Unveiling the Transformative Potential of AI-Generated Imagery in Enriching Mental Health Research

Introduction to Visual Research in Mental Health

The role of arts in improving health and well-being has been increasingly supported worldwide (Fancourt & Finn, 2019; WHO, 2023b) since the 1960s–70s when art-therapy was recognized as a profession (Van Lith, 2016). Aligned with these developments, visual methods in mental health research have been extensively explored and utilized. Currently, there is a well-established body of evidence of visual approaches in mental health research including painting and drawing (Al-Rasheed & Al-Rasheed, 2023), sculpture (Seifert et al., 2017), photography (Buchan, 2020; Milasan et al., 2022), videography (Whitley et al., 2020), digital storytelling (De Vecchi et al., 2017), animation (Coughlan et al., 2021), and collage-making (Stallings, 2016) employed as stand-alone or combined artistic techniques.

Within this context, “arts on prescription” initiatives have been emerging internationally as an alternative model of mental healthcare (Bungay & Clift, 2010; Jensen et al., 2017; Stickley & Hui, 2012). These developments contributed to generating a breadth of research revealing the multi-fold benefits of creative research methods for both researchers (Coemans & Hannes, 2017; Gillibrand et al., 2023) and research participants (Milasan, 2024; Wang et al., 2020).

For research participants, one of the most evidenced benefits is the therapeutic and cathartic effect arts can have on their physical and psychological well-being. Enhanced self-expression, reflection, self-awareness, relaxation and reduced stress, reduced blood pressure, and boosting the immune system are only a few examples supported by the literature (Crawford & Patterson, 2007; Leckey, 2011).

Empowerment is another notable benefit at both individual and community levels as a result of voicing personal concerns and experiences while actively engaging with arts (Secker et al., 2007). For example, photovoice, i.e., a photography-based research method aimed at instilling social change (Wang & Burris, 1997), is known to support mental health by reducing stigma (Tippin & Maranzan, 2022) and promoting social connectedness (Han & Oliffe, 2016).

Closely related to empowerment, the exploration of strengths and positive traits conducive to resilience and optimism has also been identified as a key aspect in a systematic review on the use of art-based methodologies with young people with complex psychosocial needs (Nathan et al., 2023). Furthermore, visual arts are known to foster self-discovery and identity exploration leading to positive psychosocial outcomes that support and facilitate mental health recovery (Appleton, 2001; Lloyd et al., 2007). Therefore, visual arts must be perceived beyond their intrinsically pleasurable and recreational aspect for those involved by recognizing the opportunities they create for self-development. Finally, visual arts allow for capturing non-verbal aspects of communication to express subjective experiences of mental distress difficult to verbalize and often lost in the bio-medical discourse around mental health focused on diagnosis and illness (White et al., 2022).

For researchers, visual methods provide flexibility through combination with research methodologies such as phenomenology (Mitchell & Meehan, 2022), grounded theory (Griffiths, 2008), ethnography (Magnus & Advincula, 2021), narrative inquiry (Lindsay & Schwind, 2015), and participatory action research (Gillibrand et al., 2023). As a result, art-based research has the potential to enhance the exploration of subjective experiences through triangulation of data (Glaw et al., 2017) and, subsequently, provide researchers with a richer understanding and interpretation of phenomena under investigation (Fraser & al Sayah, 2011). Moreover, the use of visual arts in health research is versatile due to its applicability to a range of populations including young people, adults, and the elderly, from different cultural backgrounds, literacy levels, and communication styles which make art-based research more inclusive and holistic. This is facilitated by the expansion of images as a form of communication through social and other media resulting in the development of a universal visual language as an alternative to verbal and textual literacy (Ramón-Verdú & Villalba-Gómez, 2022).

In spite of increasing popularity of visual research methods, little is known about the potential of AI-generated images in research beyond the predominantly anecdotical evidence. The topic is currently heavily sensationalized with a primary focus on fake imagery and fraud (Gu et al., 2022), dehumanization of the creative act (Lee, 2022), and other negative aspects associated with AI. This article aims to critically discuss the potential benefits of AI-generated imagery along with limitations in visual mental health research that may be applicable to other subjects utilizing visual methodologies.

