Grant us Grace and Guide us through these perilous times.
The 15 Diseases of Leadership, According of to Pope Francis
https://hbr.org/2015/04/the-15-diseases-of-leadership-according-to-pope-francis
Archives
All posts for the month August, 2025
Aug 22, 2025, 3:43 AM ET
- South Korea rolled back an initiative to use AI textbooks in classrooms.
- It followed pushback from teachers, who said they didn’t have sufficient preparation.
- While there’s a similar AI push in the US, evidence is lacking on whether it best supports student outcomes.
Humans have revolted against the machine in South Korea — and, in this battle, they’ve won.
Following pushback from teachersand parents, South Korea’s National Assembly on August 4 passed an amendment to an education bill that stripped previously sanctioned AI textbooks of their legal status as official classroom textbooks, and reclassified them as supplementary educational materials.
The Korean Federation of Teachers’ Associations said that while teachers “are not opposed to digital education innovation,” rolling out the textbooks without proper preparation and evaluation actually increased some teachers’ workloads.
The US should take note, said Alex Kotran, the founder and CEO of the AI Education Project, a nonprofit aimed at advancing AI literacy. He said the rollback of AI textbooks and the fact that teachers were involved in the pushback were “totally unsurprising.”
“Research shows that you’re going to get the best outcomes in teacher-centered classrooms, and anything that’s trying to move too quickly, focus on just the technology, without the adequate support for professional learning and development risks undermining that,” Kotran said.
The debate comes as US schools experiment with how best to use AI to fulfill their promise of more personalized learning. The Trump administration supports a public-private approach to increasing the use of the tech in education, but critics maintain that schools should be careful, given the minimal evidence on AI and student achievement, and that teacher training is key.
That’s not to say that there isn’t a place for AI, Kotran said — helping students learn AI skills will equip them for the workforce, where AI is being increasingly used in some fields. But there isn’t extensive evidence that having students learn solely from AI is the best approach.
“The bigger question is, how do you make sure the students are ready to add economic value in the labor market? And it’s not just using AI, it’s actually durable skills like the ability to communicate, problem solve, it’s critical thinking,” Kotran said. “And to build those skills, these are teacher-centered endeavors.”
The role of AI in US education
A survey released by the Korean Federation of Teachers’ Associations in July found that 87.4% of teachers reported a lack of preparation and support for using the textbook materials. The majority of respondents said that they should be allowed to choose how to use the AI textbooks to best suit their needs.
The association added in a press release that it supports efforts to advance AI usage in classrooms, but “we must not be absorbed in introducing technology while ignoring the voices of teachers.”
Some US teachers are concerned. In April, President Donald Trump signed an executive order to establish an AI task force that will establish “public-private partnerships” with AI industry organizations to promote AI literacy in K-12 classrooms. The order also called for government agencies to look into redirecting funding toward AI efforts.
Randi Weingarten, president of the American Federation of Teachers, said in a statement that the order “should be rejected in favor of what the research says works best: investing in classrooms and instruction designed by educators who work directly with students and who have the knowledge and expertise to meet their needs.
Amid concerns about AI adoption, however, some teachers have experienced positive outcomes with incorporating the technology. In an April survey of over 2,000 teachers, Gallup and the Walton Family Foundation found that among the teachers who use AI tools, 64% of respondents said that AI led to higher-quality modifications to student materials, and 61% said it helped them generate better insights on student learning and achievement.
Still, the report said that “no clear consensus exists on whether AI tools should be used in K-12 schools.”
Without comprehensive data on student outcomes using AI, it’s important to approach the topic with a focus on teacher training, not removing teachers from the equation, Kotran said. He added that, at the same time, educators and policymakers need to consider “the freight train that is barreling towards us in terms of job displacement.”
A JPMorgan analyst said there’s an increased risk that AI could replace white-collar jobs, potentially resulting in a “jobless recovery” in which that group is at higher risk of unemployment. Tech leaders are already warning of white-collar job cuts due to AI, and Kotran said the US should take this into account as Gen Zers continue to pursue those careers.
