⚡ SBS Premium Summary
AI technology is driving medical innovation by enabling personalized vaccine design and diagnosis for individual cancer patients through complex genetic analysis and protein structure prediction.
As reliance on AI in clinical settings grows, key challenges are emerging, including "deskilling"—the decline of medical professionals' practical skills—as well as the reliability of AI-provided information and data privacy leaks.
While AI-based medical services are expanding, core treatments such as cancer vaccines have not yet received final FDA approval, and technical limitations, such as the durability of therapeutic effects, still remain.
Hello. I am reporter An Hyemin, who handles and analyzes data. You may already feel firsthand how AI is changing our lives. However, there is a field where this change is felt particularly strongly abroad, including in the United States: the medical field. The era where AI designs new drugs, assists doctors with diagnoses, and creates cancer vaccines has already begun. In today's OhGraph, we will look at how AI is impacting the medical field. We will also examine, through various data and graphs, whether the future is as rosy as expected.
Cancer Vaccines Made with AI? AI Shaking Up the Pharmaceutical Industry
What if your beloved pet falls ill with a terrible disease and has to suffer continuously because there is no proper treatment? Paul Conyngham, an IT entrepreneur from Sydney, created a treatment himself for his dog, Rosie, who was diagnosed with a terminal mast cell tumor. Did Conyngham have medical knowledge? No. To create the cancer vaccine, Conyngham utilized AI.
First, starting from the second half of 2024, he used ChatGPT to find a direction for treating Rosie and located the right experts. After that, he commissioned a genomics center to analyze Rosie's healthy DNA and tumor DNA to perform gene sequencing. Gene sequencing is a technology that reads the exact sequence of DNA bases to decode a sort of blueprint of the body. After comparing normal cells with tumor cells, Conyngham identified mutant proteins present only in the tumor cells.
Next, he used Google DeepMind's AlphaFold to predict the 3D structure of the mutant proteins. This is because knowing the 3D structure of the protein is essential to understanding which part the cancer vaccine should target. Finally, he completed the final vaccine design using Grok. He delivered this data to the RNA Institute at the University of New South Wales, and in less than two months, a personalized vaccine for his dog was manufactured. The personalized vaccine was administered late last year, and it showed outstanding therapeutic effects, with Rosie's tumor shrinking by 75% by March of this year. Many people were astonished by this case of someone with no medical knowledge creating a personalized vaccine using just a few AI tools.
The creation of personalized vaccines using AI is not just an isolated incident; it is an area where actual pharmaceutical companies are also jumping in. Particularly in the case of cancer, even if cancer cells are removed through chemotherapy, recurrence is not uncommon. Therefore, the global pharmaceutical industry is focusing on "cancer vaccines" to prevent recurrence by eliminating hidden cancer cells. For reference, unlike vaccines generally used for disease prevention, cancer vaccines are closer to therapeutic agents that help the immune cells of patients who already have cancer attack cancer cells more effectively.
However, although pharmaceutical companies have been working on this for over 20 years, there has been no major progress. This was because it was difficult to accurately identify the targets of cancer cells. But whispers are spreading that utilizing AI could break through this barrier—by predicting mutant protein structures using AI like AlphaFold, just like in Rosie's case.
Currently, the frontrunner in this field is a cancer vaccine jointly developed by U.S. pharmaceutical companies Merck and Moderna. Let's look at the graph to see how effective it is.
These are the results of a five-year long-term follow-up of patients with melanoma, a type of skin cancer. It shows the recurrence and mortality rates of patients who used conventional anticancer drugs compared to those who were also administered the personalized cancer vaccine. Looking at the graph, patients who used the personalized cancer vaccine alongside conventional treatment had a 49% lower risk of recurrence or death.
Pharmaceutical companies that have felt the power of AI are actively utilizing it. Some experts even predict that AI will bring about a massive singularity across the medical and bio-economy.
The size of the AI-based drug discovery market is also growing. While forecast sizes vary slightly by analysis agency, the growth trend is projected similarly.
Market research firm Grand View Research estimated the AI-based drug discovery market at $2.3 billion in 2025. It is projected to increase to $2.9 billion this year and reach $13.8 billion by 2033. The compound annual growth rate (CAGR) is a whopping 24.8%.