A Brief Exploration of AI-Generated Imagery

The use of AI in generating images and other creative outputs is far from being novel (Boden, 1998). Examples of digital art, i.e., artistic work or practice relying on digital technology as part of the creative process, can be traced back to the 1960s (Paul, 2008). Since then, digital art emerged as a novel form of creative media, for example, 2D and 3D computer graphics, digital photography, digital collage, digital storytelling, photo painting, 2D and 3D digital painting, and pixel art (Edmonds, 2019). Alongside digital art, experiments with algorithmic imagery to create visual compositions started in the 1970s with the seminal example of AARON^®, a computer program created by artist Harold Cohen (Morbey, 1992). AARON^® represents a milestone at the intersection between AI and artistic creation through autonomous creation of drawings, paintings, and other visual artworks. A more recent example of a creative AI platform capable of generating complex visual content in various styles and mediums is DeepDream^® developed by Google engineers in 2015 and adapted and popularized later by OpenAI^® (Suzuki et al., 2017). Unlike AARON^®, DeepDream^® is heavily reliant on the advances in neural network technologies, particularly deep learning in the 1990s–2000s. The result was surreal and dreamlike images generated by amplifying and repeating patterns within existing pictures manipulated autonomously through machine learning processes able to mimic artistic styles (Santos et al., 2021). These advances marked the start of developing neural style transfer algorithms allowing AI systems to apply the artistic style of one image to the content of another with the purpose of creating novel and aesthetically pleasing compositions (Cai et al., 2023). More recently, OpenAI^® developed another platform, i.e., DALL-E^® showcasing the potential of AI to generate diverse and creative images using textual prompts and descriptions. It can be observed that, while digital images involve direct input and manipulation from humans throughout the entire creative process, AI-generated imagery relies primarily on autonomous algorithms and learned patterns from existing data to generate new visuals often beyond the artistic vision of the human creators.

While these are only a few milestones in the history of AI-generated imagery, they reflect the diversity of techniques, purposes, and capabilities of novel technologies. The level of human involvement in this process also varies from one platform to another with images being influenced to various degrees by the input images and parameters provided by humans during the generative process. For example, the human input in the case of AARON^® involves the initial programming, i.e., rules, heuristics, and algorithms, and setup of the AI system by Harold Cohen. Once the aesthetic principles and stylistic tendencies were defined, AARON^® operates autonomously with the occasional human input in the form of feedback and guidance from the artist. Unlike AARON^®, DeepDream^® requires additional human input from users who provide images on which the algorithm is based to generate visually enhanced or manipulated versions of these images. As a result, users’ choices significantly influence the final output. In what concerns DALL-E^®, the human input in creating images is central to the process in the form of detailed textual descriptions, for example, characteristics, attributes, and scenarios. The textual information is then visually interpreted through natural language processing to generate imagery corresponding to the descriptions provided.

The idea of creativity, ownership of the creative output, intentionality, and the very concept of art are challenged in the context of using AI-generated imagery by referring to the generic human–machine dyad (Coeckelbergh, 2017). However, the issue appears to be more nuanced and requires a closer inspection of the role humans have in the process of generating the final creative output. This aspect is essential when translating some of the creative outputs into mental health research data particularly in relation to art-therapy which is a process defined by intentionality in relation to facilitating personal growth and healing (de Witte et al., 2021). Some authors (Feldman, 2017; Oh et al., 2018) refer to generative art processes as human–AI co-creation in which technology is perceived as complementary to the artistic endeavor rather than a surrogate.

The Potential of AI-Generated Imagery in Mental Health Research

WHO (2023a) produced a comprehensive study on AI in mental health research including applications and challenges. However, this resource does not cover the use of AI-powered imagery and its potential for mental health practice and research. Similarly, the literature on AI in mental health research is generally focused on the prognosis and diagnosis of psychiatric conditions (Graham et al., 2019; Minerva & Giubilini, 2023). In what concerns AI-generated images, their use in mental health research appears to be predominantly clinical such as medical imaging analysis and morphometry, i.e., method for identifying macroscopic anatomical differences among the brains of different subjects (Tornero-Costa et al., 2023). These findings suggest a gap in the non-clinical use of visual methods aided by AI platforms particularly in qualitative mental health research. To address this gap, this article outlines some of the potential benefits of AI imagery to advocate for the use of this novel and innovative method in mental health research, drawing on the emerging body of literature on AI and arts.