“When it comes to education, the AI just isn’t good enough to replace teachers yet,” Kotran said. “And it’s a bad bet as a school, you’re basically saying, ‘Well, we assume the technology is going to get better and we’re going to somehow be able to get past all of the downside risks of overrelying on AI.’ These are unknown things. It’s a huge, huge risk to take. And if you’re a parent, do you really want to experiment on your kid?”
Article link: https://www.businessinsider.com/ai-in-school-south-korea-textbook-rollback-jobs-education-2025-8
The country poured billions into AI infrastructure, but the data center gold rush is unraveling as speculative investments collide with weak demand and DeepSeek shifts AI trends.
By Caiwei Chen
March 26, 2025
A year or so ago, Xiao Li was seeing floods of Nvidia chip deals on WeChat. A real estate contractor turned data center project manager, he had pivoted to AI infrastructure in 2023, drawn by the promise of China’s AI craze.
At that time, traders in his circle bragged about securing shipments of high-performing Nvidia GPUs that were subject to US export restrictions. Many were smuggled through overseas channels to Shenzhen. At the height of the demand, a single Nvidia H100 chip, a kind that is essential to training AI models, could sell for up to 200,000 yuan ($28,000) on the black market.
Now, his WeChat feed and industry group chats tell a different story. Traders are more discreet in their dealings, and prices have come back down to earth. Meanwhile, two data center projects Li is familiar with are struggling to secure further funding from investors who anticipate poor returns, forcing project leads to sell off surplus GPUs. “It seems like everyone is selling, but few are buying,” he says.
Just months ago, a boom in data center construction was at its height, fueled by both government and private investors. However, many newly built facilities are now sitting empty. According to people on the ground who spoke to MIT Technology Review—including contractors, an executive at a GPU server company, and project managers—most of the companies running these data centers are struggling to stay afloat. The local Chinese outlets Jiazi Guangnianand 36Kr report that up to 80% of China’s newly built computing resources remain unused.
Renting out GPUs to companies that need them for training AI models—the main business model for the new wave of data centers—was once seen as a sure bet. But with the rise of DeepSeek and a sudden change in the economics around AI, the industry is faltering.
“The growing pain China’s AI industry is going through is largely a result of inexperienced players—corporations and local governments—jumping on the hype train, building facilities that aren’t optimal for today’s need,” says Jimmy Goodrich, senior advisor for technology to the RAND Corporation.
The upshot is that projects are failing, energy is being wasted, and data centers have become “distressed assets” whose investors are keen to unload them at below-market rates. The situation may eventually prompt government intervention, he says: “The Chinese government is likely to step in, take over, and hand them off to more capable operators.”
A chaotic building boom
When ChatGPT exploded onto the scene in late 2022, the response in China was swift. The central government designated AI infrastructure as a national priority, urging local governments to accelerate the development of so-called smart computing centers—a term coined to describe AI-focused data centers.
In 2023 and 2024, over 500 new data center projects were announced everywhere from Inner Mongolia to Guangdong, according to KZ Consulting, a market research firm. According to the China Communications Industry Association Data Center Committee, a state-affiliated industry association, at least 150 of the newly built data centers were finished and running by the end of 2024. State-owned enterprises, publicly traded firms, and state-affiliated funds lined up to invest in them, hoping to position themselves as AI front-runners. Local governments heavily promoted them in the hope they’d stimulate the economy and establish their region as a key AI hub.
However, as these costly construction projects continue, the Chinese frenzy over large language models is losing momentum. In 2024 alone, over 144 companies registered with the Cyberspace Administration of China—the country’s central internet regulator—to develop their own LLMs. Yet according to the Economic Observer, a Chinese publication, only about 10% of those companies were still actively investing in large-scale model training by the end of the year.
China’s political system is highly centralized, with local government officials typically moving up the ranks through regional appointments. As a result, many local leaders prioritize short-term economic projects that demonstrate quick results—often to gain favor with higher-ups—rather than long-term development. Large, high-profile infrastructure projects have long been a tool for local officials to boost their political careers.
The post-pandemic economic downturn only intensified this dynamic. With China’s real estate sector—once the backbone of local economies—slumping for the first time in decades, officials scrambled to find alternative growth drivers. In the meantime, the country’s once high-flying internet industry was also entering a period of stagnation. In this vacuum, AI infrastructure became the new stimulus of choice.
