Challenge Everything, Including Artificial Intelligence

I was listening to the “People I (Mostly) Admire” Podcast this week and Steve Levitt’s guest was Sendhil Mullainathan, author of the book Scarcity (great read). I enjoy this podcast very much because it features interesting conversations between people who are no doubt smarter than I am. I always come away with an interesting idea, this week there were several.

First, he challenged the notion that Artificial Intelligence (AI) is superior to human intelligence. His view (and mine) is that it is superior for some things, but for others it is far weaker. The example he gave is to ask your Alexa to not tell you what the weather will be today. If you do she will quickly tell you what the weather is going to be. A simple example, but it provides some insight as to why AI is not, and might not ever be, a replacement for human intervention. This is why at TRC we don’t rely on things like text analytics to replace humans, but to enhance their ability to maximize the value from unstructured data. Our InTXTigator tool is great for that.

Second, he suggested that AI’s ability to analyze unstructured data provides us with a huge opportunity to gain a greater understanding of whatever we are studying. The ability to combine these data with traditional quantitative data will drive deeper insights. At TRC we not only work to combine these types of data (for example by including sentiment scores of an open end into a regression analysis) but we bring elements of each into our data collection to maximize the potential of the data.

Third, and most interesting, he suggested you need to challenge assumption, even deeply held assumptions. One example he gave was on the placement of sensors for EKG’s.  You might assume that the placement of these sensors was something that was carefully studied to maximize results.  In reality, the placement dates to the early days of EKG’s and while it generally works, it is only recently that people asked if better placement could lead to better data. For example, they are never placed on the back…could doing so provide information currently not being measured? I love this idea. I’m not suggesting we throw out long held beliefs about things like sample design, but rather that we be open to testing these assumptions…rigorously and thoughtfully. Who knows what we might be missing.

In short, this episode both validated my long held belief that curiosity is the most critical skill needed to be a good researcher AND challenged me to be even more curious!