On Combining the Use of AI with Critical Thinking

Strategic Applied Analytics Leader and Professor Dave Cameron offers commentary on combining the use of AI with critical thinking. This article originally appeared in Insight Jam, an enterprise IT community that enables human conversation on AI.

There is so much talk about artificial intelligence, as we all know. Yet, I’ve been teaching at the university level for the last 5.5 years, and have adapted my courses over that time period as the level of sophistication has improved. Yet, one of the important lessons is that using artificial intelligence is certainly important; it is key to succeeding in today’s world. But I have seen too many people put aside their “real” intelligence and succumb to “artificial” intelligence. I just finished grading finals for 3 classes I’m teaching. What differentiated the great students from the mediocre? The great students used artificial intelligence as an aid to their work and then used their innate curiosity to expand and build on it. The mediocre students submitted the artificial intelligence results.

Let me get more specific. One of the assignments is to take real data from a grocery retailer chain’s loyalty card program and use the past year’s worth of transactions to predict future spend. In this way, the grocery chain can market effectively to its best customers, where the majority of its revenue comes from. The inputs include all details on every transaction for the last 52 weeks. The output is a prediction of revenue for the upcoming 3 months if no marketing campaigns were changed.

It is easy to tell tools such as Claude, or even ChatGPT, to create a Python program that predicts optimal spend using the objectives above, and explain the content of each of the inputs used to predict. For instance, past spend is certainly a predictor of future spend.

Yet in a recent class, roughly 30 percent of the people in the class solely used artificial intelligence tools to develop their predictive equation. One of their conclusions was the the more children in the household, the less people will spend on groceries. And they submitted their final assignment asserting that. Now… think about that a minute. Does that sound right to you? I certainly hope not. The 70 percent of the students that used artificial intelligence as an aid to their work, but not the final product, looked more deeply “under the hood” and figured out what was going on. They modified their equation accordingly – removing that predictor.

See, what happened is that the linear regression developed by artificial intelligence saw that both number of people in the household and number of children in the household were positively correlated with sales and put them both in the model. Yet… those 2 inputs are highly correlated with each other. So, what happened was the AI-based equation said the more people in the household, the higher the predicted sales was one of the strongest predictors. Yet, since number of children was correlated, putting it in the model served to “cancel out” some of the impact of number of people in the household, becoming a predictor saying that the more children the less the sales.

Anyone working in the predictive analytics field for a while, realizes some of the underlying principles on how linear regression works. I taught those principles in class. 70 percent of the students did great. 30 percent never moved beyond the AI-generated approach.

Be curious, be thoughtful, yet use the tools available. I doubt most readers of this would conclude that the more children in the household, the less people spend on groceries. That would not work well in the “real world.”