Artificial Intelligence and Psychology: Limits, Complementarities, and Skepticism

Prof. Eduardo de Campos Valadares, DF/UFMG

5 minutes

“AI will do the analytical thinking, while humans will wrap that analysis in warmth and compassion.” — quotes from AI Superpowers by Kai-Fu Lee

Introduction

Artificial intelligence (AI) has advanced rapidly in various fields of knowledge, sparking debates about its role in replacing or complementing human professions, especially those involving complex interpersonal relationships, such as psychology. This essay discusses the possibility of AI replacing psychologists, the skepticism toward contemporary psychological practices, and the ethical and human implications involved in this context, through a dialogue between the author and an AI reflecting on these themes.

Psychology and Artificial Intelligence: Different Natures and Functions

AI acknowledges that psychologists play an irreplaceable role, grounded in human elements such as empathy, active listening, emotional bonding, and the interpretation of deep emotional contexts — characteristics that go beyond the current capabilities of artificial intelligence (AI). Although AI can complement the work of psychologists by providing information, initial listening, and suggestions for exercises, transformative therapeutic support depends on a qualified human relationship.

This distinction highlights the limitation of AI in creating genuine emotional bonds, even though it can recognize patterns and reflect alongside the patient. Psychology, therefore, remains an essentially human practice, whose power lies in the subjective experience shared between therapist and patient.

Skepticism Toward Psychology: A Metaphor of Alchemy

The author’s skepticism toward psychology is compared to seeing it as an “alchemy” — a hybrid field between philosophy, biology, and medicine, which still employs rudimentary tools to understand the human mind. Just as alchemy was a precursor to chemistry, contemporary psychology may be groping toward aspects of the mind and subjectivity that, in the future, might be better understood through neuroscience and other fields.

However, psychology cannot be reduced to merely decoding neural mechanisms, as subjective experience, emotional bonds, traumas, and personal narratives are dimensions that resist purely biological explanation. Considering this complexity, psychology — despite its limitations — contributes to the health and well-being of people facing mental disorders.

A comparison chart illustrating the differences between psychology and AI users, detailing input, output, and interventions with a focus on subjective models in psychology and data processing in AI.
Comparison between psychology and AI users: both deal with complex systems whose internal mechanisms they do not fully understand. In psychology, subjective models of the mind (e.g., Freudian psychoanalysis, humanistic approaches) interpret input data such as childhood experiences, social stressors, or emotional trauma, producing outputs like behavioral predictions, diagnoses, or classifications. Interventions involve introducing more or different data, such as therapeutic dialogue, new environments, or behavioral exercises. Similarly, AI users provide inputs (training data, feature sets) to neural networks, without fully understanding their internal mechanisms, and obtain outputs such as inference, prediction, regression, and classification. When the results are unsatisfactory, they adjust by supplying more or different data. Risk: Garbage In, Garbage Out — low-quality inputs lead to poor and unreliable outputs in both domains.

Critiques and Practical Limitations of Contemporary Psychology

The author’s personal experience reveals a critical aspect of current psychology: the sense that many professionals operate mechanically, as “merchants of the human soul,” offering formulaic listening and stock phrases without fostering real transformation. This critique points to the existence of practices lacking authenticity and effectiveness, leading to stagnation and frustration in patients.

In addition, a specific problem is highlighted in couples therapy conducted by two psychologists, where neutrality is compromised by unconscious biases that may reinforce divisions instead of promoting reconnection. This situation underscores the influence of power structures and subtle judgments, even in therapeutic contexts.

Another delicate issue is ethical negligence when serious clinical signs are ignored, such as the refusal to refer a patient for psychiatric evaluation, which can amount to quackery. The persistence in ineffective therapeutic models, which foster emotional or financial dependency, further undermines trust in the psychological field.

Infográfico comparando psiquiatria e engenheiros de aprendizado de máquina, detalhando como funciona cada campo, incluindo tipos de entradas, saídas e intervenções, além de destacar o risco de dependência excessiva de informações limitadas.
This figure compares psychiatry and machine learning engineering as disciplines that process input data to generate outputs, guided by knowledge of their respective systems. On the left, psychiatry integrates therapy data, medical records, patient history, and genetic screening to produce diagnoses and possible treatments, based on scientific understanding of the brain from neurobiology, genetics, and molecular biology. Interventions involve combining psychological techniques with pharmacological treatments. On the right, machine learning engineers use input data to produce inference, regression, prediction, and classification, supported by complete knowledge of AI architectures and algorithms. Interventions include adjusting rewiring, architecture, weights, biases, hyperparameters, and data conditioning. At the bottom, the risk highlights the danger of over-relying on limited or poor-quality input data without fully understanding the underlying model, which can lead to flawed outcomes in both domains.

Reflections on Love, Suffering, and the Impossible

The dialogue also addresses existential questions, such as loving someone who manifests resentment and hatred due to a psychiatric disorder. Love, in this context, requires awareness and boundaries to preserve the caregiver’s integrity, recognizing that the other’s suffering cannot be healed unilaterally.

Problems at the “threshold of the impossible” demand approaches that go beyond linear logic, including radical listening, realistic mapping, the collapse of previous models, lucid imagination, and the courage to experiment with ethical and innovative solutions. This poetic perspective emphasizes the need to give shape to what has not yet been born, transforming the impossible into the inevitable.

Conclusion

Artificial intelligence, despite its advances, does not replace psychology as a fundamental human practice, which relies on empathy, bonding, and subjective experience. Skepticism toward contemporary psychological methods is justified by practical and ethical limitations, but psychology retains its value by addressing dimensions that other sciences have yet to reach. Critical reflection and open dialogue are essential to improving psychological practice and ethically and complementarily integrating the resources of AI.

Millions of Brazilians already undergo therapy through AI — and why this is a problem” by Camila Brandalise, São Paulo, 02/07/2025 05h30 (in Portuguese)

Copyright 2025 AI-Talks.org