Rowland Illing, chief medical officer (CMO) leading the healthcare business at Amazon Web Services (AWS), the world's largest hyperscaler, warned that antibiotic resistance could become a serious problem in the medical field. Antibiotics have played a central role in disease prevention since the discovery of penicillin in 1928, but they could become useless as bacteria develop resistance, he said. His argument is that AI research must accelerate to conquer cancer and overcome bacterial resistance, amid rising expectations for AI's use in medicine after Demis Hassabis, CEO of Google DeepMind and co-developer of the protein structure prediction AI model "AlphaFold," and others won the 2024 Nobel Prize in Chemistry.
Seoul Economic Daily interviewed CMO Illing at "AWS Summit 2026," which opened at the Walter E. Convention Center in Washington, D.C., from the 30th of last month through the 1st of this month. The interview is presented below in question-and-answer format.
What does AWS do in the healthcare field?
We work to democratise access to technology so that patients can achieve better outcomes. To this end, we collaborate with end customers, healthcare providers, governments, research institutions, hospitals, and insurers. We also work with medical records service companies, genomic sequencing firms, and life science organizations developing next-generation robots and drugs to provide new treatments. AWS has invested intensively in healthcare and life science-specific services over the past six to seven years to make it easier to develop new solutions.
AI companies are jumping into new drug development. What about AWS?
We are supporting 19 of the world's top 20 pharmaceutical companies. Large pharmaceutical companies developing new drugs use AWS. We collaborate with them across all areas, from research and development to clinical trials, manufacturing, commercialization, and post-market evaluation. We also work with regulatory agencies. For example, we support the U.S. Food and Drug Administration (FDA)'s drug evaluation platform and collaborate with the European Medicines Agency.
AWS also runs several collaborative projects in Korea. We maintain a partnership with Samsung Medical Center. We built an AI platform that analyzes emergency records at 17 Korean hospitals and performs early detection of sepsis and severity assessment. We also collaborate with research institutions, healthcare service providers, and regulatory agencies.
We maintain good relationships with Anthropic, OpenAI, Cohere, and open-source AI model companies. What we are trying to do is broaden access to a variety of models.
Please introduce your collaborative projects with Korean companies.
A representative example is Korea's Inocras. Inocras is a company specializing in genomic diagnostics, whole genome sequencing (WGS), bioinformatics, and clinical interpretation services. We actually performed genomic sequencing analysis with Inocras and the U.S. Broad Institute and presented it at the American Association for Cancer Research (AACR) held in San Diego.
Cancer patients have genetic sequences. We analyzed more than 8,000 genes and over 30 cancer types, and discovered more than 250 million different variants across cancer types. There is a great deal of information that can be obtained to develop personalized treatments. Korea is leading in this technology. The remarkable technologies emerging in Korea are spreading around the world.
Can AI really cure cancer?
AI cannot directly treat cancer, but it can be used to develop models that can help treat cancer. AWS has a biological foundation model harness called Amazon Bio Discovery. It supports integrating 40 biological foundation models. For each model, you can apply test targets to examine how new drugs can be developed. It is not simply a harness for AI models, but actually serves as a link that allows models to be built in the laboratory.
Over the past year, we developed a harness for AI models with Memorial Sloan Kettering Cancer Center. Memorial Sloan Kettering Cancer Center had a pediatric brain cancer target. The center selected various models to create antibodies to attack the cancer. They created more than 300,000 different antibody molecules to attack the cancer, and narrowed them down to 100,000. Antibodies should treat cancer and not attack other things. So they must be very selective. Among many models, you have to select models that have strong activity against cancer but no activity outside of cancer, actually manufacture them in a real laboratory, and confirm their efficacy. The experimental results are then fed back into the model.
Using foundation models to develop drugs, testing them in the laboratory, and then feeding the results back into the model—repeating this process narrows down the options. As a result, we identified five antibodies that we believe are actually suitable for creating new drugs. This process took a few weeks to a few months. In the past, such analysis took years. I think it is very encouraging that biological foundation models and large language models (LLMs) can lead cancer treatment in the future.
We collaborated with the previous administration on the Cancer Moonshot project, a pediatric cancer initiative. Together with 33 major children's hospitals worldwide, we created a data platform to collect data on rare cancers, especially rare pediatric brain cancers. Five years ago, we also launched the Health Equity initiative. We provided credits to organizations working to improve health disparities and provided more than $90 million (138.6 billion won) in support. Through this initiative, we also supported pediatric programs and projects helping women suffering from postpartum hypertension. We also collaborated on a program providing children's health tracking services in India.
Can AI lower the high burden of medical costs?
If you look at a patient's entire treatment journey, it includes searching for care, guidance, and insurance linkage. One of the ways we are trying to solve the medical cost problem is by reducing wasted time and unnecessary tests during the treatment process.
Sharecare is a great example. (Sharecare is a digital healthcare company that announced on the 30th of last month it would launch a next-generation health navigation platform in collaboration with AWS.) Sharecare operates a service called Ask MD, which supports navigation of the medical system to solve cost problems that arise when patients don't know where to go. It has 75 million registered users in the U.S., of which 10 million are active.
AWS operates Amazon Connect Health, an AI solution that manages medical team task management work. We launched five agents (AI assistants) to support the medical field. You can check who your attending physician is and when they are available to book an appointment.
What do you think about AI control?
In the medical field, AI control is an issue that must be taken very seriously. AI and humans must participate together. Human involvement means being engaged in the work of AI agents. An AI agent is autonomous software that can perform tasks and achieve goals independently. But those goals must be set by humans. There can be multiple agents, but ultimately humans must bear responsibility for the results. There are sub-agents that present and test ideas. They are managed by supervisory agents. But ultimately, it must be humans who take responsibility for how AI or agents behave.
What is the biggest challenge in the medical field?
There are several, but currently I am thinking about the problem of antibiotic resistance. The problem is that many bacteria are now resistant to the antibiotics currently in use. Deaths from this are projected to reach 39 million from 2025 to 2050. So we are focusing on this problem.
We work together with the Fleming Initiative, which aims to research new antibiotics and antimicrobial resistance treatments, to safely and scalably...
