Ivan Kairatov stands at the forefront of biopharmaceutical innovation, possessing a deep-seated understanding of how technology reshapes the patient experience from the laboratory to the living room. As the healthcare industry increasingly leans on artificial intelligence to bridge gaps in care, the shadows of societal prejudice—often termed health-related stigma—have become impossible to ignore. Kairatov’s expertise in research and development provides a unique lens through which we can view the dual nature of AI: its ability to expose our darkest biases through massive data analysis and its untapped potential to foster true health equity. Our conversation navigates the landscape of digital discourse across global platforms, the psychological nuances of human-machine interaction, and the urgent need for a shift from simply measuring social exclusion to actively dismantling it through narrative-driven technology. We examine how AI is currently being used to map these social barriers, the surprising data emerging from platforms like Reddit and X, and why the anonymity of a chatbot might be the key to unlocking patient trust.
How has the application of natural language processing to massive datasets changed our understanding of how conditions like schizophrenia are perceived compared to other health issues?
The sheer scale of data we can now analyze is staggering, moving far beyond traditional surveys to capture the raw, unfiltered pulse of public sentiment. When researchers sifted through over 27,552 records and ultimately synthesized 70 key studies published between 2016 and 2025, a striking disparity emerged in the digital footprint of different conditions. We’ve observed that while obesity-related stigma is frequently discussed in public policy, it is actually less prevalent in digital corpora than the stigma surrounding schizophrenia, which carries a much heavier exclusionary tone. Using machine learning to crawl through platforms like X, Reddit, Weibo, and Facebook, we’ve found that stigmatizing content can range from less than 1% to more than 40% of the conversation depending on the community. It is a sobering realization to see how schizophrenia often triggers more fearful and exclusionary language than almost any other condition mapped by these algorithms. This tells us that the “digital ghost” of mental illness is far more pervasive and haunting than the physical biases we often prioritize in public health campaigns.
The research suggests that while we are becoming incredibly efficient at identifying stigmatizing language, we are lagging behind in actually using AI to mitigate it. Why do you think this gap exists in current clinical research?
We are currently in a phase where AI acts more like a high-powered mirror than a surgical tool, reflecting our flaws rather than fixing them. The scoping review highlights that 42 of the studies focused strictly on detecting or measuring stigma, whereas only a meager four studies actually tested AI as a means to reduce it. This imbalance exists because it is computationally “easier” to build a sentiment detection model or a text classifier than it is to design a conversational agent that safely navigates a patient’s psyche without causing harm. In the United States, which led the research with 32 studies, and even in the UK and Singapore, the focus has been on quantification because the ethical risks of a “bad” intervention are so high. We have become experts at counting the instances of prejudice, but we are still treading water when it comes to implementing long-term clinical evaluations that prove AI can actually change a person’s heart or a provider’s bias.
What role does the perceived anonymity of digital interfaces play in how patients with highly stigmatized conditions choose to interact with healthcare AI?
There is a fascinating paradox at play when a patient sits down in front of a screen rather than a human doctor. For many, the anonymity of an AI creates a safe harbor, a judgment-free zone where they feel empowered to disclose symptoms of conditions that carry heavy social baggage. The review identified 15 studies examining this trust, finding that people are often more willing to use AI for sensitive health issues because a machine doesn’t have a facial expression that might register disgust or pity. However, this isn’t universal; there is a segment of the population that remains deeply hesitant, fearing that these systems might actually increase stigma by misrepresenting their data or labeling them in a permanent, digital way. It is a delicate emotional tightrope where the machine’s lack of “humanity” is its greatest strength for some and its most frightening liability for others.
We often worry about AI bias in a general sense, but what are the specific ways these systems can inadvertently amplify existing prejudices against people with disabilities or mental illnesses?
The danger isn’t just in the code; it’s in the “garbage in, garbage out” nature of the large language models we use today. Nine specific studies highlighted how AI can actually increase stigma, showing that when prompts include references to disability or mental illness, the generated responses are significantly more negative and fearful than those triggered by neutral terms. For instance, image-generation models have been caught reproducing harmful visual stereotypes, essentially “painting” illness with a brush of social exclusion. Even more concerning is the impact on professionals; one study found that when healthcare providers were exposed to certain machine learning predictive assessments, they reported a heightened sense of fear toward their patients. This suggests that AI doesn’t just reflect bias; it can actually “prime” a human doctor to view a patient through a lens of risk and apprehension rather than care and empathy.
Among the limited studies focused on reducing stigma, what specific tactics—such as the use of first-person narratives—have shown the most promise in shifting public and professional attitudes?
The most compelling evidence we have, though preliminary, points toward the power of storytelling through conversational agents. In the four studies that successfully reduced stigmatizing attitudes, the “secret sauce” was often the use of first-person narratives where the AI shared what it is like to live with a specific health condition. When a chatbot engages a user in a mental health-related dialogue using these personal stories, it breaks down the “otherness” that fuels prejudice. We are also seeing these agents adapted as educational tools for healthcare providers to decrease bias before they even enter the exam room. While these are currently small-scale experimental studies, they suggest that if we can make AI sound more like a person with a lived experience and less like a clinical textbook, we can actually move the needle on social inclusion.
What is your forecast for the evolution of AI as a tool for health equity over the next decade?
I predict that over the next ten years, we will transition from “text-only” analysis to truly multi-modal AI systems that can detect and counteract stigma through voice, gesture, and visual interaction. Currently, our research is heavily skewed toward mental health—representing about 76% of the literature—while conditions like leprosy or rare physical disabilities are almost entirely ignored. In the coming decade, we will see a much-needed push toward cross-cultural perspectives, moving beyond the US and UK-centric data to understand how stigma manifests in different global contexts. The ultimate goal is to move AI from being an analytical observer to a proactive “equity guardrail” that sits within our healthcare infrastructure. If we can integrate clinical, social, and computational perspectives, we won’t just be mapping the landscape of stigma; we will be actively redrawing it to ensure no patient is left behind because of a label.
