By Jorge Alvarado, Director of the Master of Science in Analytics for Business Intelligence, Pontificia Universidad Javeriana (Bogotá, Colombia)
It is a common view that typical white-collar jobs will disappear with the rise of intelligent machines. While some see this as the dark side of Analytics/Big Data/IOT, it is rather a challenge for educational systems and the future role people with have in their jobs.
«The replacement of humans in jobs -ranging from to help desks to automated cars- has been going on for many years. However, it has been taking a new form in recent years. Whereas machines where traditionally replacing manual labor, they are now also used to accomplish routine information processing and even decision making under uncertainty. The reason for this is simple: Analytic techniques fed with enough information perform better in most occasions.»
Having said that, most jobs will probably not disappear but will rather be reshaped, as in previous technological revolutions. Particularly, I believe that the situation-specific surveillance, updating, and intervention of analytical models, are going to be highly valued skills in the future. This has already happened (think about autopilots in airplanes or automated car assembly), but it is going to be more important in the near future. Let me explain my point with two examples.
The pricing algorithm of Uber is a classic example of analytical models fed by big data making automated decisions in almost real time. During the 2014 Sydney Hostage Siege, the algorithm charged 4 times its usual rate, and the company had to apologize. Performance during Paris attacks during 2015 was not as outrageous, but it still showed a dubious performance; the cities Uber drivers were unable to offer fairs free of charge, as done by most traditional taxi drivers. This example calls for a new point of view on analytical models, where algorithms must be able to detect strange behaviors by themselves and call human surveillance for help. This, in turn, calls for control possibilities allowing to adapt the algorithm by human beings- whether centralized or distributed- and a new kind of job skill is, indeed, wanted and necessary.
«Recent research in the field of demand forecasting has shown that analytical models are still far away from knowing the rise and impact of unexpected and sudden environment. At the same time, forecasts based on historic track data are frequently outperformed by analytical models.»
As a consequence, an effective interaction between the model and human experts needs to be designed, adding up strengths and trying to avoid weaknesses from both sides. Psychological traits of experts, such trust and their credibility in computer forecasts, seem to play a role in this process design.
However, educational systems are slow to adapt. A second (still unpublished) research has shown that traditional forecasting methods are ineffective when addressing the common industrial task of deciding whether to override an analytical forecasting model and deciding the size of any necessary modification. Traditional forecasting training is usually focused on teaching all the steps of historical track forecasting techniques (such as exponential smoothing and ARIMA). When facing the decision whether or not to modify system forecast, most students with traditional training tried to duplicate the algorithm, leading to poor results. Students trained on algorithms, human limitations, and best practices concerning the interaction with forecast models performed much better.
Are educational systems preparing students to adapt to machine results, critically discuss it in their changing environments and modify algorithms when necessary, or are they just preparing students to duplicate the machine job? In the first case, new opportunities, jobs and wealth will be created; in the second one, failure and unemployment can be expected. Are companies rethinking their automated jobs regarding the human-computer interaction paradigm, or are they naively thinking that Big Data algorithms will solve everything by themselves? If you have so far believed in the latter approach, I challenge you to think twice.