Invisible gorillas prove you need automated text analytics. Seriously! (7-minute read and videos about inattentional blindness)
We are awful at observation. Our brains just don’t work all that well. Various psychological experiments lead to the inevitable conclusion that our brains work hard to fit whatever we perceive into known, existing categories. The part of this phenomenon that I want to cover here is called ‘inattentional blindness’.
But what has this got to do with gorillas, you may well ask? Good question. Especially if you don’t already know what this is about, please watch this video now:
I was surprised the first time
Yes, the results are surprising. And no, I did not see the special visitor the first time I saw this video some years ago. But let’s take it a little further. People who spend their lives doing detailed and specialized observations must be far better than average at this, right? Well, Dr. Trafton Drew designed a different version of this experiment and tested it on a set of radiologists; people who spend a substantial portion of their time inspecting patient X-Rays and CT scans for evidence of medical issues. This video shows what he found:
So, what does this have to do with text analytics?
Something like six years ago when Meg Whitman was CEO she called all HP VPs together in Anaheim. We spent two days together to improve our understanding and to workshop different aspects of the corporate strategy. The main meeting room was laid out ‘cabaret style’. This means round tables of about eight people each. Each table had a shared PC that was connected to the conference network. Early in the final afternoon we were asked to chat at our tables for about 20 minutes and provide a list of what we felt were three most important points covered during the conference. I was working in our Autonomy software group at the time and the exercise quickly caught my attention. I determined that a woman at the table next to mine had been given the task of inspecting all the input and establishing an overall ranking. Since all input was in free text, that seemed quite challenging.
I quickly went to our head of R&D (Sean Blanchflower) and asked whether he could use our IDOL software to do some quick text analytics to do the same classification exercise in an automated way. The answer was yes. I gave him a copy of the input file and he dashed off to do it in his hotel room. When complete, we compared the list that Sean / IDOL gave with the list my other colleague had completed. Guess what? The top five items on each list were totally different. Further discussion revealed that the IDOL list was far more accurate and that my other colleague had already partly decided what the list should be before ever looking at the data. She simply found it easier to see the items she expected than those she did not recognize. She was a victim of ‘inattentional blindness’.
Text analysis is far from perfect, but it is unbiased
Regular readers will know that the current state of text analytics is far from perfection. Even the best Natural Language Processing algorithms do not seem to be able to handle pronouns or context with any level of accuracy. Current best practice is to allow text analytics to supplement and guide human observation. The Net Promoter System and other CX methodologies gather open text input about potential improvements. Unfortunately, most humans looking at such data believe they already know what the list should be. They are trying to count the number of times the people in white pass the ball. They miss everything the people in black are doing, and of course they don’t notice the unexpected gorilla at least half the time.
Conclusion
Where you have text to analyze for CX or other purposes, combine software and human analysis to make sure you don’t miss your personal gorilla.
BTW, the gorillas seem quite upset about not being noticed…
While I have written about text analytics in our books, the above is new content. All of our books are available in paperback and Kindle formats from Amazon stores worldwide.
