Use of Artificial Intelligence in Mental Healthcare, Health Psychology, and Related Research: A Narrative Review to Address Challenges and Opportunities

ABSTRACT

Background and Aims

Artificial intelligence (AI) is the process by which a machine learns from a pool of data and can then respond to questions beyond that data set. AI has been implemented in many other fields, and a lot of interest is placed on how AI will advance patient–provider interactions. One of the greatest areas of need is in the field of psychology, as there are comparatively few providers when looking at the inordinate number of patients requiring counseling. System‐on‐chip technology, its usage in emotion‐predicting AI models, and risks of nonclinical chatbots. This review examines the integration of AI in psychology, emphasizing emotion prediction through system‐on‐chip technologies. It also highlights key concerns surrounding nonclinical chatbot use, including data bias, privacy risks, and the urgent need for responsible and ethical AI deployment.

Methods

An extensive literature search (Scopus and Web of Science) was performed in PubMed using keywords “artificial intelligence,” “psychology,” “mental health,” “cognitive psychology,” “psychological diagnosis,” and “psychology apps,” which yielded 112 articles. A total of 76 articles were excluded because of a misalignment with the focus on AI in psychology and mental care. We extracted relevant information from the remaining 36 articles, of which 6 were excluded as they did not meet the predefined criteria of AI application in this context.

Results

Although AI models have demonstrated potential in prediction, decision, and patient support, there are no FDA‐approved diagnostic tools in clinical psychology. System‐on‐chip (SOC) models show evidence for high accuracy in real‐time emotion recognition, and chatbots support patients with home symptom monitoring. However, AI in mental health is limited by biased or incomplete data sets, raising concerns about reliability. Privacy risks, particularly with nonclinical chatbots like ChatGPT, further complicate implementation. Regulatory and ethical barriers remain unresolved, and despite strong research promise, clinical adoption is still limited. These challenges highlight the need for cautious, evidence‐based integration.

Conclusions

AI has the potential to assist clinicians and researchers in psychology. Careful consideration is needed of ethical, regulatory, and methodological concerns with new AI‐based models and tools. With use rooted in evidence, integration of AI tools would enhance provider efficiency and access.

1. Background

Artificial Intelligence (AI) has seen a surge in interest due to its flexibility, being utilized in a variety of fields [1]. AI has been used in healthcare for over 50 years, with the implementation of an antimicrobial AI, “MYCIN,” in the early 1970s [2]. MYCIN could recommend antibiotics correctly dosed for the patient's body weight based on patient information [2]. AI in infectious disease has only evolved and improved. MYCIN was later developed into “INTERNIST‐1” an improved system with more parameters and information [1]. Other fields of medicine have also seen the implementation of AI. Cardiology had the first US Food and Drug Administration (FDA) approved deep learning (DL) cloud‐based AI software to analyze cardiac magnetic resonance images (MRI), named CardioAI [2]. CardioAI was able to calculate useful values for clinicians, such as ejection fraction [2]. Another area where AI has seen implementation in gastroenterology is with an AI model to predict the severity of esophageal adenocarcinoma [2]. Gastroenterology is an area where AI implementation has been observed to be remarkably effective. In a retrospective study of 150 patients with 45 clinical variables, the accuracy of diagnosis of GERD by an artificial neural network (ANN) was 100% [3]. Another ANN was able to predict the mortality of a nonvariceal upper gastrointestinal (GI) hemorrhage with a 96.8% accuracy [4]. AI's utility in gastroenterology also delves into endoscopy, where it has shown a statistically significant increase in the adenoma detection rate compared to a standard colonoscopy [5]. GI‐Genius, a computer‐aided diagnosis (CAD) system, is currently being clinically evaluated in the United States for the identification of colorectal polyps [2]. GI‐Genius has a reported sensitivity per lesion of 99.7% and can detect the lesion before the endoscopist in 82% of cases [2].

