by Dr. Steve Poulin

Over the past several years, artificial intelligence (AI) has received an increasing amount of attention (and scrutiny) because of the exciting new ways it is being used.  Many people are familiar with the technologies that AI has enabled, but how many people actually know how it works?  In this post, we'll explore what makes artificial intelligence...artificial intelligence.

Many of the technologies that artificial intelligence has enabled feel like they could be straight from a science fiction movie.  AI has enabled robots to perform like humans (although they struggle with some basic tasks like walking).  AI has helped develop new approaches to teaching (such as the intelligent tutoring system which can customize a curriculum for a student based on their individual interests and abilities).  AI has brought us closer to self-driving cars than we've ever been before (by learning traffic patterns and the conditions of the road). 

While all of those technologies feel like they are straight from the future, the reality is that artificial intelligence has been applied to predictive analytics for a few decades. It is the most recent approach in a long history of methods used for making predictions.  Regression models, such as linear regression, logistic regression, discriminant analysis, and Cox regression, have their roots in procedures developed in the 19th century.  Rule induction models such as Chi-Square Automatic Interaction Detection (CHAID) and Classification and Regression Trees (CART) were developed much later, using existing techniques such as the chi-square and Gini tests to generate if/then rules based on branching logic.  AI offers a completely new and more complex approach to predictive analytics.  The key to this advancement? The growing processing power of computers.

Predictive analytics procedures that use AI are similar to regression and rule induction models in their requirements for a target field and one or more predictors.  For instance, AI is helping doctors diagnose health problems by drawing on huge amounts of medical knowledge and identifying complex patterns in the data.  In this example, a diagnosis serves as the target field, and the characteristics of the patient are the predictors.  However, unlike regression, the relationships among the predictors are not critical for AI techniques.  And unlike rule induction models, all predictors are used for making predictions, rather than only the ones with the strongest effect on the target.

This ability to utilize an unlimited number of predictors with few constraints offers a powerful alternative to the traditional regression and rule induction models.  However, AI models are not always more accurate than the traditional models, and they are prone to generating results that are too specific to the data used for the analysis.

Now that you understand the basics of artificial intelligence, we'll take a look at how artificial intelligence actually works.

AuthorRyan Harrington