Ken Perez: Potential downsides of healthcare’s growing dependence on AI

In a recent guest essay for The New York Times, Stanford University senior Theo Baker says, “AI is everything. …. Nearly all of higher education has been overtaken by this technology.”^a

Healthcare is also being overtaken by AI. Hospitals and health systems are rapidly adopting AI to streamline operations and improve clinical decision-making.

According to a March 2024 Microsoft-IDC study, 79% of healthcare organizations are currently using AI technology.^b

In terms of generative AI, a survey of 2,174 non-federal U.S. hospitals conducted in 2024 found that 31.5% reported using it and 24.7% planned to do so in one year, suggesting that by the end of 2025, more than half of non-federal hospitals in the nation would be using generative AI.^c

And a recent analysis found that health systems’ use of AI has advanced beyond early experimentation into mission-critical infrastructure.^d

Consequently, by 2030, 20% to 30% of all healthcare spend will be affected, allocated or directed by AI, and numerous market research firms project compound annual growth rates exceeding 30% for healthcare AI spending over the next 5 to 10 years.^e To put such robust growth in perspective, the Congressional Budget Office projects the nation’s nominal gross domestic product to grow on average by 4% annually through 2036.^f

While AI is unquestionably providing some value to healthcare organizations, as it becomes integral to operations, downside risks are emerging.

Erosion of critical thinking

In Episode 7 of Season 2 of The Pitt, the hospital responds to a ransomware attack by “going analog,” shutting down its computers, internet and electronic health records. This “digital meltdown” immediately causes panic among the younger physicians and nurses in the emergency department (ED), but the most senior leaders, chief attending physician Michael “Robby” Robinavitch (played by Noah Wyle) and charge nurse Dana Evans (played by Katherine LaNasa), quickly direct the staff to switch to whiteboards, paper charts and manual methods to treat patients.

With greater use of (and arguably dependence on) AI, will hospitals and health systems be able to similarly recover from future AI outages?

More fundamentally and arguably of greater concern, with AI being integrated intentionally or unintentionally in the education and training of tomorrow’s healthcare providers, will they be able to exercise good judgment — with or without AI assistance — which is the cornerstone of medicine?

An MIT Media Lab study using electroencephalography technology found that participants who used large language models (LLMs) to compose essays showed markedly lower cognitive activity, less engagement and lower memory recall compared with other participants.^g LLM users consistently underperformed at neural, linguistic and behavioral levels. The researchers concluded, “These results raise concerns about the long-term educational implications of LLM reliance. ….”

Likewise, a recent study by Oregon State University found that tech-savvy students, who are more likely to be excessively dependent on AI, often engage in “cognitive offloading,” where they outsource mental effort to AI instead of engaging with problems and concepts on their own.^h The researchers found among these students a 66% decrease in reflection, a 41% drop in critical thinking and a 21% decline in the perceived need to understand concepts.

Of course, as realistically portrayed by The Pitt, a hospital ED is a high-pressure environment that requires healthcare professionals to engage in critical thinking, exercise sound judgment and use their foundational clinical acumen — especially when system vulnerabilities such as server outages or network downtime occur and AI tools are unavailable.

Geopolitical risk

The 2026 Iran war has brought into sharp relief how supply chain disruptions of essential resources such as oil and natural gas can have massive global ramifications. It has also illustrated how nations with controlling interests or the ability to implement “choke points” can exercise leverage over other countries whose economies depend on certain resources.

AI technology is highly dependent on “rare earths,” a group of 17 elements. For example, neodymium is among the most critical rare earths for AI infrastructure. It is used in the high-speed spinning disks of traditional hard drives and server fans that regulate data center temperatures.

Deng Xiaoping, the leader of the People’s Republic of China from 1978 to 1989, is purported to have said in 1992, “The Middle East has oil, China has rare earths.”ⁱ The report that attributed these words to Deng also cites findings by the International Energy Agency (IEA) that China has 61% of the world’s supply of rare earths and controls 92% of global processing and refining of these elements, while the U.S. Geological Survey has reported that from 2020 to 2023, about 70% of the rare earth compounds and metals in the United States were imported from China.