Enhanced Creativity Through Co-Creation Between AI and Humans

A growing concern is that AI risks to hinder human capacity for creativity and originality through automation of tasks and standardized visual representations, to the point of dehumanizing and depersonalizing art (Gosh & Fossas, 2022). However, some authors including Boden (2014), a seminal thinker on AI and arts, claim that creativity is inherent to AI through production of novel and surprising visual art. Furthermore, the potential of AI can be transformative in nature and allow humans to explore artistic ideas and conceptualizations that can serve as a source of inspiration or a blueprint for innovative and unexpected visual outputs (Fan & Liang, 2023). For example, AI can provide people who have limited creative skills with accessible tools that allow them to experiment with artistic expressions, explore new styles and techniques, and stimulate their imagination. The result could be the development of an artistic voice through collaboration and iterative dialogue with AI-powered creative algorithms that can augment human creativity by providing real-time feedback and suggestions throughout the process (Colton & Wiggins, 2012).

Co-creation seems to be a key theme in the literature on AI and art in which computational aesthetics are combined with input from humans (Wu et al., 2022). However, the ontological question remains whether AI that lacks human consciousness does indeed create art or the creative output generated through algorithms and statistical modelling is perceived as being art by an audience (Hageback & Hedblom, 2021). On this note, it is argued whether creativity lies in the final product or in the interaction between the computational systems and humans (Wingström et al., 2022). Such questions are important in art-therapy and mental health research showing that the cathartic effect of arts, e.g., personal growth, trauma healing, enhanced cognitive and neurosensory capabilities, and inter-personal relationships, results from the creative process rather than the artistic outcome itself (Shukla et al., 2022). Furthermore, the evidence on the potential negative bias people may have against AI-generated artwork compared to human-created art (Bellaiche et al., 2023) adds another level of complexity to this debate.

Accessibility and Inclusivity Throughout the Creative Process

Incorporating digital technologies in creative processes has been shown to increase accessibility to art-therapy since the 1990s. One notable example is Collie and Čubranić’s (2002) computerized system allowing people with limited mobility to engage with, and benefit from, an art-therapy group focused on trauma healing. Building on the principles of telehealth, Darewych et al. (2015) developed a similar approach with enhanced creative tools for adults with developmental disabilities. The digital media was perceived by the participants in this study as a “mess-free” therapeutic environment of particular benefit to individuals with olfactory and tactile sensitivity.

However, digital technologies for art-therapy have been criticized for being limited in what concerns the creative interface and capabilities that hinder users’ ability to fully benefit from the therapeutic process (Du et al., 2024). This shortcoming has been addressed through AI-powered solutions superior to digital and traditional art-making in terms of speed and efficiency (Fan & Liang, 2023). This is due to the complexity of AI creative tools, styles, techniques, and concepts accessible and customizable to a wide range of participants, including those with limited artistic skills, physical abilities, e.g., speech-to-image technologies, light sensitivity, and color blindness (Lin et al., 2019). Generative AI can effectively improve the accessibility of images through enhanced quality, e.g., brightness, contrast and saturation, and clarity through denoising and deblurring, and enhanced resolution, personalizing them in line with users’ needs and preferences (Chatterjee, 2022). More recent developments in this area include complex image recognition algorithms able to provide textual descriptions of images making them accessible to individuals with visual impairments (Chemnad & Othman, 2024).

Similarly, images can be made accessible to people with cognitive impairments often associated with mental health problems (Morozova et al., 2022). This can be achieved through natural language processing algorithms capable of identifying and correcting language (Le Glaz et al., 2021). As a result, AI tools open avenues for enhanced engagement, inclusivity, and participation in art-therapy and visual research in ways that are not possible through traditional art-making. It can be claimed, therefore, that AI is an alternative way to empower and democratize the access to creative tools and resources by making it easier for individuals with varying levels of artistic skill or expertise to engage in creative expression.

Considering the technological advancements and accessibility of AI-powered platforms embedded in smartphones and other portable devices, generative AI can be utilized flexibly in mental health practice and research with people of different ages and from diverse socio-cultural backgrounds, including situations when language can be a barrier to accessing psychiatric services and participating in research (Woodall et al., 2010).

Flexibility and Adaptability

Closely related to accessibility, flexibility refers to the ability to adapt and customize AI-generated imagery through adjustable parameters allowing users to control for various aspects such as artistic style, chromatics, composition, and perspective. The aim is to fine-tune and personalize the creative output to match individuals’ vision and preferences. This can be achieved through various means including interactive interfaces with real-time feedback and control over the image generation process and modular architectures to enhance customization, refine the creative output by training AI models, and transfer learning techniques (Ploin et al., 2022). The latter is successfully used in examining speech datasets to evaluate the cognitive abilities in people with dementia and other cognitive impairments, which is promising in supporting communication and comprehension (Zhu et al., 2021).

