From Pencil And Paper To Artificial Intelligence: The New Era Of Psychology

Researchers are exploring how artificial intelligence, particularly large language models, can transform psychological assessment. Instead of relying solely on questionnaires, this technology analyzes people's language quickly and in detail, offering more reliable and accessible results. This innovation can expand the reach of psychology, save resources, and reveal psychological aspects more naturally, but it also requires ethical and privacy considerations.

Accurate measurement is the foundation of psychological science. Over the last century, researchers have developed methods to validate abstract concepts such as intelligence, personality, or anxiety. These methods have enabled significant advances, but they still rely largely on questionnaires in which people answer about themselves.

This type of assessment has many limitations: it can be influenced by what a person would like others to think, by distorted memories, by cultural differences, or even by the difficulty of having a clear perception of oneself.

Therefore, psychologists have sought ways to combine different assessment methods, what we call "multimethod assessment." When different techniques are used together, the results are more reliable. But in practice, this is uncommon because it requires a lot of time, money and effort.

In recent years, advances in artificial intelligence, especially so-called large language models, have opened up new opportunities to improve and facilitate psychological assessment. Human language, which has always been one of the richest forms of expression, can now be analyzed quickly and in detail by these tools.

This is useful because language is an observable behavior (less subject to distortion), occurs in natural contexts (which increases the validity of results), and allows people to express themselves without being limited by closed-ended questionnaire questions.

Furthermore, using language as an assessment tool has practical advantages. Large language models can be applied on a large scale, saving time and resources. For example, while a set of questionnaires may require hours to complete, a short speech or text sample may be sufficient for the model to assess different psychological aspects. This makes it possible to use this resource even in emergency situations or in settings with few mental health professionals.

To understand how this technology evolved, we need to look at the history of language models. Initially, psychologists used simple word-counting programs to associate the use of certain terms with psychological states.

These programs, such as the famous LIWC (Linguistic Inquiry and Word Count), were able to show that the way we use words is linked to our emotions, attention, and even personality. But there was a problem: language is complex, full of ambiguities, and early methods couldn't capture the different meanings of the same word in different contexts.

The solution came with so-called "word embeddings," mathematical representations that allowed us to capture semantic (meaning) and syntactic (structure) relationships between words. Later, more sophisticated models emerged, such as recurrent neural networks and short-term memory networks, which learned to adapt word meanings according to context. While better, these models were slow and required a lot of computing power.

The breakthrough came with the creation of the so-called "transformer" architecture. This model introduced the "self-attention" mechanism, which analyzes all the words in a sentence in relation to each other simultaneously, capturing nuances of meaning and context much more accurately.

This enabled the emergence of large language models, capable of processing enormous amounts of data and performing complex tasks such as translation, text summarization, sentiment analysis, and even natural language generation.

Today, these models can be used in different ways: some specialize in understanding texts (such as for classifying sentiment), others in generating texts (such as writing responses or summaries), and others combine both functions (such as translating or answering questions).

In the field of psychology, large language models represent a unique opportunity to complement traditional questionnaires with richer analyses of people's language. They can broaden the scope of assessments, increase the accuracy of diagnoses, and offer alternatives in resource-poor settings. However, they also pose ethical challenges, such as data privacy and the risk of misinterpretation.

This article offers a guide for psychologists on how to use large language models in psychological assessments. The goal of this overview is to provide accessible guidance on an innovative and complex methodology. Despite rapid advances, relatively little is known about the use of AI models for psychological assessment.

Although a growing number of high-quality studies are emerging, many face limitations related to sample size, diversity, types of linguistic data, or psychological measurement. The article's authors encourage psychologists to seek solid psychometric, methodological, and interdisciplinary contributions in the evolving field of AI for psychological assessment.

READ MORE:

Large Language Models for Psychological Assessment: A Comprehensive Overview

Jocelyn Brickman, Mehak Gupta, and Joshua R. Oltmanns 

Advances in Methods and Practices in Psychological Science. 2025;8(3).

https://doi.org/10.1177/25152459251343582

Abstract: 

Large language models (LLMs) are extraordinary tools demonstrating potential to improve the understanding of psychological characteristics. They provide an unprecedented opportunity to supplement self-report in psychology research and practice with scalable behavioral assessment. However, they also pose unique risks and challenges. In this article, we provide an overview and guide for psychological scientists to evaluate LLMs for psychological assessment. In the first section, we briefly review the development of transformer-based LLMs and discuss their advances in natural language processing. In the second section, we describe the experimental design process, including techniques for language data collection, audio processing and transcription, text preprocessing, and model selection, and analytic matters, such as model output, model evaluation, hyperparameter tuning, model visualization, and topic modeling. At each stage, we describe options, important decisions, and resources for further in-depth learning and provide examples from different areas of psychology. In the final section, we discuss important broader ethical and implementation issues and future directions for researchers using this methodology. The reader will develop an understanding of essential ideas and an ability to navigate the process of using LLMs for psychological assessment.