In a world dominated by the crisis of the corona virus, we hear about the uncertainties associated with trying to predict the trajectory of the outbreak. Mathematical models are used to forecast the number of cases as well as the number of deaths. Those models are used to estimate when the outbreak of the virus will peak. However, mathematical models are based on assumptions, and assumptions may or may not be correct. This is the case because unknowns can affect the assumptions.
As a result, we have been periodically seeing media mentions of an enduring quote attributed to noted statistician George Box. Quite some time ago, Box said, “All models are wrong, but some are useful.”
As someone experienced using predictive models for forecasting in business, I’d like to offer an example to illustrate the “All models are wrong, but some are useful” statement. A good example of this is weather forecasting. Weather forecasts are based on models. Weather forecasts are sometimes wrong. Yet, those weather forecasts are right often enough to be useful. Most of us rely on weather forecasts to determine whether we should wear a heavy coat or bring an umbrella. We do this despite the fact that those forecasts are not always right.
Along similar lines, models are being used to predict the impact of the corona virus. As the government health experts have explained, the models are based on assumptions and those assumptions entail complexities. There can be unknowns. Yet, those models can be useful to help guide decisions about how to deal with the virus.
All of this has implications for the many businesses that have been striving to be more data driven and are working to apply AI (artificial intelligence). AI involves the computer learning from past data, so it can develop models used for prediction. Thus, Box’s statement, “All models are wrong, but some are useful” has important implications for AI.
Box’s statement is especially relevant to companies’ expectations about AI. His statement acts as a reminder that AI is not a magic bullet. In fact, in some cases, AI models can go terribly wrong due to factors such as bias. Thus, Box’s statement serves as a reminder to be especially careful about blindly accepting the results of AI. That’s why it is important to value human input to AI models. Yet, as Box’s statement tells us, AI models can be useful.
So, in conclusion, when applying AI in business, remember Box’s statement that “All models are wrong, but some are useful.” Since models can be wrong, be careful about blindly accepting them. But, since models can be useful, strive to improve them, remember the value of human input, and apply the models where their use is beneficial.
