories. On the other hand, autonomous vehicle has no critical issues (exclude efficiency) because it is backed by the complete Markov chain theories.
Therefore, the best feature engineering, before we have mathematic breakthrough, is to utilize manual industrial knowledge to reduce dimensions for datasets to solve overfitting issue.
RFQ Relativity of Microeconomy always pursues the best solution for AI application to economy and finance research by our manualized reduction of feature dimensions.