Is AI at odds with the scientific method? I listened to a Podcast interviewing a CEO of an AI company the other day, and (leaving out identification) he said that “AI is the scientific method turned inside out…the scientific method….has been used to build literally everything in our lives – literally everything – nothing in our lives was built with a statistical model – not one thing – not one thing…” In other words, I heard that he was suggesting that AI is at odds with the scientific method.

Hmmm…well, I would disagree that AI turns the scientific method inside out. I would argue that the methods are complementary, and not only that, but that AI is an automated step within the traditional scientific method. AI, and more specifically machine learning, recognizes patterns. This is the first step of the scientific method.

AI vs. The Scientific Method

Traditionally, a human would notice a pattern as the first step in the scientific method. For example, a doctor might recognize a pattern in which the adults she is treating seem to have more health ailments when the adults also report having experienced childhood trauma, as Nadine Burke Harris, California’s Surgeon General, observed. Asking the question, “does childhood trauma have an impact on adult physical health” helps formulate the hypothesis “yes it does.” Next, experimental (gold standard randomized controlled trials) or quasi-experimental (more often in the case of retrospective observational studies) are used to control for other possible causes and confounders to test the hypothesis. A controlled statistical model (yes, statistical models are a tradition in a scientific method) then estimates the various contribution that traumas have on physical health.

So where does AI fit? First, it can generate a massive amount of “observations,” which is the first step in the scientific method. Relying on a human to recognize patterns when data has already stored these patterns, likely for years, is inefficient. AI will also recognize more complex patterns than a human might. A challenge is that it will recognize *a lot* of patterns. However, AI can also help prioritize the patterns based on potential impact and importance, helping to identify which patterns to test first.

AI can also be use to predict future outcomes. While the scientific method is more concerned with the “Why,” AI focuses on the “What” as the why in AI is often a black box. It is true that correlation is not causality, but it is also true that the best predictor of the future is the past. The caution and challenge here is to not use AI in such a way as to repeat past mistakes, such as institutional bias…but this can be carefully and thoughtfully addressed.

As more and more data amasses, we will find patterns in entire populations. The patterns are real – why the patterns occur is the question. Implementing the strengths of all approaches and marrying them where it makes sense will result in more efficient and productive science. AI is a foundation of a more efficient scientific method.