Shweta Maniar leads the Healthcare & Life Sciences strategy at Google Cloud. She provides an update for the JDN on the impact of AI on the health sector.
JDN. In the medical field, where is AI progressing the fastest? In screening, drug discovery or diagnosis?
Shweta Maniar. 92% of drugs fail phase 1 clinical trials, meaning only 8% make it past this stage. This is where the potential of AI is most immediate and most important: in the early screening of candidate molecules, to both identify those that deserve to advance to phase 1, and predict failure before even entering it. The three use cases you cite will all progress, but it is in this upstream part of the drug development process that the impact can be the most transformative.
Will we need more computing power to achieve real breakthroughs? What are the structural limits?
The main limit is no longer computing power, it is increasing skills. Training teams and getting them comfortable with these tools as they scale is the real challenge. What I often say is that AI tools are no longer just for IT teams. We are all technology people now. The researcher, the marketer, everyone who works in these life sciences companies must understand and appropriate these capabilities. The question is how long it takes for everyone to start feeling truly comfortable. But the teams that go through the next launches will gradually become much more digitally and AI native in a way that is fundamentally different from what is being deployed today.
Are there scientific fields where AI has enabled new discoveries, impossible to obtain by traditional methods, or does it all come down to the optimization of existing processes?
We have been working with Recursion, an AI drug discovery company, for several years. We supported them with our TPUs and biomapping. Today, they have their first drug in phase 3 clinical trials, a direct result of their AI approach and this result is partly attributed to the work accomplished with Google Cloud. These are the types of stories that illustrate what we seek to make possible.
We are particularly excited about the life sciences industry because we have developed capabilities specific to this sector. Gemini can be fine-tuned on medical data, and it has the ability to read and understand genomic information, making it particularly valuable to this industry. What we are observing is a growing trend: our customers have less and less need to fine-tune these models. Sometimes they still do it, we learn from it, and these learnings gradually become integrated into Gemini.
Do you see differences in adoption of AI in health and science in general between Europe and the United States? Of what nature?
The differences are real and profound, they are fundamentally distinct markets. In Europe, as in other regions such as Asia-Pacific, we absolutely must integrate the issues of cloud sovereignty and regulatory compliance: GDPR and other local requirements. That’s why we invest a lot of time with the regulators in each market: the local equivalents of the FDA, but also the large hospital systems, to understand how they serve their markets and what their specific requirements are.