The purpose of this review is to evaluate the current and potential applications of AI in psychology, mental healthcare, and research. It highlights the significant role of AI in reducing provider burden and enhancing therapeutic outcomes with automation and provider tools. Furthermore, the review will showcase the role of AI in offering deep insights into research through behavioral analysis and its effectiveness across various medical fields.

AI is effective in many health sectors, and psychology is no exception. As psychology is one field lacking in providers, AI could take the burden off the existing providers for easily automated tasks, such as automated note‐taking, administrative tasks, and therapeutic chatbots for patients [6]. A therapeutic chatbot that works well can fill in for the providers so that every patient has access to some level of therapy while leaving the providers open for those who need a more human approach. Likewise, an AI tool that can annotate notes for the provider, as well as handle organization, filing, and schedule management, would leave them with more time to connect with and treat patients. One example of a therapeutic chatbot is the WYSA AI, which is used in the management of chronic pain and anxiety [6]. WYSA has received FDA breakthrough device approval to be used alone or with a provider for this management [6]. Providers mainly use WYSA as a way to monitor a patient's condition in between therapy sessions [6]. WYSA is showing promise, Inkster and colleagues found that a participant's amount of WYSA use was directly correlated with their mood [7]. Although this study is not conclusive, the researchers stated that further work is required to validate these findings in much larger samples and across longer periods [7]. Beyond clinical psychology, AI also has an opportunity to help researchers analyze and measure human behavior. An AI would be able to monitor certain metrics, such as the usage time of certain programs and points of focus with eye tracking. The application of AI in therapeutic psychology, both in research and practice, is growing as clinicians and researchers realize its utility.

2. Methods

A comprehensive literature search restricted to English‐language articles indexed in Scopus and Web of Science was performed in the PubMed database using keywords: “artificial intelligence” AND “psychological” OR “mental health” OR “cognitive psychology” OR “psychological diagnosis” OR “mental health therapy” OR “psychology apps.” Initially, we got 112 articles, and then we screened the titles and abstracts of all the articles to determine relevance. Among the articles, we excluded 76 articles since they did not fit with our objectives on finding the clear role of AI in psychology and mental healthcare. From the remaining 36 articles, we reviewed the full‐text reviews. Among 36 articles, we did not consider 6 articles because of not align with the predefined criteria for AI application in this context. Therefore, throughout this systematic approach, the final selection of articles ensured both relevance and methodological soundness, providing a strong groundwork for the study.

3. AI in Psychology and Mental Healthcare

3.1. AI for Psychological Diagnosis

The large increase in incidences of mental health diseases is affecting many populations around the world negatively. Many of these illnesses, namely major depressive disorder, schizophrenia, and bipolar disorder, although being mildly homogeneous in‐group, still display a great heterogeneity even when comparing two patients with the same disease. This characteristic of mental illnesses is apparent in the revisions of the Diagnostic and Statistical Manual of Mental Disorders (DSM), which was most recently revised in June 2022 [8]. Using AI in the diagnosis of mental disorders would be useful given the currently heterogeneous state of those diseases. AI is utilized for the early identification of mental health disorders, the creation of personalized treatment plans, and the effective deployment of virtual therapists [9]. AI could find patterns, or at the very least take in a large amount of data, both old and new, to find a statistical relationship between symptoms and disease at a faster rate and higher accuracy than a clinician (Figure 1). All these being said, there are currently no FDA‐approved CADs for the diagnosis of mental disorders.

Figure 1.

Use of artificial intelligence in psychology and mental healthcare, and research for positive outcomes. Figure 1 demonstrates how AI tools (red) link to improved mental health outcomes, while the blue lines represent connections between research areas, demonstrating how combining patient‐ and research‐focused AI can enhance therapeutic results.