Among AI big tech companies, some are directly entering the bio sector. A prime example is Nvidia. At the J.P. Morgan Healthcare Conference, the world's largest pharmaceutical investment event held in San Francisco last January, Nvidia participated directly, whereas the event would normally have centered around traditional large pharmaceutical companies. At this conference, Nvidia announced the establishment of a joint AI innovation research center with Eli Lilly. This research center is expected to adopt Nvidia's Vera Rubin system and BioNeMo, an AI platform specialized in drug discovery.
Nvidia has partnered not only with Eli Lilly but also with various healthcare-related companies, including clinical research service providers, genomics analysis firms, and general hospitals where actual treatments will take place.
However, looking at these moves, it is difficult to say that Nvidia is entering the pharmaceutical or medical market as a direct player. Nvidia's goal is not to become a pharmaceutical company. Rather, its goal is to make its GPUs and AI platforms an essential foundation for drug discovery. Just as its CUDA and GPUs became indispensable "shovels" in the deep learning and AI model market, it is deploying a similar strategy in the pharmaceutical sector.
On the other hand, Google DeepMind is taking a different stance. While other big tech companies are building AI pharmaceutical platforms or specialized models, DeepMind has jumped directly into the market as a drug design player. Isomorphic Labs, established in 2021 as a spin-off of DeepMind, is racing toward the goal of AI-driven drug discovery, using AlphaFold as its core foundation.
Is AI Already Better at Diagnosis? AI Enters the Clinical Setting
The influence of AI is growing not only in the pharmaceutical sector but also in personal healthcare and medical diagnostics. Healthcare-specific services have already been launched on ChatGPT and Claude, which we frequently use. OpenAI and Anthropic introduced related services in January of this year. However, these services are not yet available in South Korea and can only be used in the United States.
By using this service, you can basically have various conversations with AI about your health information. Furthermore, because it is linked to databases used in the U.S. healthcare system, you can receive assistance such as having the AI write appeal documents for insurance claims. Since its launch, a vast number of people have been utilizing healthcare-specific AI services. According to OpenAI, they handle more than 40 million health-related queries every day.
The high demand for medical AI in the United States paradoxically highlights the issue of low healthcare accessibility in the country. One-fifth of the U.S. population lives in rural areas, and people living in these regions have higher premature mortality rates than those in urban areas. Such medical blind spots are referred to as "medical deserts." This is similar to the shortage of essential medical services in rural provinces in South Korea.
We looked at how much ChatGPT's healthcare services were used in these "medical desert" areas, where it takes more than 30 minutes by car to reach a hospital. States in the West with small populations and vast geographical areas, such as Wyoming, Oregon, and Montana, showed high usage rates. Hawaii also ranked near the top, reflecting the difficulty of accessing hospitals due to its geographical characteristics as an island.
Not only individuals looking to manage their health but also doctors working in clinical settings are actively utilizing AI. The scope of application is diverse, ranging from assisting with administrative paperwork to suggesting prescriptions. The American Medical Association asked doctors how much they use AI in clinical settings. In 2023, that proportion was only 38%. However, it surged to a whopping 81% this year.
In particular, OpenEvidence, an AI service that suggests prescriptions based on medical research data, is highly popular.
In April of this year, the number of clinical-related prompts exchanged on OpenEvidence reached a whopping 27 million. This is a more than tenfold increase compared to the end of 2024. The number of U.S. doctors using OpenEvidence is estimated at around 650,000, which is about 65% of all doctors.
In some experiments, results show that AI diagnoses better than residents and specialists, and cases have been reported where AI identified diseases that doctors missed.
One patient visited an emergency room after suffering from hemoptysis (coughing up blood) and difficulty breathing for three days. At the time, the medical staff diagnosed it as simple asthma based on chest X-rays and electrocardiogram (ECG) results. However, an AI program being tested at the hospital produced a different result. The AI program indicated that the patient showed signs of severe heart damage. The hospital quickly recalled the patient for a re-examination, and the AI turned out to be correct. Fortunately, the patient underwent a heart transplant in time, preventing a tragedy.
Will AI Open a "Brave New World" in the Healthcare Market?
AI, which designs new drugs and makes quick diagnoses. Looking only at the cases discussed above, one might expect that AI can change everything. However, there are still mountains of tasks to solve, as AI-generated outputs are not perfect. An institution in Germany analyzed health information exposed on Google's AI Overview. Looking at the sources of health information provided by the AI, there were few reliable medical sources, and instead, there were many YouTube videos—to the extent that it referenced YouTube twice as much as authoritative medical sources.
