Artificial intelligence is impacting the field

[This article is part of the 2025 Trends Report]

Artificial intelligence (AI), which is driving what some call the fourth industrial revolution, has been a harbinger of change. Its power and potential have wrought excitement and fear across nearly every sector of society, from finance and transportation to education and health care.

“Because AI has surpassed the limits of pattern recognition that individuals can do, naturally it becomes impressive to us when generative AI tools work well,” said Ericka Rovira, PhD, a professor of engineering psychology at the United States Military Academy at West Point who studies human-technology interaction. “But as psychologists, because our work impacts people’s lives in such serious ways, we cannot trade magic for the explainable.”

Far from observing the AI revolution from the sidelines, psychologists are rolling up their sleeves to develop, examine, and integrate AI tools—both into the discipline and across society at large. APA’s policy statement “Artificial Intelligence and the Field of Psychology (PDF, 61KB)opens in new window,” released in August 2024, outlines how psychologists can shape AI and how the new technologies are changing the field.

In clinical practice, psychologists are exploring how large language models (a type of generative AI focused on natural language processing) can increase efficiency, expand access to mental health support, and even deepen the therapeutic relationship. On the research front, AI tools are accelerating studies across psychology while generating new courses of inquiry in the human-technology interaction field. And psychologists increasingly have a seat at the table in broader conversations about ethical data use and product development, as well as the role AI should have in our health, work, and relationships.

“As psychologists, we have the opportunity—the responsibility, even—to guide and shape the future of mental health,” said Jessica Jackson, PhD, a Houston-based licensed psychologist, chair of the APA Mental Health Technology Advisory Committee, and a member of the new U.S. Food and Drug Administration Digital Health Advisory Committee. “We do not have to observe this process of development from the sidelines.”

Bringing AI into the clinic can be a daunting task—one that requires technological know-how and ethical deliberation—but experts say it’s time to do so. One helpful tool for clinicians is the “Companion Checklist: Evaluation of an AI-Enabled Clinical or Administrative Tool (PDF, 59KB)opens in new window” created by APA’s Office of Health Care Innovation.

“The time is now to start integrating AI,” said David Luxton, PhD, a clinical psychologist and an affiliate professor at the University of Washington’s School of Medicine, who is writing an APA guidebook on integrating AI into practice, due out in 2026. “Behavioral health professionals need to be competent on the boundaries and risks of AI but also on how it can benefit their practice.”

The forthcoming guidebook explores how AI can help with everything from administrative and back-office tasks to assessment, direct care, and patient self-care (for specifics, see the Trends article on how tech is reducing burnout and increasing access to care). It is intended as a practical manual to give psychologists and other providers of mental health care the knowledge they need to get started.

One tool that uses large language models and may soon become standard: chatbots that assess for depression, suicide risk, and other issues in a conversational and empathetic way. Luxton is one of several researchers developing such a chatbot, which he said could offer advantages over the standard nine-item Patient Health Questionnaire (PHQ-9), which can feel redundant for patients over time, affecting the accuracy of their responses.

“What we hope to assess is: How do both clinicians and clients feel about using an automated interactive conversational tool? Could these systems actually be more accurate and reliable than traditional screening tools and assessments?” Luxton said.

More and more therapists are also using AI to transcribe and analyze their session notes. While it can improve efficiency, psychologists should also consider the very real privacy concerns with storing patient data in the cloud, said Margaret Morris, PhD, a clinical psychologist in private practice and an affiliate faculty member at the University of Washington who studies digital mental health interventions. For example, patients who live in states where abortion is banned might hesitate to discuss the procedure during a session.

“If you as a patient don’t feel your conversations with a therapist are really your own, I think the whole pursuit changes,” Morris said.

Clinicians using such tools should explore whether patient data must be stored in the cloud. Morris encourages clinicians to ask questions such as:

• Who has access to the data?
• Who will have access to it in the future?
• Is it used to train an AI model?
• Can patients later retract their data?

It’s also important to consider how switching to automated transcription may change the workflow and risk shortchanging patients, Morris said. How can providers continue to do the critical analysis and formulation they previously did while writing notes?

Promising advances include opportunities to run AI transcription software on powerful local computers, using open-source large language models, such as Llama from Meta, tuned for mental health. Seeing clinicians opt for that method is promising, Morris said, because it shows a sense of agency in using tools to serve a specific purpose rather than simply adopting a popular new product.

“If you can put the therapeutic relationship first and then use the technology in service of that relationship, it’s possible to get some value from these tools—and not be the product yourself,” she said.

Morris and a team at the University of Washington, led by graduate student Zachary Englhardt and computer scientist Vikram Iyer, PhD, are also studying how generative AI could help therapists analyze data from patients’ daily lives, for example, from wearable devices. In a study where clinicians interacted with ChatGPT around its analysis of wearable data, they envisioned using such data to foster coinvestigation with patients. Clinicians wanted to query the AI collaboratively with patients, with the patients deciding what data to share and how to interpret patterns (Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, Vol. 8, No. 2, 2024opens in new window).

