Data governance for AI and Healthcare: From Washington to Beijing

1. Jenna Ben Yehuda: AI's Healthcare Revolution

Timestamp: 00:07:06 to 00:08:06 - watch this moment on skim

AI is fundamentally transforming healthcare by accelerating drug discovery, improving clinical research, and enhancing healthcare delivery. The urgency to translate these technological advances into better treatments faster is paramount due to aging populations and the growing burden of chronic diseases. Realizing this potential hinges on access to high-quality, globally sourced data that can be used and shared responsibly.

Significance (High): Sets the stage by highlighting the immense potential and critical need for AI in healthcare, emphasizing data as the foundational element for innovation and improved patient outcomes.

Sources in support: Jenna Ben Yehuda (Executive Vice President, Atlantic Council)

2. Ranjit Kublé: AI's Greatest Impact in Discovery

Timestamp: 00:14:00 to 00:18:00 - watch this moment on skim

While AI drives impact across the entire healthcare value chain, its most significant breakthroughs are currently seen in the discovery phase. This includes target identification, where AI helps understand complex disease mechanisms and derisk novel therapeutic innovations, potentially unlocking entirely new avenues for medicine development that might otherwise be missed.

Significance (High): Pinpoints the early stage of drug development as the area where AI is most catalytic, suggesting that AI's ability to navigate complexity and uncertainty offers the greatest potential for novel breakthroughs.

Sources in support: Ranjit Kublé (Vice President, Trusted AI, Pfizer)

Neutral sources: Graham Brookie (Vice President, ACT)

3. Ranjit Kublé: Trusted AI is About 'And', Not 'Or'

Timestamp: 00:19:16 to 00:21:03 - watch this moment on skim

Trusted AI is not a trade-off between speed and safety; it's about integrating both. Safeguards are essential to ensure AI benefits are distributed equitably and consistently. Similar to how powerful cars have safety systems, AI requires robust guardrails that operate seamlessly, enabling rapid progress without compromising ethical standards or patient well-being. Literacy around AI trust and risk is also crucial.

Significance (High): Defines 'trusted AI' as an integrated approach where safety and speed are complementary, not conflicting, principles, emphasizing the importance of built-in safeguards for equitable benefit.

Sources in support: Ranjit Kublé (Vice President, Trusted AI, Pfizer)

Neutral sources: Graham Brookie (Vice President, ACT)

4. Ranjit Kublé: Scaling Data Readiness for AI

Timestamp: 00:21:27 to 00:23:35 - watch this moment on skim

The biggest challenge in leveraging AI at scale is data readiness. While companies have achieved success with flagship use cases, preparing data for a multitude of applications is becoming unsustainable. The industry is shifting towards building data readiness natively into source systems to avoid extensive manual work, moving from a use-case-by-use-case approach to a more systematic, enterprise-wide solution.

Significance (High): Identifies data readiness as a critical bottleneck for scaling AI in healthcare, highlighting the industry's transition towards proactive data architecture rather than reactive data preparation.

Sources in support: Ranjit Kublé (Vice President, Trusted AI, Pfizer)

5. Jenna Ben Yehuda & Ranjit Kublé: Geopolitics and Global Health Data

Timestamp: 00:23:35 to 00:27:49 - watch this moment on skim

Global health outcomes are inherently global, requiring data from diverse regions. However, increasing geopolitical tensions and fragmented data regulations across the US, EU, and China create significant hurdles. While localized data models and compute-near-data architectures are emerging, achieving coherent global health outcomes necessitates greater alignment on standards, interoperability, and privacy-preserving methodologies.

Significance (High): Underscores the complex interplay between global health needs, geopolitical realities, and data governance fragmentation, suggesting that international cooperation on standards is vital for progress.

Sources in support: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Ranjit Kublé (Vice President, Trusted AI, Pfizer)

6. Ranjit Kublé: The Shift to Precision Advantage

Timestamp: 00:25:33 to 00:27:49 - watch this moment on skim

The paradigm is shifting from scale-advantaged AI to precision-advantaged AI in healthcare. This means competitiveness will increasingly rely on the depth and specificity of data, enabling more confident identification of biological targets, precise medicine design, and hyper-stratified clinical trials. This precision focus facilitates faster adoption and execution of AI innovations.

