Boon or bane?

Every day, users of digital media capitulate to the demands of the dominant platforms. Although users of smartphones, tablets and laptops are aware that every single session on the internet leaves behind a record of their online activities, very few people seem willing to alter their behavior to take account of this fact. A study carried out by American psychologists suggests what this information can reveal about the individual user: The study’s authors were able to construct psychological profiles of users of social networks based solely on their choice of likes on these platforms.

This type of inference is made possible by machine learning algorithms, that are designed to recognize patterns in large quantities of data. In order to perform such tasks, algorithms must be trained first. They must learn to identify potentially informative associations between variables, using sufficiently large datasets. In the mentioned study, the subjects were first subjected to a personality test. “And the results of these tests enabled the algorithm to learn which patterns of likes matched up with which the self-reported personality traits,” as LMU psychologist Dr. Clemens Stachl explains. At the end of the training period, a dataset comprised of 260 likes sufficed to allow the program to provide a more accurate description of an individual‘s personality than that person’s family and friends could supply.

LMU psychologist Clemens Stachl is trying to assess the informational value of big data.

Stachl works with Professor Markus Bühner, who holds the Chair of Psychological Methodology and Diagnostics at LMU. Bühner is engaged in exploring the potential, and the risks associated with the use of new digital technologies for research on personality structure. He begins with a qualification in relation to the significance of the studies so far performed in the field. “There is one point one must be careful about. People always refer to these studies as if it were proven that big data can predict an individual’s personality. But what is actually predicted is the result of a personality test based on self-assessment, and we know that such tests have a certain error rate.” For example, not everyone who fills out such a questionnaire does so honestly, others have no precise recollection of their own behavior in certain situations or tend to choose only the most extreme of the proposed responses. Nevertheless, tests like this are currently regarded as the most reliable way of describing an individual’s personality.

One of the urgent research topics in Bühner’s department at the moment is the search for approaches that provide better descriptors of personality. Big data is one possibility. “As researchers, we also hope to gain new insights into the structure of psychological traits in this way.” After all, machine learning algorithms can process far larger datasets than could be analyzed with classical statistical methods.

“The technological developments also open up new ways to collect data,” says Stachl, who is himself currently investigating how much smartphone use can tell us about their owners’ personalities. In such projects, psychologists collaborate with colleagues in the Institute of Media Informatics, who help to develop dedicated apps designed to address the specific issues in question. In this context, data security and the protection of subjects’ privacy play a central role. For example, so far, the smartphone study has focused on collecting abstract quantitative data – such as frequency and duration of smartphone use – in order to avoid any risks to users’ privacy. “Often, it is no longer necessary to ask the subject directly about his or her behavior, one can simply record it,” says Stachl. “Another approach called experience sampling also has great potential, because it is possible via the smartphone to interview people in the situation in which they find themselves at that particular point. So one can now ask a question like ‘how are you feeling?’ several times per day or week, and then evaluate the impact of being in different situations (e.g. at home vs. at work) with regard to the person’s mood.”

Probing personalities The use of big data in the selection and management of personnel is often referred to nowadays as ‘people analytics’. Many employers see the clever use of digital data as an ideal way of finding the best candidates for open positions or identifying the people with the qualities required of business leaders. However, Markus Bühner is skeptical in this respect. “An overemphasis on big data in this area can bring a lot of problems in its wake,” he remarks.

First, there is the matter of data protection: What sorts of data may be collected for what purposes and for how long? “The primary criterion for the use of a given method of personnel evaluation and selection is its relationship to the demands made by the post on offer,” says Bühner. In Germany, the standards that govern the assessment of professional aptitude are set out in a DIN norm. This specifies that all the information and data collected about the candidates must be related to the demands associated with the position in question. “So anyone who wishes to use big data in this context must first decide what sorts of information are of relevance to the position to be filled,” says Bühner. “One can, of course, arrange to have candidates observed 24/7. But what can one do with the knowledge that someone gets too little sleep?” Even when one has a clear idea of the kinds of data to be assessed, the next problem turns up. Developing algorithms that can parse data correctly requires a lot of expertise. Moreover, one must keep in mind the fact that algorithms can rapidly go out of date. – And continued use of obsolete algorithms increases the risk that one ends up with the wrong person for the job.

Furthermore, partly as a result of demographics, all the signs indicate that the labor market will contract in the coming years, making it even more difficult for companies to find the right job candidates. In this light, Bühner regards a naive faith in technological solutions as problematic. “In my view, we need to find approaches to the selection and management of personnel that are better than the digital surveillance of staff or digital evaluation of job candidates. The challenge in the coming years will be to find ways to enable staff members to employ their skills and develop their capabilities on the job, and to enjoy a long and healthy working life.” According to Bühner, analyses based on big data may well have a place among the tools used in personnel management, but they are unlikely to replace the older set. And he emphasizes the need for a wide-ranging discussion on data protection and the ethical use of data. “Just because something is possible does not necessarily mean that it should be done,” he says. With respect to other potential applications for big data, he is more optimistic, citing their promise as a tool for the prevention of depression as an example.

In light of China’s embrace of big data, his assessment of the risks becomes particularly pertinent. In 2020 China plans to introduce a Social Credit System, which will then apply to all citizens, after having been tested in selected regions over the past few years. From then on, the behavior of every citizen will be continuously monitored. “And correct behavior will be rewarded. That represents a strong incentive to behave in conformity with expectation and to follow the recommended patterns, for those who don’t obtain no social credits,” says Clemens Stachl. For Markus Bühner the threatened withdrawal of rewards is the key, “the magic word – that almost always works.” He sees the program as an experiment in social psychology. It also presents researchers with a dilemma: How should they handle data collected in this way? Bühner is confident that not every researcher will want to work with all the data available. “It is also, as always, a question of ethics.”