There are also research findings indicating problems with medical information obtained through conversations with AI chatbots.
A study presented various scenarios to 1,298 British adults.
"A 31-year-old pregnant woman is experiencing chest pain and shortness of breath. What condition is suspected, and what is the appropriate action?"
When examining the performance of AI alone in identifying the disease, the score is high like this.
However, when ordinary people conversed with the AI, the accuracy rate was actually lower. It was even lower than the score obtained by simply Googling without using AI.
Why did this result occur? The researchers point out the limitation that, from the perspective of the general public, AI does not perfectly provide the information necessary to make a medical judgment. Furthermore, users fail to extract the exact medical information they need from the AI's responses.
Because of these results, having medical AI supplement medical blind spots may not be entirely positive. This is because it is difficult for people without medical knowledge to accurately judge the medical information generated by AI.
Then, is it fine if doctors with medical knowledge use AI? An experiment was conducted on experienced doctors who have performed colonoscopies for a long time. Before the introduction of AI, their detection rate for polyps that could develop into cancer in the future was 28.4%. However, after AI was introduced, that rate dropped to 22.4%. Although AI provides assistance, a phenomenon called "deskilling"—where practical skills decline the more one relies on AI—was confirmed. This suggests that continuous exposure to and high reliance on AI can lead to a decline in skills.
Another point we should pay attention to in medical AI is our medical data processed by AI. In fact, personal medical information is highly sensitive personal data. In the United States, where medical AI is currently expanding, there is a growing problem of medical institution staff uploading patient information to AI without hesitation. Because many doctors enter patients' personally identifiable information into OpenEvidence, which was introduced earlier, some hospitals have officially requested doctors not to upload patients' medical information.
Among these, genomic data used for personalized vaccines is likely to be particularly sensitive. Since my DNA will never change, once it is leaked, it is highly sensitive personal information that poses a lifelong risk. Furthermore, because one's DNA can reveal some genetic information of their parents, siblings, and children, the risk is even higher.
To continue the story of personalized vaccines, while it is true that AI offers hope, it is still a long way off. Although many companies are jumping in, not a single one has received final FDA approval yet. There is also a limitation that mutations can differ from cell to cell even within a single tumor.
Rosie, the dog mentioned earlier, also had some tumors that did not respond to the personalized cancer vaccine. Rosie showed signs of improvement initially after receiving the personalized cancer vaccine. However, her condition recently deteriorated to the point where treatment became difficult as the tumor spread rapidly. Due to her advanced age and the rapid spread of the tumor, Rosie was in extreme pain. Ultimately, Conyngham made the hardest decision: euthanasia.
In the EU, the AI Act classifies medical AI as a high-risk system. This means that because it directly affects life and health, it must be handled with greater caution than in any other field. As in the story of Paul Conyngham, who created a cancer vaccine with AI for Rosie, the potential of AI medicine certainly exists. However, safety and reliability issues still remain at the same time. What will the future of medicine solved by AI look like? That is all for today's OhGraph. Thank you very much for reading this long article to the end.
References
- Paul Conyngham@paul_conyngham) | X
- Tech boss uses ChatGPT to create cancer vaccine for his dog | The Australian
- Intismeran Autogene Plus Pembrolizumab Versus Pembrolizumab Alone in High-Risk Resected Melanoma | Adnan Khattak (2026)
- What If AI Doesn’t Ruin Everything? Understanding Artificial Intelligence With Reid Hoffman | This is Gavin Newsom, Reid Hoffman
- Artifical Intelligence In Drug Discovery Market | Grand View Research
- Isomorphic Labs is using Google Cloud’s AI Hypercomputer to reimagine drug discovery | Google Cloud
- OpenAI: AI as a Healthcare Ally [Jan 2026] | OpenAI - A case of artificial intelligence-enhanced diagnostics leading to heart transplantation | Heidi S. Hartman(2026)
- Reliability of LLMs as medical assistants for the general public | Andrew M. Bean(2026)
- Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy | Krzysztof Budzyń(2026)
Written by
: An Hyemin
Designed by
: Ahn Jun-seok
Interns
: Kim Su-young, Shin Yeon-seong
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.