3.2. AI in Cognitive Psychology

One area of psychology that is changing this, though, is cognitive psychology. Two typical models used in cognitive psychology, face attraction and affective computing, are being used to advance AI and the scope of its implementation (Figure 1) [10]. Face attraction models examine the perception and evaluation of facial attractiveness. These evaluations and perceptions are fundamental to attention, memory, and social judgment. Facial symmetry and expression mold responses in cognition and emotion [11]. Important elements of mental healthcare and health psychology, such as impressions, connections, and self‐image, are all understood through the lens of face attraction models [11]. Acknowledging how facial attractiveness can impact provider–patient relationships through feelings of anxiety or body image issues is essential for developing strategies to reach those with low self‐esteem or who are isolated.

Another model of affective computing, otherwise known as emotions, sentiments, or feelings, has been studied and analyzed to help AIs better understand human emotion [10]. This is necessary for the broader implementation of AI as emotions play a large role in decision‐making, and in how one responds to information. One such affective computing AI model, an AI system on chip (AI SoC) design, was able to predict what emotion a test subject was feeling 77.41% of the time based on physiological and multimodal sensors [10]. Unlike other models, the AI SoC was able to make these predictions in real‐time using an electroencephalogram (EEG), electrocardiogram (ECG), and photoplethysmogram (PPG) [12]. Regardless of significant accuracy, misclassification remains a concern. These findings have questionable generalizability due to the controlled test setting. In addition, the reliance on monitors such as EEG, ECG, and PPG limits their scalability and practicality.

3.3. AI in Mental Health Therapy

Although there is still much work that needs to be done in the advancement of AI, it has seen some implementation in the clinic to assist healthcare providers. A prime example of its use is its potential in the diagnosis, treatment, and management of mental health conditions. First off, AI can facilitate understanding of mental health disease progression. Recent publications revealed the release of advanced support systems to monitor patients with psychotic disorders using AI in smartwatch devices [12]. One publication that specifically addresses this proposes a system called “E‐Prevention,” in which it records biometric data using a smartwatch, video recordings of patients during clinical visits, and the ability to predict relapse and prediction [12]. Once again, addressing the shortage of healthcare practitioners, it is often difficult for patients to obtain the right care. So, if AI can help in the analysis of individuals the way “E‐Prevention” can, it can potentially lead to a cost‐friendly and efficient method. In other cases, a psychologist may not necessarily be available to everybody. So, an AI robot might be programmed in a way that can resemble the interaction of a psychologist or psychotherapist. This is not just an elementary idea, but something that has been discussed frequently in research articles [13], which not only will make up for the shortage of therapists, but also create a better environment for those in need of help (Figure 1). Despite promising applications, AI in mental health remains limited by validation, wearable and recording devices, and uncertain effectiveness.

3.4. AI‐Based Decision Support Systems for Mental Disorders

As previously mentioned, AI can provide support in catching early symptoms of psychotic disorders and can give potential solutions to the cause [12]. This alone can provide major assistance to patients, making them more aware of their issues. In the future, this can become an astronomical tool not only for patients but also for mental health providers. Before the introduction of AI, as well as currently, mental health professionals have difficulty diagnosing patients accurately using diagnostic tools. In a previous study including 217 psychiatric consultations, it was found that the most accurate diagnosis was for cognitive disorders at 60% [14]. For depression, anxiety, and psychosis, they were at 50%, 46%, and 0%, respectively [14]. With misdiagnosis rates being so high, this is an issue that needs to be addressed. Through several studies regarding AI in neuropsychology, it was found that AI can diagnose mental disorders with a 68%–100% accuracy [15]. This is a significant accuracy increase, and this can become a great tool for therapists and other mental health professionals to at least use as a point of reference (Figure 1). These results, however, are based on a single practice setting, which limits their generalizability to other settings and populations. Likewise, the study design limited researchers' ability to assess the initial diagnostic impressions of the referring provider. The reliability of the findings is also called into question as the psychiatric diagnoses were not validated through a structured clinical interview [15].

