If you would like to be notified when new content is posted to our website, subscribe to our email alerts.Subscribe Now
Recently, in an article in Scout magazine entitled: “The Rise of the Weaponised AI Propaganda Machine”, a series of claims was attributed to Michal Kosinski, one of the leaders in personality assessment using social network activity linguistic analysis:
“According to Zurich’s Das Magazine, which profiled Kosinski in late 2016, “with a mere ten ‘likes’ as input his model could appraise a person’s character better than an average co-worker. With seventy, it could ‘know’ a subject better than a friend; with 150 likes, better than their parents. With 300 likes, Kosinski’s machine could predict a subject’s behavior better than their partner. With even more likes it could exceed what a person thinks they know about themselves.”
The claim is based upon evidence reported in a series of academic articles showing that the accuracy of personality assessment in this way meets or exceeds that of self-report questionnaires. A couple of these articles are referenced below.
The Scout magazine article makes it very clear that personality assessment based upon Facebook and Twitter ‘digital footprints’ is now a reality, in use, and deployed in earnest by at least one organization (Cambridge Analytica).
I also explored the Das Magazine article: which provided even more information about how far this ‘personality psychographics’ work is being picked up by politicians, marketeers, and commercial entities.
These are must-reads – actually riveting reads. Not only are they informative about the data-scraping of personal information resulting in the personality profiling of 220million US adults, whose profiles are now being sold by Cambridge Analytica, but just how far political leaders in other countries are now buying what’s on offer, for ‘influencing the population’ reasons.
And the Psychometrics Centre at the University of Cambridge, UK is now advertising the “Apply Magic Source” PredictionAPI service for which they claim:
“By analysing aspects of a given user’s online behaviour, our PredictionAPI engine can forecast a range of variables that includes personality, happiness, intelligence, entrepreneurial potential and more.”
If you can bypass the self-reports altogether – and go directly for:
Behaviour ==> personality attribute ==> criterion outcome analysis
then you basically have the most accurate methodology available for predicting the criterion outcomes – because it cuts out the potential for purposeful impression-management and the lack of personal insight of some who complete self-report questionnaires.
As soon as large organizations dealing with huge applicant lists realise they can buy the applicant assessment results they need from organizations like Cambridge Analytica or the Psychometrics Centre, without ever having to engage-with/ask candidates to complete any self-report questionnaires, the market for self-report will begin to shrink substantially in favour of computer-based direct behaviour-based assessment reports of candidate psychological attributes and outcome/risk predictions.
Basically, the applicant process and shortlisting can be entirely automated using judicious target-candidate profile-matching using ‘discrete’ AI-based assessments. And, as Cambridge Analytica have shown us, this new technology can be applied by governments just as well as commercial organizations.
The commercial and I/O implications of these realities are significant, not just for test publishers who rely for much of their profitability on selling web-delivered self-report questionnaires, but for applicants themselves. Because if an applicant does not possess a digital footprint/searchable activity (e.g. Facebook, Twitter, LinkedIn, Yahoo! Answers, etc.), they may find themselves excluded altogether from entire sectors of the job-market.
Two test publishers have already publicly announced their activity and new products in this brave new world:
Hogan-X – who also list Michal Kosinski as an adviser to their project:
“Hogan X is translating digital behavior into insights that people will use to understand themselves in greater depth. We’re creating tools to apply those insights to work, relationships, teams and well-being.”
“… provides you the most comprehensive source of global talent demand and supply data, predictive analytics and insights into real-time job market, location, and competitive intelligence to help you make smarter talent planning and recruiting decisions”
Some might say that personality assessment based on social media responses seems rather superficial, biased, “circumstantial”/secondary, part of the persona / part of social masking. Indeed, one might inquire whether a “like” indicates support of a person or agreement with a point of view or interest in a topic. It all seems too vague, too remote from what we might consider as relevant indicators of an individual’s personality. Yet, the empirical results indicate this kind of assessment can do just as well, if not better, than self-report.
However, these results are all based upon ‘aggregation’. That is, the results are all about aggregates of individuals, based upon averages computed over thousands of individuals. So, what may work for some will be inaccurate for others, but the overall trend is that in terms of the statistical property of ‘explained variance’ in some criterion attribute, the psychographics approach will be equivalent and sometimes better in predictive accuracy than a self-report questionnaire assessment.
Ultimately, what the Big Data analysts are trying to do is create a new form of assessment which is more convenient and more automated than a self-report assessment, less or completely unaffected by faking/impression management, but also more accurate in terms of predicting relevant outcomes.
But this kind of application is predicated upon two big assumptions:
That first assumption relies upon either the usual “intrapsychic trait as cause” (we all possess some magnitude of a neurologically-mediated trait-attribute which causes us to behave in certain ways) or a version of “past behaviour is predictive of future behaviour”, where past linguistic behaviour on the internet is inferred to be predictive of future workplace outcomes. No unambiguous empirical evidence exists to substantiate the first interpretation. The second interpretation relies upon personality per se being predictive of workplace outcomes. It is, but according to the latest evidence reported in Schmitt(2014) and Schmidt, Oh, and Shaffer (2016), only marginally so.
The second assumption seems implausible from first principles let alone the empirical evidence supporting the opposite. However, adaptation may not endure if an underlying feature of your psychology is having to be continually ‘suppressed’ in order to maintain the required behaviours.
Interestingly, where all this analytics activity may have less impact is the market in which ‘deep psychology’ assessment-providers inhabit. Why? Because when needing to select high-stakes, high-autonomy individuals, you need to know far more about the substantive functional psychology of an individual than can ever be gleaned from self-report or data-scraped linguistic-behavioural analysis.
Park, G., Schwartz, H.A., Eichstaedt, J.C., Kern, M.L., Kosinski, M., Stillwell, D.J., Ungar, L.H., & Seligman, M.E.P. (2015). Automatic personality assessment through social media language. Journal of Personality and Social Psychology, 108, 6, 934-952.
Schmidt, Frank L. and Oh, In‐Sue and Shaffer, Jonathan A. (2016) The Validity and Utility of Selection Methods in Personnel Psychology: Practical and Theoretical Implications of 100 Years of Research Findings (October 17, 2016). Fox School of Business Research Paper.
Schmitt, N. (2014). Personality and cognitive ability as predictors of effective performance at work. Annual Review of Organizational Psychology and Organizational Behavior, 1, 45-65.
Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments are more accurate than those made by humans. PNAS, 112, 4, 1036-1040).