AI's Diagnostic Capabilities: Unspoken Success

Exploring what ChatGPT Health reveals about your well-being and the limitations it faces.

Have you ever turned to ChatGPT for health guidance, perhaps because of an unexplained skin condition or muscle stiffness post-exercise? I faced both instances and remarkably, ChatGPT identified my winter-induced rash as cold urticaria before my own doctor did.

OpenAI reports that over 230 million people consult ChatGPT weekly with health inquiries. While online health research is nothing new, the method has shifted. Instead of navigating countless search results, users now engage in what feels like a direct dialogue.

A New Trend in AI Health Tools

Recently, major AI players embraced this conversational health shift. OpenAI introduced ChatGPT Health, a specialized segment where users can link medical records and health data for customized insights. Initially limited to select users, it promises wider accessibility soon. Anthropic soon introduced a similar tool for Claude, alongside products for health professionals.

Despite disclaimers advising professional consultation, these tools have not deterred millions from using them to decipher their symptoms.

AI as a Diagnostic Tool: Overlooked Potential

AI shines in diagnosis—a process of matching symptoms to diseases, a method in line with AI training. AI models process input to link with diagnosed conditions, echoing the decades of knowledge from clinical experiences.

In a study, GPT-4 exceeded 90% accuracy on complex cases, notably surpassing doctors who averaged 74%. This margin affirmed AI's potential in identifying rare conditions over medical practitioners.

Challenges arise in treatment, where considerations extend beyond medication efficacy to include patient compliance, cost, and accessibility—nuances AI models cannot navigate.

Adam Rodman, a Harvard Medical School professor, asserts that management lacks definitive solutions, revealing the complexities AI can't yet emulate.

Health Monitoring: Real vs. Perceived Benefits

ChatGPT Health and Claude position themselves as health analysts, allowing users to track vital statistics. Yet, without independent studies to verify AI's interpretation of data trends, these observations remain ambiguous.

OpenAI's HealthBench aims to test AI in explaining lab data, although it relies on script-like interactions rather than real patient engagement.

While these AI features may assist in spotting lifestyle patterns, akin to tracking migraine triggers, the science backing their effectiveness in health predictions isn't concrete yet.

The Hidden Risks of AI in Health

Sharing health data with AI tools prompts privacy concerns, particularly since they aren’t governed by HIPAA laws meant to protect medical info. This vulnerability might strip your data of legal safeguards.

Given the current threats to reproductive and gender-affirming care in certain states, the risk of your data becoming a legal liability isn't hypothetical.

AI's tendency to mirror user emotions can be problematic, as it might adjust responses to match a user’s concern, a behavior professionals are trained to avoid.

Despite efforts by companies to make AI ask follow-up questions or direct users towards consultations, their real-world effectiveness remains unproven.

Moreover, there is a risk of perilous advice, as some models have delivered potentially harmful recommendations.

Tragic examples, such as the death of a teenager from AI-provided drug advice, underline the potential consequences of overreliance on AI for critical health decisions.

Making a Decision: To Use or Not to Use?

With long wait times for medical appointments, many turn to chatbots for immediate, albeit basic, answers. Chatbots are constantly available and persistent.

While AI can effectively interpret lab results and clarify medical terms, it's untested in its ability to extrapolate valuable insights from long-term health data. Crucially, AI cannot replace the nuanced understanding of a human health professional.