Clearly, China has a near-monopoly of the rare earths supply chain.

Rare earths have been described as China’s trump card (pun intended), serving as geopolitical bargaining chips. And those chips are becoming more powerful as the U.S. economy becomes increasingly dependent on AI. The result is a potential geopolitical perfect storm.

During the May 14-15 summit in China between U.S. President Donald Trump and President Xi Jinping of China, Xi made it clear that Taiwan is a high priority for his nation. Could China threaten a rare earths embargo of the United States so it can take Taiwan without American intervention? There is a historical precedent for such an action. In 2010, a Chinese fishing trawler operating in disputed waters collided with Japanese Coast Guard patrol boats near the Senkaku Islands. In the ensuing dispute, China imposed a sudden embargo on exports of rare earths to Japan that lasted about two months, as well as strict export quotas that remained in effect until January 2015.^j

And lest one thinks that China’s use of rare earths is from a bygone era, on Jan. 6, China targeted Japan with new export restrictions on dual-use technologies, including rare earths, permanent magnets and other critical minerals in a move widely seen as a response to comments made in November 2025 by Japanese Prime Minister Sanae Takaichi, who stated that an attack on Taiwan by China could constitute an “existential threat” to Japan and justify a military response. ^k

While a rare earths embargo by China would not impact healthcare’s application of AI immediately, a sustained embargo would threaten the implementation of longer-term AI initiatives. In short, healthcare is a caboose on the AI train, vulnerable to upstream supply chain disruptions that could result from geopolitical conflicts.

Hospitals and AI: How to minimize risk

So what can healthcare organizations do to minimize these downside risks?

Just as 100% of U.S. hospitals have backup generators and an emergency plan in the event of power outages, it would be prudent for them to have backup plans so they can continue to operate without access to AI. In the face of the AI revolution, training of healthcare professionals needs to reemphasize the importance of resilience and true human intelligence as characterized by critical thinking and sound judgment.

Given the importance of AI, managing the rare earths geopolitical risk is a national security issue that clearly requires the leadership of the U.S. government and the involvement of the technology sector. To minimize the risk, the United States should follow the Japanese playbook that was produced after the 2010 embargo. It included pursuit of supply chain diversification, reduction or elimination of the use of rare earths in manufacturing processes, recovery of rare earths from used electronics and massive stockpiling. Unfortunately, to date, the United States has not followed that playbook.

Today’s AI euphoria is ubiquitous. And healthcare organizations absolutely should seek to derive value from the careful application of AI. But caveat emptor — dependence on AI poses significant risks for the healthcare workforce as well as the nation. 

Footnotes

---

a. Baker, T., “What AI did to my college class,” The New York Times, May 17, 2026.
b. Dahdah, R., “Microsoft makes the promise of AI in healthcare real through new collaborations with healthcare organizations and partners,” Official Microsoft Blog, March 11, 2024.
c. Everson, J., Nong, P., and Richwine, C., “Uptake of generative AI integrated with electronic health records in U.S. hospitals,” JAMA Network Open, Dec. 12, 2025.
d. Guerra, S., et al. “State of health AI 2026,” Bessemer Venture Partners, Jan. 22, 2026.
e. Kopko, M., and Duque, S., “The coming AI share of Medicare spend,” Pearl, February 2026.
f. CBO, The Budget and Economic Outlook: 2026 to 2036, February 2026.
g. Kosmyna, N., et al., “Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task,” MIT Medi Lab, June 10, 2025.
h. Knapp, D., “Oregon State study raises concerns about AI’s impact on student thinking skills,” KATU 2 ABC,
May 29, 2026.
i. Wei, S.T., “[Big Read] Can China use rare earths to checkmate the US?” ThinkChina, June 18, 2025.
j. Bradsher, K., “How China took over the world’s rare earths industry,”  The Straits Times, April 19, 2025.
k. Baskaran, G., and Schwartz, M., “China’s rare earth campaign against Japan,” Center for Strategic & International Studies, Jan. 13, 2026.