“AI felt like a shot of adrenaline,” says Li. “A lot of money that used to flow into real estate is now going into AI data centers.”
By 2023, major corporations—many of them with little prior experience in AI—began partnering with local governments to capitalize on the trend. Some saw AI infrastructure as a way to justify business expansion or boost stock prices, says Fang Cunbao, a data center project manager based in Beijing. Among them were companies like Lotus, an MSG manufacturer, and Jinlun Technology, a textile firm—hardly the names one would associate with cutting-edge AI technology.
This gold-rush approach meant that the push to build AI data centers was largely driven from the top down, often with little regard for actual demand or technical feasibility, say Fang, Li, and multiple on-the-ground sources, who asked to speak anonymously for fear of political repercussions. Many projects were led by executives and investors with limited expertise in AI infrastructure, they say. In the rush to keep up, many were constructed hastily and fell short of industry standards.
“Putting all these large clusters of chips together is a very difficult exercise, and there are very few companies or individuals who know how to do it at scale,” says Goodrich. “This is all really state-of-the-art computer engineering. I’d be surprised if most of these smaller players know how to do it. A lot of the freshly built data centers are quickly strung together and don’t offer the stability that a company like DeepSeek would want.”
To make matters worse, project leaders often relied on middlemen and brokers—some of whom exaggerated demand forecasts or manipulated procurement processes to pocket government subsidies, sources say.
By the end of 2024, the excitement that once surrounded China’s data center boom was curdling into disappointment. The reason is simple: GPU rental is no longer a particularly lucrative business.
The DeepSeek reckoning
The business model of data centers is in theory straightforward: They make money by renting out GPU clusters to companies that need computing capacity for AI training. In reality, however, securing clients is proving difficult. Only a few top tech companies in China are now drawing heavily on computing power to train their AI models. Many smaller players have been giving up on pretraining their models or otherwise shifting their strategy since the rise of DeepSeek, which broke the internet with R1, its open-source reasoning model that matches the performance of ChatGPT o1 but was built at a fraction of its cost.
“DeepSeek is a moment of reckoning for the Chinese AI industry. The burning question shifted from ‘Who can make the best large language model?’ to ‘Who can use them better?’” says Hancheng Cao, an assistant professor of information systems at Emory University.
The rise of reasoning models like DeepSeek’s R1 and OpenAI’s ChatGPT o1 and o3 has also changed what businesses want from a data center. With this technology, most of the computing needs come from conducting step-by-step logical deductions in response to users’ queries, not from the process of training and creating the model in the first place. This reasoning process often yields better results but takes significantly more time. As a result, hardware with low latency (the time it takes for data to pass from one point on a network to another) is paramount. Data centers need to be located near major tech hubs to minimize transmission delays and ensure access to highly skilled operations and maintenance staff.
This change means many data centers built in central, western, and rural China—where electricity and land are cheaper—are losing their allure to AI companies. In Zhengzhou, a city in Li’s home province of Henan, a newly built data center is even distributing free computing vouchers to local tech firms but still struggles to attract clients.
Additionally, a lot of the new data centers that have sprung up in recent years were optimized for pretraining workloads—large, sustained computations run on massive data sets—rather than for inference, the process of running trained reasoning models to respond to user inputs in real time. Inference-friendly hardware differs from what’s traditionally used for large-scale AI training.
GPUs like Nvidia H100 and A100 are designed for massive data processing, prioritizing speed and memory capacity. But as AI moves toward real-time reasoning, the industry seeks chips that are more efficient, responsive, and cost-effective. Even a minor miscalculation in infrastructure needs can render a data center suboptimal for the tasks clients require.
In these circumstances, the GPU rental price has dropped to an all-time low. A recent report from the Chinese media outlet Zhineng Yongxian said that an Nvidia H100 server configured with eight GPUs now rents for 75,000 yuan per month, down from highs of around 180,000. Some data centers would rather leave their facilities sitting empty than run the risk of losing even more money because they are so costly to run, says Fan: “The revenue from having a tiny part of the data center running simply wouldn’t cover the electricity and maintenance cost.”
“It’s paradoxical—China faces the highest acquisition costs for Nvidia chips, yet GPU leasing prices are extraordinarily low,” Li says. There’s an oversupply of computational power, especially in central and west China, but at the same time, there’s a shortage of cutting-edge chips.