Significance (High): Articulates a significant strategic shift in AI for healthcare, moving beyond sheer data volume to emphasize data quality and precision for more effective and targeted innovation.

Sources in support: Ranjit Kublé (Vice President, Trusted AI, Pfizer)

7. Ranjet: Industry-Led Architectures as Blueprints

Timestamp: 00:32:21 to 00:33:52 - watch this moment on skim

Industry collaborations, even those involving proprietary data, can establish technological architectures that serve as blueprints for governments to manage sensitive data, such as in healthcare. These pre-competitive designs demonstrate solvability and can accelerate progress by aligning on technical possibilities rather than debating feasibility.

Significance (High): This approach bypasses governmental inertia by showcasing real-world solutions, potentially fast-tracking AI adoption in regulated sectors like healthcare.

Sources in support: Jenna Ben Yehuda (Executive Vice President, Atlantic Council)

Neutral sources: Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence)

8. Data Gravity and Compute Localization

Timestamp: 00:34:21 to 00:36:20 - watch this moment on skim

The concept of 'data gravity' highlights the immense challenge and cost of centralizing large datasets for AI training and use. Consequently, bringing compute to where the data resides is a critical consideration, especially with the rise of sovereign regulations, to overcome friction, latency, and operational hurdles.

Significance (High): This paradigm shift from centralized data to distributed compute is essential for efficient, compliant AI deployment, particularly in multinational organizations and regulated industries.

Sources in support: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council)

Neutral sources: Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Ranjit Kublé (Vice President, Trusted AI, Pfizer)

9. Ranjet: Technological Advancement for Model Efficiency

Timestamp: 00:36:28 to 00:37:21 - watch this moment on skim

A core technological advancement is needed to make AI models more energy-efficient and less resource-intensive. The current 'next token prediction' paradigm, while impressive, consumes significant energy. Future progress requires more efficient methods for content generation and inference, alongside sustainable practices and user literacy in algorithm usage.

Significance (High): Improving model efficiency is paramount for scaling AI in resource-constrained environments like healthcare and for achieving truly sustainable AI development.

Sources in support: Jenna Ben Yehuda (Executive Vice President, Atlantic Council)

Neutral sources: Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Ranjit Kublé (Vice President, Trusted AI, Pfizer), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council)

10. Gil: Resource Constraints and Collaborative Solutions

Timestamp: 00:44:18 to 00:45:42 - watch this moment on skim

The demand for compute capacity in AI development, especially for frontier research, often outstrips available resources. To address this, collaborative approaches, such as international consortia and shared databases, are vital for pooling compute power and overcoming data sovereignty issues, enabling more comprehensive research like that for TB genomes.

Significance (Medium): This highlights that AI's potential, particularly in global health, can only be fully realized through shared resources and cross-border cooperation, mitigating individual limitations.

Sources in support: Graham Brookie (Vice President, ACT)

Neutral sources: Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council)

11. Ranjet & Steven: Navigating Data Residency and Workload Flexibility

Timestamp: 00:45:45 to 00:48:05 - watch this moment on skim

Data residency and localization requirements significantly impact AI workloads across the value chain, from research to commercialization. While decades of considerations existed, these restrictions are now more pervasive, requiring flexibility in setting up systems for local, regional, or global access, influenced by regulations, licensing terms, and contracts.

Significance (High): The increasing pervasiveness of data localization mandates complicates AI deployment, demanding sophisticated infrastructure and policy management to balance compliance with operational needs.

Sources in support: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council)

Neutral sources: Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Ranjit Kublé (Vice President, Trusted AI, Pfizer)

12. Daria: The Ethical Tightrope of Open-Weight Models

Timestamp: 00:51:49 to 00:53:57 - watch this moment on skim

Open-weight AI models, while rapidly advancing, present significant ethical and legal dilemmas due to their nebulous data processing and guardrail implementation. Malicious actors can strip safety features, creating harmful content, which forces organizations like Dreadnote to define clear lines of responsibility for model usage and data governance, especially when self-hosting or using on-prem environments.

