What AI Can Teach Us About Transforming Healthcare

Solving complex challenges

Many Western healthcare systems are nearing a breaking point. In the Netherlands, healthcare spending has increased from 10% of GDP in 2000 to 14.5% in 2021. Reports increasingly warn of a looming "healthcare gridlock" where demand outstrips the system’s ability to deliver care. This raises many questions.

How do we fix it? Are our goals clear? Do incentives drive the right behaviors? Are we systematically learning from experimentation to find what could work at scale? And how can we balance costs, safety, innovation, and outcomes?

It may be time to draw inspiration from an unexpected source: artificial intelligence (AI). Specifically, we can learn from AlphaFold, a groundbreaking AI system that solved one of science’s most complex challenges. What principles from its success can healthcare leaders apply to drive system-wide innovation?

"The question is no longer if AI can help us transform healthcare, but how we can best use the technology and its lessons to ensure better outcomes for everyone."

What is AlphaFold?

AlphaFold, developed by DeepMind, solved a decades-old problem in biology: predicting the 3D structure of proteins from their amino acid sequences. This breakthrough is essential for understanding diseases and developing new treatments.

The AI model followed three core principles:

1. Clear objectives: Predict protein structures accurately.

2. Aligned incentives: Use a reward system to reinforce correct predictions.

3. Optimization under constraints: Operate effectively within limits such as computational power and available data.

Interestingly, Nobel laureate Denis Hassibis, founder of DeepMind, started his career in game design. AlphaGo, the self-taught AI that beat the world’s best player in the most complex game, preceded AlphaFold. Deepmind combined the insights from this field with AI capabilities to solve complex problems.

This approach can enable the 5 Ps: Precise, Personal, Predictive, Preventative, and Pro-active healthcare.

What are the Lessons for Healthcare

If AI can tackle the complexity of protein folding, let’s explore how we may apply similar principles to address healthcare system challenges.

 Clarity of Objectives: While the overarching goal of any healthcare system is straightforward—delivering the best outcomes at the lowest cost—we often lack consistent and transparent methods to measure outcomes.

Intermountain Healthcare in the U.S. has implemented robust systems to measure care outcomes systematically. By tracking and acting on these metrics, they’ve reduced hospital readmission rates and improved the quality of care. Marc Harrison, the former CEO who led this strategy, has single-mindedly focused on optimizing the overall value of the healthcare system.  

Germany’s Martini Klinik has become the world leader in prostate cancer care. In the words of its founder, Hartwig Huland: “Our recipe for success is no secret – quite the opposite. Ever since our clinic was established, we have adhered to the basic principle of optimizing outcome quality and remain firmly convinced that this overarching goal should guide the actions of every medical facility.”

Systematic outcome measurement and the insight that can be gleaned trigger continuous improvement and innovation. It creates transparency, fosters trust, and helps healthcare organizations identify and replicate success.

Aligning Incentives with Desired Outcomes: AlphaFold’s rapid learning stemmed from its reward system, which incentivized correct predictions. In contrast, healthcare systems often reward providers for production (e.g., the number of consults, tests, and procedures performed) rather than the quality of outcomes.

 The UK NHS (National Health Service) has experimented with outcomes-based payment models. Providers are rewarded for meeting specific health goals, such as reducing complications or improving patient recovery times.

Shifting to value-based reward systems can realign incentives, motivating providers to focus on the quality and efficiency of outcomes rather than volume. These systems could foster innovation by rewarding approaches that optimize care delivery within the resource, regulatory, and ethical constraints.

Embracing Constraints and Trade-offs: AlphaFold operated under strict limits—finite computational power and data—forcing it to balance speed and accuracy. Healthcare systems face similar constraints, including limited budgets, staff, and time. However, healthcare often fails to view these constraints as opportunities for innovation.

The Cleveland Clinic has used operational digital twins—virtual replicas of workflows—to optimize patient flow and reduce wait times. They've achieved measurable efficiency gains by simulating resource allocation and care delivery.

Embracing constraints encourages creativity and forces organizations to optimize decision-making. Digital tools like simulation models can help healthcare leaders test and refine interventions before deploying them in real-world settings. Not all constraints are fixed; some are artificial and can be stretched. A clear understanding of those constraints can help identify the system boundaries.

Broader Applications Beyond Healthcare

These approaches aren’t limited to healthcare. They apply to any complex system with competing priorities and constraints.

Education institutions can measure student learning outcomes systematically, reward educators for closing gaps, and use digital twins to design, simulate, and test teaching strategies. Before launching yet another law or regulation, governments can use simulation models to optimize urban planning, resource allocation, or policy decisions within limited budgets.

Leaders in these sectors can learn how complex problems can be solved through clear goals, proper incentives, and constraints. They can then explore, simulate, and test new ways to tackle longstanding challenges.

AlphaFold’s rapid iterative improvement relied on transparent, measurable results. Healthcare leaders should adopt this approach by tracking and publishing patient outcomes. For example, Intermountain Healthcare’s transparency in outcome measurement has encouraged best practices and fostered accountability.

DeepMind simulated millions of protein structures to identify patterns and improve predictions. Similarly, healthcare systems can use simulations to evaluate the impact of interventions. Digital twins of hospitals or care systems can model patient flow, resource allocation, and treatment scenarios. Virtual patient models can help personalize treatments and predict outcomes before applying them in real life.

Personalize Risk Analysis and Prevention: AlphaFold relied on detailed data to predict protein structures. To create personalized patient risk profiles, healthcare systems can use granular diagnostic tests and monitoring data. This approach can enable the 5 Ps: Precise, Personal, Predictive, Preventative, and Pro-active healthcare.

Designing Reward Systems for Value: AlphaFold optimized rewards within its constraints, driving continuous improvement. Similarly, healthcare reward systems must evolve to prioritize outcomes over inputs. Providers could be incentivized not just to deliver care but to efficiently provide the best patient outcomes within constraints such as time, resources, and patient safety.

This approach acknowledges the trade-offs inherent in any complex system and encourages innovation by embracing limitations.

AI as a Playbook for Healthcare Transformation

The example shows us that solving complex problems requires more than raw computational power. It demands clear goals, aligned incentives, and the ability to learn iteratively within constraints. By adopting these principles, healthcare leaders can reimagine systems and tackle the challenges of delivering better, more sustainable care.

The question is no longer if AI can help us transform healthcare, but how we can best use the technology and its lessons to ensure better outcomes for everyone.