However, not all brokers were looking to make money from data centers in the first place. Instead, many were interested in gaming government benefits all along. Some operators exploit the sector for subsidized green electricity, obtaining permits to generate and sell power, according to Fang and some Chinese media reports. Instead of using the energy for AI workloads, they resell it back to the grid at a premium. In other cases, companies acquire land for data center development to qualify for state-backed loans and credits, leaving facilities unused while still benefiting from state funding, according to the local media outlet Jiazi Guangnian.
“Towards the end of 2024, no clear-headed contractor and broker in the market would still go into the business expecting direct profitability,” says Fang. “Everyone I met is leveraging the data center deal for something else the government could offer.”
A necessary evil
Despite the underutilization of data centers, China’s central government is still throwing its weight behind a push for AI infrastructure. In early 2025, it convened an AI industry symposium, emphasizing the importance of self-reliance in this technology.
Major Chinese tech companies are taking note, making investments aligning with this national priority. Alibaba Group announced plans to invest over $50 billion in cloud computing and AI hardware infrastructure over the next three years, while ByteDance plans to invest around $20 billion in GPUs and data centers.
In the meantime, companies in the US are doing likewise. Major tech firms including OpenAI, Softbank, and Oracle have teamed up to commit to the Stargate initiative, which plans to invest up to $500 billion over the next four years to build advanced data centers and computing infrastructure. Given the AI competition between the two countries, experts say that China is unlikely to scale back its efforts. “If generative AI is going to be the killer technology, infrastructure is going to be the determinant of success,” says Goodrich, the tech policy advisor to RAND.
“The Chinese central government will likely see [underused data centers] as a necessary evil to develop an important capability, a growing pain of sorts. You have the failed projects and distressed assets, and the state will consolidate and clean it up. They see the end, not the means,” Goodrich says.
Demand remains strong for Nvidia chips, and especially the H20 chip, which was custom-designed for the Chinese market. One industry source, who requested not to be identified under his company policy, confirmed that the H20, a lighter, faster model optimized for AI inference, is currently the most popular Nvidia chip, followed by the H100, which continues to flow steadily into China even though sales are officially restricted by US sanctions. Some of the new demand is driven by companies deploying their own versions of DeepSeek’s open-source models.
For now, many data centers in China sit in limbo—built for a future that has yet to arrive. Whether they will find a second life remains uncertain. For Fang Cunbao, DeepSeek’s success has become a moment of reckoning, casting doubt on the assumption that an endless expansion of AI infrastructure guarantees progress.
That’s just a myth, he now realizes. At the start of this year, Fang decided to quit the data center industry altogether. “The market is too chaotic. The early adopters profited, but now it’s just people chasing policy loopholes,” he says. He’s decided to go into AI education next.
“What stands between now and a future where AI is actually everywhere,” he says, “is not infrastructure anymore, but solid plans to deploy the technology.”
Fragile Progress, Continuing Disparities
AUTHORS
David C. Radley, Kristen Kolb,Sara R. CollinsDOWNLOADS
RELATED CONTENT
- 2023 Scorecard on State Health System Performance
- 2022 Scorecard on State Health System Performance
- 2024 State Scorecard on Women’s Health and Reproductive Care
- Advancing Racial Equity in U.S. Health Care: The Commonwealth Fund 2024 State Health Disparities Report
Scorecard Highlights
- Topping the 2025 Scorecard’s overall health system rankings are Massachusetts, Hawaii, New Hampshire, Rhode Island, and the District of Columbia, based on 50 measures of health care access and affordability, prevention and treatment, avoidable hospital use and costs, health outcomes and healthy behaviors, income disparity, and equity.
- The lowest-ranked states are Mississippi, Texas, Oklahoma, Arkansas, and West Virginia.
- Uninsured rates fell to record lows in all states by 2023, and differences in health coverage and access to care narrowed between states. These improvements were in all likelihood due to the Affordable Care Act’s coverage expansions, recent state expansions of Medicaid eligibility, and more affordable marketplace plan premiums.
- The number of children receiving all doses of seven recommended early childhood vaccines fell in most states between 2019 and 2023. In five states, including Nebraska and Minnesota, the decline exceeded 10 percent.
