Artificial intelligence in neurosciences and mental health

Introduction

Artificial intelligence (AI) as a concept was born in 1950, with the publication of the article “Computing Machines and Intelligence” by Alan Turing. The “father of artificial intelligence” wondered if a computing system could think like a person thinks, later abandoning the concept due to the limitations of neuroscience at the time and the complexity of neurocognitive functions. The term artificial intelligence was proposed in 1956 by John McCarthy, who defined it as “the science and engineering of making intelligent machines”⁽¹⁾.

Artificial intelligence represents the simulation of human intelligence by a computer system, with the ultimate goal of reaching the human level of decision-making and problem-solving⁽²⁾. There are currently three types of AI in use: natural language processing (NLP), machine learning (ML), and deep learning (DL)⁽¹⁾.

The process of simulating human intelligence is most frequently based on machine learning, a concept derived from informatics, statistics and linguistics, through the use of structured, unstructured and semi-structured big databases. By being exposed to an enormous amount of information, AI analyzes these structures to identify patterns, create logical (organized) content, and form correlations for data prediction^(2,3).

The COVID-19 pandemic has created a spike in the prevalence of mental disorders globally (especially affective and anxiety disorders). Even today, these conditions cause a significant level of disability, as measured by the disease burden index (DALY)⁽⁴⁾. The need to implement auxiliary methods to support the public health system is thus observed, an attractive variant being represented by the augmentation of the medical act through AI.

If we refer to psychiatry, artificial intelligence involves the use of computer techniques and algorithms for the diagnosis, prevention and treatment of mental disorders. Lately we have seen an increase in the use of this cutting-edge technology in the field of mental health⁽⁵⁾.

Objectives

This article critically assesses the advantages, disadvantages, effectiveness and use of different forms of artificial intelligence in mental disorders and their prevention.

Materials and method

Articles, mainly systematic reviews and reviews of different forms of AI used in mental health, published between 2019-2024, from the PubMed and ResearchGate databases, were used to conduct the study. The keywords included: “artificial intelligence”, “gpt”, “neuroscience”, “psychiatry”, “mental disorders”. A total of 15 articles relevant to the requirement were selected.

Results

The use of artificial intelligence in psychiatry has increased dramatically in recent years. Some experts are saying that it could be the most disruptive technology, compared with the invention of electricity or internet. This has arisen as a result of the need to cope with an ever-increasing number of patients and to increase the availability of mental health services to the general public. Access to AI health services can be done through the Internet, mobile applications or video games⁽⁶⁾.

Artificial intelligence has utility in medical data processing, disease classification, detection of relevant biomarkers and early stages of various conditions, real-time monitoring of evolution, and discovery of new therapies and interventions (both in psychiatry and other medical specialties)⁽⁴⁾.

In the field of neuroscience and mental health, AI can be found in various forms (Table 1).
 

I.  Neuroscience deals with the study of the structures and functions of the nervous system. Starting from the biological organization of neural networks, information technology engineers have developed complex network architectures used in various applications. Artificial intelligence is capable of developing complex multivariate models for identifying the relationship between some specific neural activity and the associated cognitive function. This means that AI has the potential to decipher how a neural network (circuit) determines a cognitive response (ex.: attention, memory, socialization, interpersonal interactions). Using the MVPA (multivariate pattern analysis) technique, AI was able to demonstrate that the circuits of the prefrontal cortex are responsible for the process of generating working memory (in contrast to the previous theory that the hippocampus is the center responsible for this function)⁽²⁾.

In the last two decades, attempts have been made to integrate neuroimaging and AI studies in order to classify patients with mental disorders. It was thus observed that by means of functional MRI interpreted by artificial intelligence, the rate of correct positive diagnosis in patients with schizophrenia is 85.5%⁽⁶⁾. Using neuroanatomical and neurofunctional data, AI has been shown to exhibit a high level of reliability in the diagnosis of psychiatric disorders. Computational models combining deep learning methods (with mono- or bidimensional neural networks) and EEG have been used to detect depression, with a success rate of over 90%. If we refer to psychoses, machine learning algorithms have begun to identify language disorders and conceptual disorganization, suggestive elements of these conditions. The correct positive diagnosis was made in over 70% of cases⁽⁷⁾.