Significance (High): This underscores the critical need for robust data security and ethical frameworks that extend beyond model development to encompass the entire lifecycle of AI deployment, particularly with accessible open-weight technologies.

Sources in support: Ranjit Kublé (Vice President, Trusted AI, Pfizer)

Neutral sources: Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council)

13. Model Flexibility and Cost

Timestamp: 00:56:41 to 00:56:53 - watch this moment on skim

Snowflake emphasizes providing customers with model flexibility, allowing them to choose the best-fit AI model for their business and optimize its implementation within the Snowflake ecosystem. This approach acknowledges that different models serve different needs and that cost and performance are key considerations.

Significance (High): Empowers users to select optimal AI solutions, potentially reducing costs and improving performance by aligning models with specific business objectives.

Sources in support: Jenna Ben Yehuda (Executive Vice President, Atlantic Council)

Neutral sources: Graham Brookie (Vice President, ACT), Ranjit Kublé (Vice President, Trusted AI, Pfizer), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council), Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

14. Dreadnought's 'Dread Index' for Security

Timestamp: 00:56:57 to 01:00:02 - watch this moment on skim

Dreadnought offers the 'Dread Index,' an online tool that compares AI models (frontier and open-weight) from different countries based on their offensive security capabilities, cost, and complexity. This helps security practitioners assess risks and understand the cyber capabilities of models being integrated into their ecosystems.

Significance (High): Provides a valuable resource for security professionals to conduct risk assessments and make informed decisions about adopting AI models, enhancing cyber defense strategies.

Sources in support: Graham Brookie (Vice President, ACT)

Neutral sources: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Ranjit Kublé (Vice President, Trusted AI, Pfizer), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council), Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

15. Fit-for-Purpose in Life Sciences

Timestamp: 00:59:50 to 01:00:53 - watch this moment on skim

Rajie states that the ideal in life sciences is a 'fit-for-purpose' AI model, where the choice of model and hosting method offers specific advantages. The field is rapidly evolving, blurring the lines between frontier and open-weight models, necessitating a flexible approach to match applications with the right model-hosting combination.

Significance (High): Emphasizes the need for tailored AI solutions in specialized fields like life sciences, adapting to rapid technological advancements and ensuring optimal application performance.

Sources in support: Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council)

Neutral sources: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Graham Brookie (Vice President, ACT), Ranjit Kublé (Vice President, Trusted AI, Pfizer), Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

16. AI Governance: Agile and Evolving

Timestamp: 01:02:09 to 01:06:01 - watch this moment on skim

Stephen argues that governance frameworks are essential, even if not explicitly visible, and must be flexible to accommodate AI's rapid evolution. Ignoring governance can lead to embedded, unseen issues, while an agile approach, even with exceptions, prioritizes derisking breakthroughs without halting progress.

Significance (High): Highlights the critical need for adaptable governance structures that balance innovation with risk management, preventing potential downstream conflicts and ensuring quality.

Sources in support: Ranjit Kublé (Vice President, Trusted AI, Pfizer)

Neutral sources: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Graham Brookie (Vice President, ACT), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council), Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

17. Daria on Flexible Governance and Risk

Timestamp: 01:06:27 to 01:09:31 - watch this moment on skim

Daria stresses that AI governance must be flexible due to AI's continuous evolution. She advocates for a solution-focused approach to exceptions, prioritizing derisking breakthroughs without slowing them down. This involves understanding what agents have access to and building governance around risk tolerance, citing incidents where models exceeded their tasking.

Significance (High): Underscores the necessity of dynamic governance and proactive risk assessment, ensuring AI systems operate within defined boundaries and safeguarding against unintended consequences.

Sources in support: Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence)

Neutral sources: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Graham Brookie (Vice President, ACT), Ranjit Kublé (Vice President, Trusted AI, Pfizer), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council), Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

18. Interoperability of Healthcare Cohorts

Timestamp: 01:10:29 to 01:13:05 - watch this moment on skim

Jill Alterovich addresses the transportability between healthcare cohorts like the Million Veteran Program (MVP) and the All of Us platform. She notes that while standards exist, differences in population sampling and data types require careful validation and modeling to draw reliable conclusions across datasets.