- The infant mortality rate (deaths within the first year of life) worsened in 20 states between 2018 and 2022, with considerable variation across states.
- Premature, avoidable deaths vary considerably across states — the rate in West Virginia is more than twice as high as the rate in Massachusetts. Not only are avoidable mortality rates higher in the United States than in other high-income countries, but they are also on the rise, even as they fall elsewhere.
- Wide racial disparities in premature deaths are the norm in most states. In 42 states and D.C., avoidable mortality for Black people is at least two times the rate for the group with the lowest rate.
- When it comes to having affordable health coverage, good-quality care, and the opportunity to live a healthy life, where you live matters in the U.S. Targeted, coordinated federal and state policies are needed to raise health system performance across the nation.
Read the 2025 Full Report: https://www.commonwealthfund.org/publications/scorecard/2025/jun/2025-scorecard-state-health-system-performance?
Models from four major AI firms will be available for immediate testing upon launch. Notably, Elon Musk’s xAI Grok chatbot will not be one of these four.
AUGUST 14, 2025
The General Services Administration will roll out a new governmentwide tool Thursday that gives federal agencies the ability to test major artificial intelligence models, a continuation of Trump administration efforts to ramp up government use of automation.
The AI evaluation suite, titled USAi.gov, will launch later Thursday morning and allow federal agencies to test various AI models, including those from Anthropic, OpenAI, Google and Meta to start, two senior GSA officials told FedScoop.
The launch of USAi underscores the Trump administration’s increasing appetite for AI integration into federal government workspaces. The GSA has described these tools as a way to help federal workers with time-consuming tasks, like document summaries, and give government officials access to some of the country’s leading AI firms.
The GSA, according to one of the officials, will act as a “curator of sorts” for determining which models will be available for testing on USAi. The official noted that additional models are being considered for the platform, with input from GSA’s industry and federal partners, and that American-made models are the primary focus.
Grok, the chatbot made by Elon Musk’s xAI firm, is notably not included on the platform for its launch Thursday. xAI introduced a Grok for Government product last month, days after FedScoop reportedon the GSA’s interest in the chatbot for government use.
FedScoop reported last month that GSA recently registered the domain usai.gov.
How USAi.gov will work
The USAi tool builds upon GSA’s internal chatbot, GSAi, which was rolled out internally in March to give GSA employees access to different enterprise AI models. Zach Whitman, GSA’s chief AI officer and data officer, hinted last month that the GSA was exploring how it could implement its internal AI chatbot in other agencies.
Once an agency tests the model on USAi, it has the option to procure it from the normal federal marketplace, one of the officials said. In other cases, an agency may stay on the USAi platform in the wake of changing market dynamics but can still access the model for testing, the official added.
The platform appears to directly coincide with the GSA’s ongoing rehaul of the federal procurement process, which is focused on consolidating the government’s purchasing of goods and services.
“What we don’t want is to get into this situation where we buy a few licenses for something here and a few licenses for something there, so being able to blanket our entire workforce with the same market-leading capabilities was hugely valuable to us, right off the bat,” Whitman said in an interview with FedScoop about the USAi launch.
GSA has announced a number of new collaborations this month with firms like OpenAI, Anthropic and Box, to offer their products at a significantly discounted price to federal agencies. And FedScoop reported this week that the GSA is considering prioritizing the review of AI technologies in the procurement process.
The USAi launch comes on the heels of the White House’s AI Action Plan, which calls on the GSA to establish an “AI procurement toolbox” to encourage “uniformity across the federal enterprise.” The plan, released last month, mandates that federal agencies guarantee any employee “whose work could benefit” from frontier language models has access to them.
Building trust with models
Whitman said GSA is hopeful federal users will have more trust to work with a platform like USAi, noting public tools on their own can prompt fears around working with sensitive materials.
Dave Shive, GSA’s chief information officer, said in an interview with FedScoop that the agency is “not just prototyping technology.”
“We’re also prototyping new ways to do business and it made a bunch of sense for us to build … a ‘model garden’ — a portfolio of models that our users can choose from, because they all have different strengths and weaknesses,” Shive said. “And those are models across a variety of vendors, because they’re trying to think of new, creative ways to do 21st-century business at GSA.