3.5. AI‐Based Psychology Apps

In the United States from 2019 to 2020, it was estimated that 20.78% of adults were experiencing mental illness [16]. In quantity, that is about 50 million Americans, and 54.7% of that population does not receive any treatment [16]. These people may want to seek therapy but may be afraid of being judged, afraid of speaking their feelings, the cost, and several other factors. Through the addition of AI‐based psychology apps, these factors may be wiped out completely. Speaking with an AI designed to be judgment‐free and with similar knowledge to a professional may be extremely reassuring. An app with the name Woebot, which was designed as a web‐based cognitive behavioral therapy, has shown to have such results in past years [16]. Providing the data, it was shown to significantly reduce the study group's symptoms of depression as compared to a control group [17]. Lack of statistical synthesis and the exclusion of performance metrics, compounded by the overlapping primary studies across reviews, may have introduced a duplication error in the reported classifier performance ranges [17].

4. AI in Psychological and Mental Health Research

4.1. AI Facilitates Understanding of Disease Progression

Looking at different cognitive cases, AI can be used to differentiate impairment from normal states according to an umbrella review of AI‐driven technologies [15]. For example, it was found that AI had a 60%–98% accuracy in differentiating Mild Cognitive Impairment from healthy controls (HC) [15]. In more specific cases, AI differentiated Schizophrenia (SCZ) from HC with accuracy ranging from 61% to 98.6% [15], suggesting AI is capable of identifying neurological disease progression with great accuracy, handling complex data sets and utilizing them in cases of mental health disorders, and ensuring future possibilities that represent an emerging area of research. If institutions can identify mental health illnesses due to this machine‐learning software, it will not be long until individuals who are too afraid to step up for their mental health are helped. According to Mental Health America, over half of adults with mental illness (about 28 million people) do not seek help in the United States [16]. It is even more dreadful if they live with it over extended periods (Figure 1). Generalizability is limited with this study as it is retrospective and single‐site, with a lack of structured diagnostic validation.

4.2. AI Helps in the New Treatment of Mental Health Disorders

As outlined previously, there are applications such as Woebot that can reduce a study group's depression with just the act of communicating with them [17]. Since then, there have been many advancements in the field of psychology using online resources. One such advancement is Virtual Reality (VR) technology in treating anxiety and other mental disorders [18]. VR is the use of a computer system to simulate a 3‐dimensional world in the eyes of the viewer. A randomized controlled trial in 91 schizophrenia patients showed that receiving therapy in VR components resulted in an improvement in conversation skills and assertion compared to traditional methods of treatment [18]. If this were used with the combination of AI trained to speak as a mental health professional, it may cause an increased interest in therapy. Additionally, social barriers for individuals with mental health disorders can be significantly lowered, resulting in greater confidence for them in general. The VR study was limited by its sample size, the time frame for its intervention, as well as postintervention follow‐up, which weakens its analysis of the sustained effects and mediation. In addition, the engagement and dose–response data were unavailable, further limiting its application. This is just one of the emerging applications of AI for the treatment of mental health disorders (Figure 1).

4.3. AI in Cognitive Modeling

Regarding cognitive modeling, AI has been implemented in various professions [19]. Intelligent Cognitive Assistant (ICA) technology is used to imitate human behavior, and due to its capability of using individually tailored language, they are considered to be a crucial tool in digital mental health services [19]. ICAs have been successful in guiding people's thought processes and actions through cognitive analysis and are used to support those with stress, anxiety, and depression (SAD) [19]. ICAs are also used for attitude and behavioral change therapy (ABC) and are unique in the fact that the patients tend to reveal personal information more freely, establishing a better relationship and allowing the ICAs to learn and achieve ABC [19]. With the capabilities that ICAs have with SAD and ABC, they are the ideal vessel for mental health help and pave the way for new technology in the field of mental health (Figure 1) [19]. Evidence for ICAs in managing SAD remains limited. Small sample sizes, short‐term studies, and lack of longitudinal analysis raise questions about the strength of the available evidence [19].

