A few years ago, then HSBC US CRO Rhydian Cox had a conversation with an AI expert. After the long talk on AI given by the expert, Rhydian, with his usual humor and smily face, said “You have so many smart people working for you. You don’t need artificial intelligence. You’ve got real intelligence.”

The term AI is used very loosely to mean many things, including many traditional statistical methods and ideas. This post is a start on a few less heard of AIs.

Static AI

Techniques manually trained offline, or whose parameters are explicitly coded and then used to make estimates. This is what we have done at many places traditionally and still doing it.

Examples:

• credit risk models using Gradient Boosting Machine (GBM)
• Long short-term memory (LSTM) forecasting model for anticipated attrition rate
• Classification/regression using Random Forests.

Dynamic AI

Automatically retrain parameters periodically during production.

Examples:

• Self-updating recommendation/fraud-detection systems using Neural Networks
• Adaptive clustering based on trends and activities

Auto AI

Dynamic AI + capability of automatically changing their basic structures (e.g. hyperparameters, input variables).

Cognitive AI

Techniques that can autonomously make decisions and take actions accordingly, even on matters for which they were not specifically trained.