“And if they have that full suite of models, instead of being limited to just one vendor, it allows them to do that business level, business architecture, prototyping, the very things that we’re all expecting AI can help with,” he added.
In addition to the chatbot and API testing features on USAi, agency administrators can also view GSA’s data-based evaluations for the models to determine which are best for their specific use cases, one of the officials said.
“You can define ‘best’ in any number of ways, from cost implications, from speed implications, from usability implications to bias sensitivity implications,” the other official said, adding that “we have all this kind of decision criteria across a vast number of domains that go into them.”
The GSA said it is offering USAi to all civilian federal agencies, along with the Defense Department. A person familiar with the matter said that as of late Wednesday afternoon, chief AI officers had not yet been briefed about the launch of the USAI.gov platform.
Three evaluations take place prior to a model being available for testing on USAi, one of the officials explained. The first focuses on safety, such as looking at whether a model outputs hate speech, while the second is based on performance at answering questions and the third involves red-teaming, or testing of durability.
The safety teams reviewing the report are specific to USAi, the official noted, emphasizing that this process is not intended to “overstep the role or function of a USAi platform” that welcomes agency input.
Rebecca Heilweil contributed reporting.
Article link: https://fedscoop.com/usai-general-services-administration-artificial-intelligence-google-meta-anthropic-claude/?
Comparing Performance in 10 Nations
AUTHORS
David Blumenthal, Evan D. Gumas,Arnav Shah, Munira Z. Gunja,Reginald D. Williams IIDOWNLOADS
RELATED CONTENT
- International Health Care System Profiles
- High U.S. Health Care Spending: Where Is It All Going?
- U.S. Health Care from a Global Perspective, 2022: Accelerating Spending, Worsening Outcomes
- The Cost of Not Getting Care: Income Disparities in the Affordability of Health Services Across High-Income Countries
- Mirror, Mirror 2021: Reflecting Poorly
- Americans, No Matter the State They Live In, Die Younger Than People in Many Other Countries
Abstract
- Goal: Compare health system performance in 10 countries, including the United States, to glean insights for U.S. improvement.
- Methods: Analysis of 70 health system performance measures in five areas: access to care, care process, administrative efficiency, equity, and health outcomes.
- Key Findings: The top three countries are Australia, the Netherlands, and the United Kingdom, although differences in overall performance between most countries are relatively small. The only clear outlier is the U.S., where health system performance is dramatically lower.
- Conclusion: The U.S. continues to be in a class by itself in the underperformance of its health care sector. While the other nine countries differ in the details of their systems and in their performance on domains, unlike the U.S., they all have found a way to meet their residents’ most basic health care needs, including universal coverage.
SECTIONS
- 01Performance Overview
- 02Access to Care
- 03Care Process
- 04Administrative Efficiency
- 05Equity
- 06Health Outcomes
- 07What the U.S. Can Do to Improve
- 08How We Conducted This Study
- 09How We Measured Performance
Introduction
Mirror, Mirror 2024 is the Commonwealth Fund’s eighth report comparing the performance of health systems in selected countries. Since the first edition in 2004, our goal has remained the same: to highlight lessons from the experiences of these nations, with special attention to how they might inform health system improvement in the United States.
While each country’s health system is unique — evolving over decades, sometimes centuries, in tandem with shifts in political culture, history, and resources — comparisons can offer rich insights to inform policy thinking. Perhaps above all, they can demonstrate the profound impact of national policy choices on a country’s health and well-being.
In this edition of Mirror, Mirror, we compare the health systems of 10 countries: Australia, Canada, France, Germany, the Netherlands, New Zealand, Sweden, Switzerland, the United Kingdom, and the United States. We examine five key domains of health system performance: access to care, care process, administrative efficiency, equity, and health outcomes (each is defined below).
Despite their overall rankings, all the countries have strengths and weaknesses, ranking high on some dimensions and lower on others. No country is at the top or bottom on all areas of performance. Even the top-ranked country — Australia — does less well, for example, on measures of access to care and care process. And even the U.S., with the lowest-ranked health system, ranks second in the care process domain.