4.4. AI Methodology in Biomarker Detection

Biomarkers have been defined into various categories and different areas of application by the US FDA [20]. In more recent years, mental health orders have become a new avenue for biomarker discovery, but currently, only a few tests have been approved for clinical use [20]. However, further identification of biomarkers for mental health could lead to the development of an advanced clinical description support system (CDSS) powered by AI, resulting in advanced personalized medicine [20]. Currently, Molecular Diagnostic (MolDx) panels, used for clinical cancer care, are becoming more common and coupled with genetic counseling [19]. In the mental health field, the GeneSight test is used to assess genetic variants for the treatment of psychiatric disorders, and such advances lead to the creation of diagnostic panels for mental health disorders [19]. Those panels can then be coupled with the CDSS alongside patient history to offer medications and treatments, which can then be used by clinicians to create a treatment plan with minimal adverse effects [20]. With the use of CDSS, personalized medicine for mental health disorders would be driven forward (Figure 1) [20]. Although biomarkers are promising for use in psychiatry, few are approved for clinical use. Many lack large‐scale validation, making them unsuitable for clinicians to use. Compared to other fields of medicine, MolDx panels in psychiatry are underdeveloped and face challenges in standardization and integration. Tools like GeneSight may not capture the relation between genetic, environmental, and comorbid conditions that impact psychiatric disorders.

4.5. AI in Big Data Analytics for Mental Health

As aforementioned in Section 3.4, AI can be used for mental disorders, assisting in detecting symptoms and providing treatment plans [13]. In terms of larger analysis for mental illness, AI can perform actions that human healthcare professionals cannot, such as analyzing patient information across various sources—medical records and medical devices, for instance—and identifying the patterns in the data to develop a diagnosis [13]. AI has been used for mental healthcare analysis for ‐three main reasons: digital phenotyping, natural language processing, and chatbots [13]. Digital phenotyping is using digital data to measure and monitor a patient's mental health by examining material from personal records like social media and medical records, and using that to detect behavioral changes that can be associated with mental health issues [13]. Natural language processing algorithms track and analyze patterns used in conversations, including chats and emails, and detect changes and how they correlate with the progression of a patient's mental health, whether it is progressing or regressing [13]. Chatbots can be used to inquire about a patient's mental health, similar to how a mental healthcare professional would, questioning their mood, stress levels, sleep, and so on, and then analyzing it and suggesting therapies or seeking out medical intervention [13]. It is already seen in smart glucose trackers, where when the levels fluctuate by a significant margin, the medical term is alerted [13]. There is still uncertainty about whether AI will improve mental healthcare; however, it cannot be denied that it helps with the analysis and detection of mental disorders (Figure 1) [13]. Despite promising data, the effectiveness of AI in providing mental healthcare is unknown. Many tools, such as digital phenotyping and natural language processing, lack validation. Reliance on personal data, such as social media, raises concerns about data set bias. Generalizability should also be considered in diverse populations [13].

5. Methods of Protection to Ensure Patient Privacy, Transparency, and Equity

Psychologists may be among the most qualified to determine methods of patient protection, with training on various research methodologies, ethical treatment of participants, psychological impact, and more. They can help companies understand the values, motivations, expectations, and fears of diverse groups that might be impacted by new technologies. Moreover, they can help recruit participants with rigor based on factors such as gender, ancestry, age, personality, years of work experience, privacy views, neurodiversity, and more. Official policies have not yet been established by the American Psychological Association (APA), but there is work in progress for monitored use. Intentional and responsible implementation is necessary for the effective and beneficial implementation of AI. The absence of recommendations for the implementation of AI by regulatory bodies, such as the APA, leaves many without guidance.

