3 ways generative AI is reshaping the patient experience

Since at least the 1970s, artificial intelligence (AI) has been hailed as the solution to some of healthcare’s most intractable problems. So far, the reality has fallen short of the hype.

Generative AI, specifically large language models such as those underpinning ChatGPT, offers a powerful new modality to improve global health and healthcare. The World Economic Forum and ZS interviewed leaders spanning healthcare, technology, academia and beyond to assess generative AI’s transformational potential in healthcare, the most significant adoption barriers and actions stakeholders must take to overcome them.

The findings are the subject of the Forum and ZS joint white paper Patient-First Health With Generative AI: Reshaping the Care Experience (January 2024), which sheds light on the relevant timing of generative AI solutions, their potential and what needs to happen to see generative AI in healthcare deployed successfully.

While generative AI is already being used to discover new drug candidates, revolutionize clinical trials and more, its biggest near-term opportunity may be helping to reimagine patient experience.

The COVID-19 pandemic exacerbated existing healthcare ills, including skyrocketing doctor burnout and increasing financial strain on global health systems. The severe shortage of healthcare workers – the World Health Organization estimates the shortfall at 15 million today, roughly 95% of which is in low- and middle-income countries – poses the greatest risk to global health as uneven access to treatment widens health disparities and adds to the growing burden of care from treating patients with chronic conditions.

Meanwhile, the consumerization of health continues to catalyze healthcare transformation as patients increasingly seek a more active role in managing their health. But this phenomenon poses risks as patients with low health literacy can fall prey to inaccurate or misleading health information online, and increasing use of digital health tools to track and manage health overwhelms doctors’ ability to use new data streams.

Generative AI – with its ability to mimic human speech, process unstructured data (which makes up 80% of all medical information) and solve a wide variety of problems with minimal training – can tackle these challenges by communicating directly with healthcare consumers, thereby shoring up the troubling gap in healthcare workers.

In a survey conducted before ChatGPT became widespread, 94% of US healthcare consumers said they go online for health information. But studies consistently have found that few health websites meet established quality thresholds and that most US healthcare consumers have limited health literacy skills. Large language models trained on high-quality health data in all languages can offer health information that’s correct and accessible, and therefore actionable, for users regardless of their location, education level or cultural context.

Given the healthcare worker shortage, health systems need help shifting resources away from non-urgent patients to those who require timely, high-touch care. Early “virtual care assistants” have failed to scale due to a lack of open-source models for the more than 7,000 languages spoken globally and predictive AI’s limited grasp of nuances across cultural contexts and health system archetypes. With specific and culturally diverse training, large language models can help healthcare providers safely stratify patients entering the healthcare system and route their care accordingly.

In their first year of treatment, up to 60% of chronically ill patients miss doses, take the wrong dosage or abandon treatment altogether, costing health systems hundreds of billions of dollars per year. While traditional algorithms can predict when a patient is likely to drop off treatment, they struggle to suggest (let alone execute) effective interventions to keep patients on track. Pairing classical AI with generative AI can create a powerful tool to manage chronic illness by keeping patients on therapy.

As with classical AI, mistrust is the most formidable barrier to adopting generative AI in healthcare. For reasons that are frequently unclear, large language models occasionally generate incorrect or misleading answers. Meanwhile, insufficient or biased training data can produce dangerously biased outputs, and it’s inherently risky to feed private health data into an AI model.

At the same time, large language models are currently very expensive to train, fine-tune and run, potentially putting them out of reach for many health systems in low- and middle-income countries. If we’re not careful, generative AI risks widening existing health disparities as those in high-income countries get the spoils.

These barriers aren’t insurmountable. There are four steps stakeholders can take to ensure everyone benefits from advances in generative AI.

A person’s health is inherently precious. Any stakeholder developing, deploying or vouching for a tool that could put even one patient’s health at risk should be cautious. However, the risk of medical or ethical malpractice stemming from the responsible use of generative AI pales compared with the benefit of ignoring this modality and its boundless potential for patient impact.

Generative AI could be the most profound technological breakthrough in our lifetimes. However, society must make a concerted effort to put appropriate protections in place to ensure its safe and responsible use. We have the tools to do so, but now we need to summon the collective will.