Nevertheless, in the aggregate, the nine nations we examined are more alike than different with respect to their higher and lower performance in various domains. But there is one glaring exception — the U.S. (see “How We Conducted This Study”). Especially concerning is the U.S. record on health outcomes, particularly in relation to how much the U.S. spends on health care. The ability to keep people healthy is a critical indicator of a nation’s capacity to achieve equitable growth. In fulfilling this fundamental obligation, the U.S. continues to fail.
PREVIOUS EDITIONS OF MIRROR, MIRROR
Mirror, Mirror 2021: Reflecting Poorly
Mirror, Mirror 2017: International Comparison Reflects Flaws and Opportunities for Better U.S. Health Care
Mirror, Mirror on the Wall, 2014 Update: How the U.S. Health Care System Compares Internationally
How We Measured Performance
Our approach to assessing nations’ health systems mostly resembles recent editions of Mirror, Mirror, involving 70 unique measures in five performance domains. The data sources for our assessments are rich and varied. First, we rely on the unique data collected from international surveys that the Commonwealth Fund conducts in close collaboration with participating countries.1 On a three-year rotating basis, the Fund and its partners survey older adults (age 65 and older), primary care physicians, and the general population (age 18 and older) in each nation. The 2024 edition relies on surveys from 2021, 2022, and 2023.
We also rely on published and unpublished data from cross-national organizations including the World Health Organization (WHO), the Organisation for Economic Co-operation and Development (OECD), and Our World in Data, as well as national data registries and the research literature.
Mirror, Mirror 2024 differs from past reports in certain respects:
- It covers 10 countries instead of the previous 11, after Norway exited the Commonwealth Fund’s international surveys. Norway was the top-ranked country in the 2021 edition of Mirror, Mirror.
- It accounts for the impact of COVID-19 on health system performance, as we are able to use data collected since the onset of the pandemic and do not use data pre-2020.
- It investigates several dimensions of equity. In addition to comparisons between residents with above-average and below-average income, this edition examines health system performance differences based on gender (limited to male and female because of insufficient sample size to include additional gender identities) and location (rural and nonrural) as well as patients’ experiences of discrimination, as reported by physicians. Comparisons of performance with respect to race and ethnicity were not possible because of data limitations: many countries do not collect information on these variables and the constructs of identity vary from country to country. To allow for continuity and comparison with previous editions, we present separate analyses for those based only on income and those based on income, gender, and geography combined. Only the analysis based on income was included in our overall rankings. For further detail, see “How We Conducted This Study.”
Article link: https://www.commonwealthfund.org/publications/fund-reports/2024/sep/mirror-mirror-2024
Anthropic, Google, and others are developing better ways for agents to interact with our programs and each other, but there’s still more work to be done.
August 4, 2025
A growing number of companies are launching AI agents that can do things on your behalf—actions like sending an email, making a document, or editing a database. Initial reviews for these agents have been mixed at best, though, because they struggle to interact with all the different components of our digital lives.
Part of the problem is that we are still building the necessary infrastructure to help agents navigate the world. If we want agents to complete tasks for us, we need to give them the necessary tools while also making sure they use that power responsibly.
Anthropic and Google are among the companies and groups working on exactly that. Over the past year, they have both introduced protocols that try to define how AI agents should interact with each other and the world around them. These protocols could make it easier for agents to control other programs like email clients and note-taking apps.
The reason has to do with application programming interfaces, the connections between computers or programs that govern much of our online world. APIs currently reply to “pings” with standardized information. But AI models aren’t made to work exactly the same every time. The very randomness that helps them come across as conversational and expressive also makes it difficult for them to both call an API and understand the response.
“Models speak a natural language,” says Theo Chu, a project manager at Anthropic. “For [a model] to get context and do something with that context, there is a translation layer that has to happen for it to make sense to the model.” Chu works on one such translation technique, the Model Context Protocol (MCP), which Anthropic introduced at the end of last year.
Related Story
The next big thing is AI tools that can do more complex tasks. Here’s how they will work.
MCP attempts to standardize how AI agents interact with the world via various programs, and it’s already very popular. One web aggregator for MCP servers (essentially, the portals for different programs or tools that agents can access) lists over 15,000 servers already.
