Modern Psychometrics Drafts

These chapters have been drafted in anticipation of a forthcoming 5th edition of Modern Psychometrics, once loosely projected for around 2030. In today’s fast-moving AI environment, however, waiting that long no longer makes sense—and neither would waiting until 2027. The relevant ideas are live now, and the interaction between psychometrics and AI is evolving quickly enough that the writing needs to evolve with it.

These chapters are intended as methodological interventions, not final position statements. Their purpose is to bring psychometric concepts and constraints into clearer contact with contemporary AI practice—and, equally, to clarify what psychometrics can learn from the ways AI systems now behave (especially in adaptive and interactive settings).

How the Chapters Fit Together

The provisional chapters approach the relationship between psychometrics and artificial intelligence from different but connected directions.

• Current Limits to AI-Assisted Test Development examines a problem arising from contemporary methods of item generation and calibration. It argues that samples may be demographically diverse while remaining cognitively unrepresentative, particularly when experienced online participants are used to estimate how items will function among less practised test-takers.
• Item Response Theory and Adaptive Testing provides the broader measurement framework. It considers how calibrated item banks, adaptive administration and continuing parameter revision can support forms of assessment that are more responsive than conventional fixed tests.
• AI-Supported Statistical Inference develops one possible response to the limitations identified in the first chapter. It asks whether artificial intelligence, curriculum evidence and established psychometric knowledge might provide provisional structural hypotheses and item-parameter priors, which would then be tested, corrected and anchored using carefully selected human data.

Together, the chapters address a common question: How can psychometrics make constructive use of artificial intelligence without losing contact with the human cognition it is intended to measure?

Why Publish Working Chapters?

In a slower-moving field, these arguments might reasonably have remained unpublished until the completion of a new edition of Modern Psychometrics. Artificial intelligence is developing too rapidly for that approach to be satisfactory. Methods, assumptions and practical possibilities are changing while the chapters themselves are being written. Publishing them as working drafts allows the arguments to remain visible, testable and open to correction. It also provides a record of how psychometric thinking is developing during a period in which the relationship between human and artificial intelligence is still taking shape. The chapters will therefore continue to be revised as new evidence, methods and objections emerge. They may be cited, provided that they are identified as developing contributions rather than settled texts.