November 28, 2022

AI – someone needs to know what’s going on!

The point of the questions is to make us think about statistics and correlations and consider the cause and effect and whether the relationship is meaningful.

---

You may recall, a few years back, a deep internet discussion concerning the lethality of cheese [1]. Maybe not. It was an interesting discussion of a case from a website devoted to finding interesting correlations in disparate data. Fig. 1 shows the chart which generated a lot of interaction: the annual per capita consumption of cheese in the USA and the number of annual fatalities from people being entangled in their bedsheets. Does eating more cheese put us at increased risk of bedsheet fatality?

Some of the discussion was purely statistical – is there a third factor which is a common cause? Does the data merit a statistical correlation analysis at all [2]? I happened to present the chart at a couple of events just to promote data analysis discussion and got a generally amused but dismissive response from the audience: “it’s just a coincidence”. So, I posed a question to the audience: “How do you know what your cheese is doing right at this moment in your refrigerator?” A question I feel is especially reasonable as the refrigerator light goes out once you shut the door. And I didn’t stop there… “How do you know cheese isn’t plotting your demise right now? Plotting with the mayonnaise?” A brief discussion of monitoring/surveillance occasionally follows. The point of the questions is to make us think about statistics and correlations and consider the cause and effect and whether the relationship is meaningful. Is the relationship useful in predicting outcomes based on new data? Who can tell? With the cheese, we all likely have some relevant experience and some a priori knowledge. In other cases, we may need someone who knows what is going on… a subject matter expert (SME) who can make sense of what we find while mining data.

Subscribe to Transformers Magazine