The Promise and Limits of AI in Healthcare: Anastassia Žukova’s Insights from the Biohacking & Anti-Aging Conference
The world is undergoing a fundamental shift in how it perceives technology. Just a few years ago, tech giants competed over whose artificial intelligence was more advanced. Today, the paradigm has changed, and Silicon Valley’s defining mantra has become: “AI is everything.” Neural networks are rapidly making their way into every industry, and investors are readily funding the integration of intelligent algorithms into virtually every area of business and everyday life.
The health and biohacking industry has emerged as one of the most promising frontiers for AI-driven solutions. But how practical are today’s AI tools? Are algorithms truly ready to take on the responsibilities of diagnosing disease and managing human health?
This thought-provoking topic was explored by Anastassia Žukova, Head of an innovative preventive medicine project in Tallinn. Anastassia attended a major international conference dedicated to anti-aging medicine and evidence-based longevity. The insights she brought back offer a fresh perspective on the capabilities of modern AI while dispelling many of the popular myths surrounding “digital biohacking.”
Why Preventive Medicine Needs AI and nd How Mathematics Measures the Strength of Scientific Evidence
Integrating AI into preventive medicine is not about chasing technology trends. It is a practical and rational response to the overwhelming volume of scientific information being produced today.
Hundreds of medical papers and clinical trial results are published every single day. Given the limits of human capacity, it is simply impossible for healthcare professionals to review and evaluate all of them in a timely manner. As a result, preventive medicine faces a significant challenge: many high-profile studies promising “revolutionary treatments” ultimately prove to have weak scientific evidence.
One compelling example of addressing this challenge is the Nestarenie (Nestarenie.net) project created by Dmitry Veremeenko. The project team has implemented AI to automate the assessment of the reliability and scientific quality of medical publications.
How does it work?
A specialized algorithm applies mathematical models to perform an in-depth audit of scientific data, effectively separating evidence-based science from subjective speculation.
How AI filters scientific evidence:
- Evaluating the study population: The algorithm distinguishes between studies conducted on humans and those performed exclusively on laboratory animals, recognizing that animal research cannot automatically be translated into clinical outcomes for humans.
- Calculating statistical significance: AI determines the actual rate of successful outcomes rather than relying solely on the percentages reported by study authors.
- Filtering out the placebo effect: The algorithm estimates the probability that improvements observed in study participants resulted from placebo responses or random chance rather than the intervention itself.
The speaker noted that the algorithm uses only publications with a reliability score of 80% or higher when generating medical recommendations.
Using this “cleaned” evidence base, the AI platform performs its primary practical function – automated risk assessment. Users upload the results of comprehensive medical examinations, including blood tests, ultrasound scans, and MRI reports. The neural network instantly compares their health markers against validated scientific evidence, calculates personalized disease risks, and provides individualized lifestyle recommendations.
The 28 Blood Test Experiment: Where Popular AI Models Fall Short
Another major topic discussed at the conference was the growing use of large language models (LLMs) by both physicians and patients.
“Today, virtually every patient comes to their appointment after already interpreting their lab results with ChatGPT,” Anastassia noted.
To evaluate how well general-purpose AI models perform in real-world diagnostics, one of the conference speakers conducted an experiment. He submitted the exact same prompt, containing his own 28 blood test results, to all of the leading AI models currently available, including DeepSeek, GPT, Claude, and Groq.
The results were sobering: the models consistently made critical errors.
The key weaknesses of today’s LLMs in medicine:
- A tendency to hallucinate. During testing, the base GPT model occasionally fabricated medical studies or cited seemingly authoritative sources that, upon verification, turned out to be opinion articles with no connection to evidence-based medicine.
- Long-context degradation. When generating complex therapeutic protocols, AI models demonstrated instability. They could suddenly “forget” essential information provided earlier in the conversation and produce inaccurate recommendations (for example, confusing medication schedules or dosages), which makes continuous human oversight essential.
- Differences in model quality. According to the experiment participants, Claude currently delivers more accurate, balanced, and nuanced analyses of complex medical texts than the standard version of ChatGPT. However, GPT also becomes a highly capable tool when extensively customized and trained for specific medical applications.
Why AI Makes Mistakes: The Conflict Between “Reference Ranges” and “Optimal Health Ranges”
The primary reason why general-purpose AI often produces inaccurate medical analyses lies in the data on which it was trained. Standard AI models typically rely on conventional laboratory reference ranges – the average values considered acceptable for generally healthy individuals, or more precisely, values indicating that hospitalization is not yet necessary.
Preventive medicine, however, focuses on optimal health ranges, which can reveal hidden deficiencies long before disease develops.
The limitations of AI in health assessment:
- Superficial analysis: AI recognizes that a biomarker falls within the laboratory’s reference interval and concludes that everything is normal. It fails to detect latent deficiencies, such as early-stage iron deficiency, that may be the true cause of chronic fatigue, low energy, or burnout.
- Lack of context: High-quality diagnostics require AI to integrate multiple layers of information. Blood biomarkers should be interpreted alongside genetic data, family medical history, sleep patterns, and extensive health questionnaires consisting of at least 160 questions.
General-purpose AI simply lacks sufficient clinical context. Without specialized medical training, it delivers only a superficial interpretation of laboratory results, and relying on such analysis without professional oversight may pose significant health risks.
The Key Takeaway
Summing up the discussion, Anastassia Žukova articulated a fundamental principle that applies equally to medicine and business management:
“Today, AI in medicine is an exceptionally powerful accelerator of operational processes. It is indispensable for processing enormous volumes of information, filtering scientific literature, and automating the initial assessment of medical data.”
However, no machine can replace human reasoning, clinical experience, or the ability to perform deep, systems-level analysis. AI is a powerful assistant, but the final judgment and responsibility must always remain with the human expert.
This article summarizes the core insights from the Baltic Business Club meeting. The content is for informational purposes and reflects the practical frameworks shared by the invited experts and club members.
