What’s New In Assessment?
What’s New In Assessment? Tomas Chamorro-Premuzic, Contributor Thu, April 23, 2026 at 7:56 PM UTC
Add Yahoo Tech on Google
Thu, April 23, 2026 at 7:56 PM UTC How AI and related technologies are reshaping talent identification and recruitment
Giant robot flicking tiny man. Ai technologies and unemployment problem concept. Vector illustration.
getty Despite a century of scientific progress on how to identify, assess, and select employees, managers, and leaders, a persistent gap remains between what we know and what organizations do. In practice, most recruiters and hiring managers still rely on intuition or favor unproven methods, such as unstructured interviews, dubious psychometrics, subjective referrals, and politically tinted preferences, rather than evidence-based tools that reliably predict performance and explain meaningful differences in potential.
Recent years have brought a wave of technological advances that promise to change this. Some scale established, science-based methods; others offer cheaper, faster, and more user-friendly alternatives. Crucially, these innovations do not necessarily improve predictive accuracy or explanatory power. Yet they may still represent progress relative to the status quo, if only because the baseline is so low. For example, algorithmically scored video interviews may not outperform well-designed structured interviews conducted by trained assessors, but they often surpass the typical unstructured interview conducted by an untrained one. Similarly, generative AI tools that infer personality or leadership traits from digital footprints may fall short of validated psychometric assessments, but they are not necessarily worse than having no assessment at all, or relying on popular but scientifically weak instruments such as the MBTI .
Advertisement Advertisement That said, from both a business and ethical standpoint, it is not enough to adopt a method simply because it is marginally better than a flawed baseline. If autonomous vehicles reduced annual road fatalities from 1.2 million to 1.1 million, few would consider that an acceptable outcome. Likewise, while reducing the proportion of incompetent hires from 30% to 25% may be an improvement, it is hardly sufficient if more rigorous, evidence-based approaches could bring that figure closer to 10%. The real opportunity is not incremental improvement over poor practice, but a decisive shift toward methods that are both scientifically grounded and practically scalable.
Importantly, regardless of how emerging technologies, including AI, evolve, science still provides the most robust criteria for evaluating their value. Concepts such as reliability and validity remain the closest thing we have to a north star for assessing the quality of any tool, no matter how novel or impressive it may appear. Ultimately, what matters is not how sophisticated a method looks, but how consistently and accurately it predicts meaningful outcomes, such as job performance, leadership effectiveness, or productivity. At its core, assessment remains a probabilistic exercise, and some measures are demonstrably better predictors than others. Crucially, prediction alone is not enough. The best assessments also explain , offering a theoretically grounded account of why certain traits, scores, or behaviors are linked to subsequent outcomes.
Consistent with this, academic research on emerging assessment technologies, including the many variants now grouped under the umbrella of AI, has become a rapidly expanding domain within industrial-organizational psychology . While the proliferation of tools, platforms, and start-ups makes it difficult for research to keep pace with every new product, most innovations can be grouped into broader categories of assessment (as we first did a decade ago ). This allows us to evaluate their likely validity and utility based on established evidence, rather than novelty alone.
OK, so let’s take a look at some of the salient innovations around assessment, and what we can conclude about their accuracy or potential based on actual scientific evidence…
Advertisement Advertisement 1) Video interviews (AI-scored or asynchronous)
Research on structured interviews remains unequivocal: when standardized and scored systematically, they are among the best predictors of job performance. AI-enabled video interviews aim to scale this logic, often by coding verbal and non-verbal cues. However, the evidence here is mixed. While studies show that structured asynchronous interviews can achieve acceptable reliability and moderate validity, claims that facial expression or “micro-emotion” analysis predicts performance are not supported by robust evidence. Recent reviews caution against overinterpreting these signals. In short, the gains come from structure and standardization, not from AI per se .
(2) Digital footprint scraping (social media, online behavior, passive data)
The idea that personality and ability can be inferred from digital traces has gained traction following work by Kosinski and colleagues , showing that Facebook “likes” can predict personality with moderate accuracy. However, translating these findings into selection contexts is far more problematic. Meta-analytic evidence is still sparse, and concerns around construct validity, stability, and adverse impact remain substantial. Reviews suggest that while signal exists, it is noisy, context-dependent, and often inferior to direct psychometric measurement. At present, this approach is better seen as supplementary rather than substitutive.
(3) AI simulations, immersive, and gamified assessments
High-fidelity simulations and work samples have long been among the most valid predictors of performance (often exceeding .50 in meta-analyses). The current wave of AI-driven simulations, including virtual assessment centers and immersive environments, seeks to scale these methods. Evidence is cautiously positive. Studies on simulation-based assessment suggest strong criterion-related validity when tasks closely mirror job demands. However, the addition of AI (e.g., automated scoring, adaptive scenarios) has not yet consistently demonstrated incremental validity beyond well-designed simulations. The promise lies in scalability and standardization, not necessarily in improved prediction.
(4) Natural language processing (NLP)
NLP has emerged as one of the more promising areas. A growing body of research shows that linguistic patterns in written or spoken responses can predict personality, cognitive ability, and even job performance proxies. For example, studies using open-ended responses and essay-based assessments report moderate correlations with traditional measures. Similarly, language-based indicators of cognitive ability show incremental validity in some contexts. That said, results vary significantly depending on the model, training data, and task. Importantly, NLP tends to work best when grounded in theoretically meaningful constructs, rather than as a purely data-driven exercise.
Advertisement Advertisement (5) Game-based and AI-enabled assessments
Game-based assessments aim to infer traits such as cognitive ability, risk tolerance, or persistence from behavioral data captured during gameplay. Early studies suggest that such tools can achieve acceptable reliability and modest validity, particularly when designed to measure well-defined constructs like cognitive ability. However, the evidence is uneven. Reviews indicate that many commercial offerings lack transparency and validation. When properly designed, these tools can approximate traditional measures; when not, they risk becoming engaging but uninformative proxies.
(6) Other approaches
Two additional categories deserve mention. First, algorithmic composite models, which integrate multiple data sources (e.g., CV data, assessments, interviews) using machine learning, have shown promise. Evidence from personnel selection research consistently shows that combining predictors improves validity, often substantially. Second, continuous or longitudinal assessment, enabled by workplace data (e.g., performance metrics, collaboration patterns), represents a shift from one-off selection to ongoing evaluation. While still emerging, this approach aligns with evidence that performance is dynamic and context-dependent.
The bottom line
Across all these innovations, a consistent pattern emerges. Technology can enhance scale, efficiency, and candidate experience, but it rarely substitutes for sound measurement principles. The best-performing tools are not those that look the most innovative, but those that adhere most closely to established scientific standards: clear constructs, reliable measurement, and demonstrable links to real-world outcomes.
Advertisement Advertisement In that sense, the future of assessment may look radically different on the surface. Underneath, however, the fundamentals remain remarkably unchanged.
This article was originally published on Forbes.com
View comments About our ads
