'Human expert augmented by AI is the best hiring formula': Dr Tomas Chamorro-Premuzic

Dr Tomas Chamorro-Premuzic is professor of business psychology at University College London and Columbia University

48-Dr-Tomas-Chamorro-Premuzic

Interview/ Dr Tomas Chamorro-Premuzic, professor of business psychology at University College London and Columbia University

DR TOMAS CHAMORRO-PREMUZIC, psychologist and author, has been instrumental in building science-based tools that help companies predict performance of employees and people’s ability to understand themselves. As an expert in AI, he has often cited studies showing how “AI can make more accurate estimates of our personality than not just our friends but also ourselves”.

The best tests don’t take your answers at face value. Think of assessments as an invitation to present yourself in a positive way. If your scores predict your performance, the assessment is valid and useful, irrespective of whether you think you were honest.

In his book, I, Human: AI, Automation, and the Quest to Reclaim What Makes Us Unique, he recalls that 20 years ago, when he was running research experiments for his doctoral thesis, he had to beg people to take psychometric tests. However, in the AI age, he says “the behavioural DNA of our habits, including our most intimate preferences, our deepest, most private thoughts and our guilty pleasures, has been turned into a vast reserve of information so that algorithms can learn all there is to know about us”.

Based in Brooklyn, New York City, he cofounded DeeperSignals and Metaprofiling. An international authority in psychological profiling, he was the former CEO at Hogan Assessment Systems, which is known for its state-of-the-art personality assessments. He is also the professor of business psychology at University College London and Columbia University. With roots in Buenos Aires, Chamorro-Premuzic established his career in London before moving to the US. In the past, he has held academic positions at the New York University and the London School of Economics and frequently lectures at Harvard Business School, Stanford Business School, London Business School and International Institute for Management Development.

He has bagged awards from the American Psychological Association, the International Society for the Study of Individual Differences and the Society for Industrial-Organizational Psychology, to which he is a fellow.

Currently serving as chief talent scientist at Wisconsin-based Manpower Group, one of the largest staffing firms globally, he spoke to THE WEEK exclusively on the evolution of personality tests and how AI will impact the personality testing industry. Excerpts:

Q/ What are the most significant changes you have observed in the personality testing industry?

A/ No major changes in the past 50 years, just cosmetic changes: shortening the tests, adding pictures, using AI to improve the scoring key and so on. There has not been true innovation in the field and most of the so-called innovations have made them worse. The most significant advancement in the field is that you can now scrape passive data to infer people’s personality without having to give them an assessment. In that sense, the biggest innovation has been to make assessments irrelevant (though legal constraints regulate against the use of passive data scraping, keeping traditional assessments in business).

Q/ Are today’s personality tests successful in cherry-picking high potential employees (HiPos)?

A/ Personality tests can certainly enhance HiPo identification by going beyond past performance and zeroing in on potential, soft skills and leadership abilities, even when candidates have not managed people before. They are also better than alternatives such as managerial ratings, CVs, past performance or experience, qualifications and 360 [degree feedback]. They are also better than IQ tests, though they mostly complement them.

Q/ How reliable is the use of AI in hiring, whether it is personality tests or interviews?

A/ AI is too big to answer this question. It includes CV-parsing tools, natural language processing (NLP), gen AI, machine learning, neural networks, deep learning, and many other ways to turn data into insights. It is more interesting to look at the data in question: how reliable is video interview data, voice, speech, language, behaviour in X or Y context and so on.

In general, AI has been useful in identifying some signals that predict potential where other data science methods don’t, but the overarching finding so far is that it is mostly helpful in terms of costs, speed and efficiency, including improving user experience, rather than accuracy. The best AI is more accurate than most humans, an average AI is as accurate as a competent human, but a human expert augmented by AI is the best formula to date. Ultimately, it all depends on the quality of data.

If you train AI to predict who gets promoted in a firm, it may select Machiavellian sociopaths who are confident rather than competent. If you train it to identify the people who add most value to a company, it may identify the best workers. But they will be dismissed by human raters who are used to politicians and alpha males.

Q/ How do modern assessments tackle test-takers who pick socially desirable responses to score well?

A/ Well-designed assessments invite people to fake; they assume they will. Success in every area of life involves impression management, deception and socially agreeable behaviour. When an interviewer asks you if you love working with others, the honest answer is: “it depends on who they are”. But if you answer that, you are probably antisocial and shouldn’t get the job.

Likewise, the best tests don’t take your answers at face value. Think of assessments as an invitation to present yourself in a positive way. If your scores predict your performance, the assessment is valid and useful, irrespective of whether you think you were honest.