The AI-powered image generation process can be adjusted to individual skills by creating visual material from scratch or based on existing images accessible from the Internet or personal archives. The visual content can be manipulated to stylize and abstract images that may also benefit the research from an ethical perspective by removing identifiable elements in the existing images (Wiles et al., 2008).

Apart from being flexible, the digital creative process is also more efficient, less time-consuming, and resource-intensive compared to conventional arts although high costs may be associated with accessing more performant AI solutions. Furthermore, AI art may be generally perceived as being more sustainable with some authors claiming that the computational power required for training and running complex AI models can actually have a significant environmental impact (Dhar, 2020).

Limitations of AI-Generated Imagery in Mental Health Research

Undoubtedly, AI presents with a range of potential benefits for both research participants and visual researchers. AI-powered imagery creates opportunities for expressing experiences of mental health by bringing subconscious thoughts and emotions into focal awareness aided by digital visual aids and textual prompts. However, this process is not without challenges and limitations.

For example, it is known that traumatic experiences are often difficult to express verbally and textually (Talwar, 2007) which may pose some challenges with producing prompts and descriptions to generate the desired images with the aid of AI. A potential solution could be the use of visual metaphors generated through text that can be shaped iteratively through customization and interaction with the AI software until the desired outcome is achieved.

As shown in the previous section, AI-generated imagery may be conducive to more complex and versatile artistic expression beyond the artistic literacy of the users. This is particularly relevant for subjective experiences of mental distress such as psychosis characterized by auditory and visual hallucinations (Hanevik et al., 2013). However, considering that psychotic experiences are often related to depersonalization and derealization (Farrelly et al., 2016), it is uncertain what the impact of using a virtual environment as compared to a real-life creative setting would be on an individual’s mental health (Ricci et al., 2023). The use of online art workshops, for example, to facilitate a sense of connection and community with others may address this shortcoming while supporting the social function arts have in mental health (Wang et al., 2020).

Another question to be answered is whether the value of AI-generated art lies in the creative outcome or the process of creation which may vary from study to study and from individual to individual. It is known, for instance, that art-therapy is process-oriented led mostly by registered professionals who facilitate reflection on the created artwork (Hu et al., 2021). Replacing this interaction with a machine may impact the human nature of the process. In relation to this, recent research focused on anthropocentric creativity beliefs reveals that creative processes tend to be perceived as an inherently and uniquely human characteristic (Millet et al., 2023), while some authors perceive this as a myth (Arielli & Manovich, 2022). Nonetheless, there is a need for researchers to reflect on the potential sense of depreciation toward AI-generated art compared to traditional art-making.

AI-generated art is subjected to numerous ethical controversies and dilemmas (Stark & Crawford, 2019). Firstly, determining who owns the right to the AI-generated artwork can be complex and ethically problematic. Additionally, AI algorithms are trained on specific datasets that risk perpetuating and amplifying societal biases and stereotypes in generating art with potential implications on how experiences of mental health and distress are portrayed (Straw & Callison-Burch, 2020). Moreover, the content generated using AI platforms may include, intentionally or inadvertently, inappropriate or controversial aspects, for example, cultural appropriation. Such issues may cause further distress and discomfort to the research participants who have limited control over how the digital artwork is generated. Finally, as some AI art software allow the use of pre-existing images, the generated artwork may contain personal or sensitive information that needs to be managed responsibly by researchers (Grba, 2023).

AI has the potential to restore works of art by analyzing patterns and helping to recreate damaged or missing content (Gaber et al., 2023). Metaphorically, such restorative capabilities of AI platforms may extend to human beings with lived experiences of mental distress in search of creative outlets for self-expression and therapeutic exploration of emotions. However, considering the pitfalls presented by AI technologies, further research and philosophical discussions are needed to shape AI-driven imagery as a methodological avenue for mental health research.

Conclusion

AI-powered creativity has witnessed an exponential growth with considerable influence in the way contemporary art is understood and experienced. Despite heated debates whether AI may diminish the value of human creativity, AI capabilities a complex reality that is transforming the way art is conceptualized and, consequently, will soon influence how art-based research is conducted. Within this context marked by blurry boundaries between computational and human creativity, researchers need to develop a deeper understanding of this subject. For this purpose, this article explored several benefits of, and limitations to, using AI-driven imagery in mental health research that may be applicable to other areas of health and social care. As this is an emerging topic, more research on the effectiveness and therapeutic value of AI-generated art compared to conventional artistic approaches is required, from the perspective of the researchers and research participants to shape novel and innovative visual methodologies utilizing artificial intelligence.