For example, the meaning of activity and phone usage metrics vary from one person to the next. A patient who experiences high anxiety at the start of new romantic relationships might track the number of times they unlock their phone screen as an indicator of their shifting emotional states.

“What I love about this is that it’s putting AI in the service of this human relationship, the therapeutic alliance that we know to be so important for therapy,” Morris said.

Large language models can also analyze session transcripts to aid in supervision and training, including by highlighting instances where a trainee deferred from a treatment protocol, then providing real-time feedback. For example, mpathic and Lyssn are AI models trained by mental health care providers to recognize qualities such as empathy during conversations, including reflective listening, affirmations, and open-ended questions. Such technology can help busy supervisors target key moments for feedback—or the AI itself can offer trainees an alternative way to phrase a response, for instance by rephrasing a directive as a question.

While products on the market today are relatively simplistic—for instance, scoring adherence to a cognitive behavioral therapy protocol—Morris is hopeful that future tools will provide insights on psychodynamic and other insight-based approaches. Other training opportunities include simulations that afford therapists-in-training a safe place to explore various approaches.

“For example, imagine a trainee interacting with a simulated patient—one that has realistic facial expressions and body language and responds in predictable ways,” said Tara Behrend, PhD, a professor at Michigan State University’s School of Human Resources and Labor Relations who studies human-technology interaction.

It’s well known that when generative AI is trained on incomplete data, it can draw incorrect conclusions that perpetuate bias in society. But AI can also help improve health equity if used strategically, said psychologist Adrian Aguilera, PhD, an associate professor at the University of California, Berkeley, and the University of California, San Francisco, who studies how technology can reduce health disparities.

For instance, psychologists might prefer to build culturally adapted mental health interventions from the ground up, but resources to do so are often limited. With much less effort, large language models can help providers tailor an intervention for a specific population—not just translating a protocol into Spanish, for example, but adapting it for Spanish speakers from a Latin American country, complete with relevant cultural metaphors.

“That becomes a deeper adaptation that certainly needs to be reviewed, updated, and refined by humans, but it allows us to create these tailored interventions with much less effort than needed in the past,” Aguilera said.

At mpathic, researchers are also exploring how AI can be used to detect high versus low cultural attunement during therapy sessions, then provide real-time feedback when needed.

AI has already been a major boon for the scientific research process, including through generating R code, a programming language often used for statistical analysis, and to create experimental stimuli such as text or video vignettes. But generative tools such as ChatGPT also invent false references (often referred to as AI hallucinations) and may even fabricate data, Rovira said.

Another open question is to what degree AI can or should be used for manuscript writing. Doing so could save time, accelerate scientific discovery, and make science more accessible to nonnative English speakers, Rovira said, but it is not without its risks. Journals may require disclosure when AI is used for manuscript writing, but psychology as a field should also consider what degree of generative AI use is acceptable at any stage of the research process (Tay, A., Nature Index, August 17, 2021opens in new window).

“This explosion in manuscripts could break the peer review system as we know it,” Rovira said. “So, what is our role as a society of professionals in deciding the limits of AI for these tasks?”

Rovira and other human factors researchers are also asking new questions about human-technology interaction that could prove critical to using AI safely. For example, past research shows that people with better working memory and attentional control tend to be less easily duped when automation fails (Proceedings of the Human Factors and Ergonomics Society Annual Meeting, Vol. 63, No. 1, 2019opens in new window). Does that trend hold true for AI agents, such as when a ChatGPT provides bogus references? In another line of research, Rovira and her colleagues are exploring whether people are more likely to trust AI agents when they use good social etiquette.

“People need to stay engaged in the decision-making, which can be hard when these systems operate incredibly quickly and smoothly,” she said. Human factors psychologists can help structure human-AI interactions in a way what keeps people in the loop to provide context for safety reasons.

Before using AI tools in mental health care and beyond, psychologists have a responsibility to look past marketing language and evaluate each product with a critical eye, Behrend said. They can also help shape ethical development principles set by government regulators and oversight organizations, such as collecting the least amount of data necessary and providing transparency about exactly how it will be used, Aguilera said.

“All of this is moving so rapidly, so that regulation needs to be an iterative process,” he said. “But having basic principles can at least provide some guard rails.”

It would be foolish to ignore the dangers AI can pose when used with bad intent. Social media algorithms already manipulate us in a very personalized way, Luxton said, and similar data can be used for reputational harm and psychological manipulation. Disparities in who has access to and control of AI are also likely to cause problems across society, he said. Far from a deterrent, he hopes those truths can galvanize psychologists to get involved.

“Change brings anxiety, and being critical is extremely important,” Aguilera said. “But the AI revolution is happening, so let’s engage and try to steer things in a direction that’s for better rather than for worse.”