II. Artificial intelligence applications can be used for the diagnosis of psychiatric disorders, tracking symptoms and the evolution of the condition, prediction of exacerbations, along with psychoeducation⁽⁶⁾.

In 1966, the first virtual psychotherapist (ELIZA) capable of emulating conversational capabilities was developed, with the aim of letting the patient perform the cognitive interpretation of events. This experiment formed the basis of NLP (natural language processing) and quickly led to the development of AI communication⁽⁶⁾.

Virtual psychotherapy can be done through apps (Tess) and chatbots (Woebot, Sara, Wysa) that use text messaging services⁽⁷⁾.

For example, Tess (based on NLP) mimics conversations with a psychologist and teaches patients coping strategies for stressful moments, with good efficacy in relieving anxiety and depression^(6,7).

Woebot is another automated conversational application that emulates the process of cognitive behavioral therapy. It helps users identify their dysfunctional cognitions and emotions and how to cope with stress, while reducing anxiety and depressive symptoms^(6,7).

MindLAMP is an application that collects information about patients’ experiences in order to establish the prediction rate of remission of disorders. BiAffect aims to predict manic and depressive episodes in patients with bipolar affective disorder (by analyzing typing dynamics, errors or pauses)⁽⁶⁾.

Replika is a smartphone application that allows users to discuss themselves with an avatar, confidentially and non-discriminatingly, thereby discovering positive qualities and insight into their own personality⁽⁶⁾.

III. GPT (generative pre-trained transformer) is a multimodal language model created by OpenAI that is based on deep learning (a subset of machine learning). Chatbot programs such as now famous ChatGPT have appeared using this language model. This can be support in completing medical records, augmenting doctor-patient communication, finishing academic papers, preparing scientific presentations, and analyzing databases for research.

For integration into routine psychiatric work, however, improvements to the GPT language are needed, such as the introduction of empathy functions, emotional intelligence, assessment of personality traits, and detection of psychiatric symptoms and signs (especially emergency ones). Empathy is based on the theory of mind. GPT-4 shows theory of mind abilities, which reflects the possibility that the empathy function will be implemented in the future⁽⁸⁾.

Mental health professionals must verify and supervise the content entered by ChatGPT as it does not understand the complex requirements, the data are not always correct, and it lacks clinical experience or judgment (omits relevant clinical information).

A recent study highlighted the fact that ChatGPT has a relatively good ability to analyze affective-emotional functions, average for assessing suicidal risk, and poor for personality traits.

Given the level of stress and burnout in psychiatry (using the CBI scale the prevalence of burnout has been observed to be close to 50%), ChatGPT and other GPT-type programs can reduce this burden by streamlining the way observation sheets are completed and other medical documents (for example, by using the speech-to-text dictation function)⁽⁸⁾.

IV. Avatar therapy for patients with schizophrenia is an area of ​​interest. This therapy shows utility in reducing the severity, frequency and distress associated with auditory hallucinations in psychosis. The psychiatrist assists the patient in creating an audio-visual entity (an avatar) that reproduces the characteristics of the psychoproductive content. Through repeated exposure to this avatar, the patient learns coping mechanisms, gains insight, and begins to control hallucinations. Sessions can be recorded, and the patient can view them again at home for exercise and pain relief in the event of an exacerbation. However, study results have not demonstrated a statistically significant effect at this time, so more evidence is needed for avatar therapy⁽⁹⁾.

V. In addition to avatar therapy, virtual reality (VR) assisted therapy can also be used in schizophrenia, which encourages patients to become aware of hallucinations and become critical of them, with the consequent increase in quality of life⁽⁷⁾.

VI. Robotic technology (Nao, Kaspar, and Zeno) can facilitate the development of cognitive, emotional, and social skills in children with autism through imitation, attention-stimulating play, emotional recognition, and conversation⁽¹⁰⁾. Robotic AI therapy (at the intersection of AI and robotics) features unlimited accessibility, no scheduling, no time limit, no breaks, and is nondiscriminatory. Thus, some human limitations can be overcome with a high rate of safety and consistency⁽¹¹⁾.

