Google Researcher Seeks AI Solutions for Medical Breakthroughs, Inspired by Father's Parkinson's Battle

A Google researcher is working to harness the power of artificial intelligence in medicine, driven by a personal experience with his father's battle against Parkinson's disease. Vivek Natarajan, a DeepMind researcher at Google, has dedicated himself to developing AI systems that can assist doctors and scientists in diagnosing diseases, proposing experiments, and speeding up treatment discovery.

Natarajan's interest in using AI for medical breakthroughs began when he witnessed his father struggle with Parkinson's over several years. His father continued working until retirement age despite the physical and cognitive symptoms of the disease, leaving a lasting impression on Natarajan. This experience influenced his career choice and led him to ask himself where AI could have the most significant impact.

After completing his master's degree in the US, Natarajan joined Meta in 2017 but soon became drawn to using AI for healthcare. He had previously developed an app called 'Ask the Doctor Anytime, Anywhere' during his undergraduate studies and was aware of Google and DeepMind research pointing towards medical applications.

Natarajan joined Google Brain in 2019 after connecting with Greg Corrado, a founder of Google Brain. At Google, he found that healthcare progress depended on earning trust from physicians, patients, regulators, and policymakers, making it challenging to implement AI systems.

He began focusing on deeper questions such as reliability, uncertainty, generalization, and interactivity in AI systems for medicine. Natarajan realized that an AI system simply outputting a probability score was not sufficient; doctors needed explanations, while patients required conversations. Medicine is inherently contextual and deeply human.

Natarajan collaborated with Alan Karthikesalingam, a physician-scientist who shared his ambition to use AI in medicine. They were inspired by Google's significant scientific breakthroughs, including AlphaGo and AlphaFold. The pair proposed an initiative that eventually led to the development of Med-PaLM, which demonstrated the potential for large language models to learn from limited examples.

The team's work on Med-PaLM was a catalyst for further medical AI research at Google Brain. However, Natarajan and Karthikesalingam were not satisfied with simply passing medical exams; they aimed to create systems that could take patient histories, reason through diagnoses, and communicate empathetically. This led to the development of AMIE and Co-Clinician, which envisions AI as a collaborative member of a care team.

The focus expanded from medicine to science itself when Natarajan's team was approached by Stanford professor Gary Peltz with a question: Could their systems generate scientific hypotheses? The result was Co-Scientist, a Gemini-based multi-agent system designed to help researchers generate, debate, rank, and refine hypotheses. Initial tests showed promising results in areas such as antimicrobial resistance and cancer drug repurposing.

Natarajan acknowledges the risks associated with AI-generated hypotheses but emphasizes that his team's goal is to accelerate medical breakthroughs safely. He believes they now have a 'line of sight' towards understanding disease mechanisms broadly, which could lead to finding cures for many diseases.