Significance (High): Clarifies the complexities of cross-cohort research, emphasizing the need for rigorous validation to ensure the accuracy and generalizability of findings derived from disparate data sources.

Sources in support: Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

Neutral sources: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Graham Brookie (Vice President, ACT), Ranjit Kublé (Vice President, Trusted AI, Pfizer), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council), Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence)

19. Ranjit Klé: The Legal Lag

Timestamp: 01:20:05 to 01:21:57 - watch this moment on skim

Legal frameworks are struggling to keep pace with AI development, with precedents dating back to 2002 predating current AI capabilities. This creates a significant challenge in establishing clear lines and consensus between engineers and lawyers.

Significance (High): This legal lag creates uncertainty and potential for misuse, as existing laws may not adequately address novel AI-driven issues. The difficulty in bridging the gap between technical and legal language hinders the development of effective governance.

Sources in support: Graham Brookie (Vice President, ACT)

Neutral sources: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Ranjit Kublé (Vice President, Trusted AI, Pfizer), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council), Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

20. Steven Moon: Guardrails and Communication

Timestamp: 01:22:14 to 01:23:04 - watch this moment on skim

Establishing appropriate guardrails for AI agents, models, and data is essential, but equally concerning is the panic and fear often surrounding AI incidents, which can lead to less fact-based conversations. Effective communication is crucial for managing public perception and preventing uncontrolled escalation.

Significance (High): This point underscores the dual challenge of AI governance: technical safeguards and public understanding. Without a balanced approach to communication, fear can undermine rational policy-making and hinder beneficial AI adoption.

Sources in support: Ranjit Kublé (Vice President, Trusted AI, Pfizer)

Neutral sources: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Graham Brookie (Vice President, ACT), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council), Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence), Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

21. Trey her: Knowledge as Supply Chain

Timestamp: 01:25:34 to 01:26:20 - watch this moment on skim

The 'knowledge' component, including research data and the ability to distill it rapidly with AI, is a critical part of the AI supply chain, akin to physical materials in nuclear technology. This exponentially increases individual capabilities and must be considered in critical infrastructure discussions.

Significance (High): This reframes the security and governance discussion for AI, highlighting that intellectual capital and data accessibility are as vital as physical resources. It implies a need for new policies to manage the dissemination and potential misuse of AI-driven knowledge.

Sources in support: Gil (Associate Professor at Harvard Medical School and President of the Presidential Innovation Fellows Foundation)

Neutral sources: Jenna Ben Yehuda (Executive Vice President, Atlantic Council), Graham Brookie (Vice President, ACT), Ranjit Kublé (Vice President, Trusted AI, Pfizer), Trey Herr (Non-Resident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council), Ranjet (Vice President of Trusted AI, Fizer Digital and Technology AI Center of Excellence)

Industry Providers Shaping AI Policy (Medium severity)

Type: Commercial

Participants from Fizer, Snowflake, and Dreadnote, companies that develop or provide AI solutions, are discussing AI governance and policy. Their commercial interests may influence their perspectives on the ideal regulatory and technical frameworks for AI.

Significance: This raises questions about whether the proposed solutions and policy recommendations are truly in the public interest or if they subtly favor the business models and growth of these AI-centric companies. The audience must consider if the 'art of the possible' demonstrated by industry is being presented without a full accounting of potential downsides or alternative approaches.

Platform Advocacy (Medium severity)

Type: Commercial

Speakers representing companies like Snowflake and Dreadnought naturally advocate for their platforms and solutions, potentially influencing their perspectives on AI model flexibility, governance, and cost-effectiveness.

Significance: This commercial interest could color the discussion, leading to a focus on proprietary advantages or downplaying limitations of their own offerings compared to competitors or open-source alternatives. Audiences should consider these potential biases when evaluating the presented solutions.

Open vs. Frontier Model Debate (Medium severity)

Type: Professional

The discussion around open-source versus frontier (closed) AI models involves differing professional philosophies and potential business models, with some speakers favoring the flexibility of open models and others the perceived security or capability of closed ones.

Significance: This ongoing debate shapes the direction of AI development and adoption. The differing viewpoints highlight the complex strategic decisions organizations face regarding AI investment, security, and innovation, potentially influencing which types of models gain traction.