Kaspar and Nao are able to help children with autism spectrum disorders in developing social skills and facial recognition of emotions, as well as improving eye contact⁽⁶⁾. These robots can also have negative effects (initial fear, dependence on the robot, difficulty switching learned elements to a human subject).

The initial study results are promising, with children with autism having a superior response to robotic therapy compared to human psychotherapists, with improvements in spontaneous language, social skills, and empathy⁽⁷⁾. However, most of the results of these clinical studies must also be confirmed in the environments close to the patients (within the family, the school environment)⁽¹⁰⁾.

VII. Animal robotic therapy (Paro, eBear) has achieved good results in reducing stress, agitation, loneliness, as well as in improving mood and socialization level, being a real support for patients with major neurocognitive disorders and unipolar affective disorders^(6,7).

A 2023 study showed a reduction in levels of anxiety, depression and agitation, decreased use of pharmacotherapy, but no statistically significant effect on sleep. Robotic therapy via Paro reduces apathy and enhances instinct and social adaptability in dementia patients⁽¹²⁾. Another study observed a reduction in the level of pain and the use of analgesics in the context of Paro therapy, but with no significant differences in the level of anxiety, depression or agitation⁽¹³⁾.

VIII. Video games can be effective in delivering psychoeducational information, lowering stress levels, and tracking symptoms with high availability⁽⁶⁾. They are mostly cheap, accessible, and provide a sense of pleasure. The benefits of video game use (with moderate use, without reaching the threshold of addiction) are shown in Table 2⁽¹⁴⁾.
 

Video games are a form through which AI succeeds in connecting with the user and ameliorating the symptoms of two of the most prevalent mental disorders: depression and anxiety (especially by reducing anhedonia and feelings of isolation)⁽¹⁴⁾.

Conclusions

In recent years, artificial intelligence is taking on an increasingly important role in medicine, neuroscience and mental health. Considering the significant impact of mental disorders on the functionality and quality of life of patients, the improvement of these algorithms and computer techniques for the diagnosis, prevention and treatment of psychiatric conditions is constantly sought.

Artificial intelligence works through methods of classification, hypothesis generation, prediction and detection. In order to use AI responsibly, ethical aspects must also be taken into account, as well as the need to implement strict measures to protect patients’ personal data⁽¹⁵⁾. These interventions can become available to a large number of patients under the conditions of a limited number of specialists.

 A significant advantage is represented by the access of people from disadvantaged environments or located at significant distances (e.g., rural areas) to mental health services (for example, through chatbots). By promoting free apps, patients who are uninsured (or whose insurance does not cover mental health services) can also be assessed and helped⁽¹¹⁾. Patients may be reluctant to disclose problems to their doctor, but the process could be eased through an AI interface. Therapies through chatbots, virtual psychotherapists and robotic companions present clear advantages: reducing stigma, creating a comfortable environment, a good cost-effectiveness ratio, increasing accessibility, efficiency and constant availability (lack of breaks)^(6,7). Along with the multiple advantages of artificial intelligence intervention in psychiatry, there are also new problems to which answers must be sought:

• the minimization and methods of risk assessment;
• the lack of guidance on the development of applications and their integration in clinical activity;
• paucity of training doctors in order to use AI as a support;
• ethical and regulatory issues;
• hypothetic replacement of current services by AI (and associated consequences);
• AI oversight (a must, in our opinion);
• protecting patients’ autonomy;
• transparency in the use of algorithms;
• short- and long-term effects of AI use in mental health (including addiction)^(7,11).

Despite the obvious limitations and challenges that AI poses, it is seen to present an increasingly important role and possible support for mental health professionals.

Overall, maybe the safest mode of operation remains the supervision of the artificial intelligence by the mental health specialist, with periodic evaluation of both the patients and the optimal operating parameters of the algorithms. The psychiatrist should take full responsibility for any AI errors involved in the medical act⁽⁸⁾.

Autori pentru corespondenţă: Ovidiu Alexinschi E-mail: alexinschi@yahoo.com

CONFLICT OF INTEREST: none declared.

FINANCIAL SUPPORT: none declared.

This work is permanently accessible online free of charge and published under the CC-BY.