6. The Misuse of Nonclinical Chatbots and Methods to Overcome

A major risk with the implementation and advancement of AI is the misuse of nonclinical, publicly available chatbots. Misuse in this context is the use of a nonclinical AI chatbot to supplement the role clinical psychologists play in patient care. ChatGPT, a new browser‐based AI service, has seen widespread attention due to its advanced problem‐solving and conversational capabilities. There is a tangible risk in using ChatGPT in clinical psychology. ChatGPT has not been approved by the FDA for any indication or use and is prone to misinformation as well. When asked to generate 30 abbreviated medical papers, out of 115 references, 47% were fraudulent, 46% were authentic but inaccurate, and 7% were both genuine and accurate [21]. Likewise, an ML AI model such as ChatGPT is completely dependent on the data being supplied to it. Therefore, without careful monitoring and consideration, harmful biases and discrepancies may arise in conversation. Arguably, the biggest concern regarding the misuse of nonclinical chatbots is patient privacy. To use this service, users must first make an account, and from this account, OpenAI collects account, communication, and social media information [22]. Through utilizing the software, log, usage, analytics, and device information are also collected. Although OpenAI must legally comply with data privacy regulations, ChatGPT can share users' data with a third party without first notifying the user [22]. Some strategies to manage the danger of misusing nonclinical chatbots would be public awareness of the risks of using ChatGPT as a supplement to the role of a clinical psychologist. Likewise, increased awareness and implementation of WYSA, an FDA‐approved chatbot, would limit the number of people sharing their information with a nonclinical chatbot, which has no proven efficacy or robust privacy standards.

7. Limitations

AI has a plethora of uses in the mental healthcare system; however, there are limitations, privacy risks, ethical barriers, regulatory considerations, and data bias are all considerations that must be addressed. Despite its promise in research, evidence‐based integration in clinical applications is necessary, as well as a robust method of implementation. Conversational AI can engage in conversations with patients, gather information, and provide interventions based on the evidence gathered [23]. However, regardless of the practicality, conversational AI is unlikely to replace human therapists in the future due to the challenge of planning and executing collaborative tasks between AI and human healthcare professionals [23]. Conversational AI can put both patients and providers at risk of unintentional consequences; however, there is also the possibility of how it can be tailored to a patient's specific needs and provide superior care [23]. Conversational AI can be used to supplement the lack of clinicians in the mental healthcare field and address one of the current tensions in care, the lack of incentive to engage in meaningful but lengthy conversations [23]. Clinical professionals have no financial incentive to engage in those conversations, so they tend not to [23]. However, AI does not need financial incentives and has nonconsumable time, so it is an effective and attractive alternative [23]. Conversational AI would be suited to addressing barriers to mental healthcare access [23]. Patient privacy, bias, and transparency impair the integration of AI into clinical practice. Likewise, clinician training, change bias, difficulties in integration, regulatory bodies, and high costs also significantly restrain utilization [9]. There is a significant gap between the utility of AI in research and the utility of AI in clinical practice. Critical perspectives on the translational challenges are important to the successful implementation and impact of AI in mental healthcare.

8. Conclusions

AI can increase the productivity of existing mental healthcare providers through automated note‐taking and chatbots. There are also methods that AI can use on the go in terms of smartwatches and mobile phones, which can be used to improve patient conditions by managing and monitoring their mental health. When implemented in combination with providers, AI can improve the assessment of lifestyle characteristics. Although acknowledgment of AI's limitations in this area is necessary.

Author Contributions

Md. Ashrafur Rahman: conceptualization, data curation, writing – original draft. Evangelos Victoros: conceptualization, data curation, writing – original draft. Rob Davis: conceptualization, data curation, writing – original draft. Tariq Duaa: conceptualization, data curation, writing – original draft. Yeasna Shanjana: conceptualization, supervision, visualization, writing – review and editing. Md. Rabiul Islam: conceptualization, supervision, visualization, writing – review and editing.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Transparency Statement

The lead author Md. Ashrafur Rahman affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Rahman M. A., Victoros E., Davis R., Duaa T., Shanjana Y., and Islam M. R., “Use of Artificial Intelligence in Mental Healthcare, Health Psychology, and Related Research: A Narrative Review to Address Challenges and Opportunities,” Health Science Reports 8 (2025): 1–7, 10.1002/hsr2.71595.

Data Availability Statement

Data sharing is not applicable to this article as no data sets were generated or analyzed during the current study.