Working out how to govern how AI agents interact with each other is arguably an even steeper challenge, and it’s one the Agent2Agent protocol (A2A), introduced by Google in April, tries to take on. Whereas MCP translates requests between words and code, A2A tries to moderate exchanges between agents, which is an “essential next step for the industry to move beyond single-purpose agents,” Rao Surapaneni, who works with A2A at Google Cloud, wrote in an email to MIT Technology Review.
Google says 150 companies have already partnered with it to develop and adopt A2A, including Adobe and Salesforce. At a high level, both MCP and A2A tell an AI agent what it absolutely needs to do, what it should do, and what it should not do to ensure a safe interaction with other services. In a way, they are complementary—each agent in an A2A interaction could individually be using MCP to fetch information the other asks for.
However, Chu stresses that it is “definitely still early days” for MCP, and the A2A road map lists plenty of tasks still to be done. We’ve identified the three main areas of growth for MCP, A2A, and other agent protocols: security, openness, and efficiency.
What should these protocols say about security?
Researchers and developers still don’t really understand how AI models work, and new vulnerabilities are being discovered all the time. For chatbot-style AI applications, malicious attacks can cause models to do all sorts of bad things, including regurgitating training data and spouting slurs. But for AI agents, which interact with the world on someone’s behalf, the possibilities are far riskier.
For example, one AI agent, made to read and send emails for someone, has already been shown to be vulnerable to what’s known as an indirect prompt injection attack. Essentially, an email could be written in a way that hijacksthe AI model and causes it to malfunction. Then, if that agent has access to the user’s files, it could be instructed to send private documents to the attacker.
Some researchers believe that protocols like MCP should prevent agents from carrying out harmful actions like this. However, it does not at the moment. “Basically, it does not have any security design,” says Zhaorun Chen, a University of Chicago PhD student who works on AI agent security and uses MCP servers.
Bruce Schneier, a security researcher and activist, is skeptical that protocols like MCP will be able to do much to reduce the inherent risks that come with AI and is concerned that giving such technology more power will just give it more ability to cause harm in the real, physical world. “We just don’t have good answers on how to secure this stuff,” says Schneier. “It’s going to be a security cesspool really fast.”
Others are more hopeful. Security design could be added to MCP and A2A similar to the way it is for internet protocols like HTTPS (though the nature of attacks on AI systems is very different). And Chen and Anthropic believe that standardizing protocols like MCP and A2A can help make it easier to catch and resolve security issues even as is. Chen uses MCP in his research to test the roles different programs can play in attacks to better understand vulnerabilities. Chu at Anthropic believes that these tools could let cybersecurity companies more easily deal with attacks against agents, because it will be easier to unpack who sent what.
How open should these protocols be?
Although MCP and A2A are two of the most popular agent protocols available today, there are plenty of others in the works. Large companies like Cisco and IBM are working on their own protocols, and other groups have put forth different designs like Agora, designed by researchers at the University of Oxford, which upgrades an agent-service communication from human language to structured data in real time.
Many developers hope there could eventually be a registry of safe, trusted systems to navigate the proliferation of agents and tools. Others, including Chen, want users to be able to rate different services in something like a Yelp for AI agent tools. Some more niche protocols have even built blockchains on top of MCP and A2A so that servers can show they are not just spam.
Both MCP and A2A are open-source, which is common for would-be standards as it lets others work on building them. This can help protocols develop faster and more transparently.
“If we go build something together, we spend less time overall, because we’re not having to each reinvent the wheel,” says David Nalley, who leads developer experience at Amazon Web Services and works with a lot of open-source systems, including A2A and MCP.
Google donated A2A to the Linux Foundation, a nonprofit organization that guides open-source projects, back in June, and Amazon Web Services is now one of the collaborators on the project. With the foundation’s stewardship, the developers who work on A2A (including employees at Google and many others) all get a say in how it should evolve. MCP, on the other hand, is owned by Anthropic and licensed for free. That is a sticking point for some open-source advocates, who want others to have a say in how the code base itself is developed.
“There’s admittedly some increased risk around a single person or a single entity being in absolute control,” says Nalley. He says most people would prefer multiple groups to have a “seat at the table” to make sure that these protocols are serving everyone’s best interests.
However, Nalley does believe Anthropic is acting in good faith—its license, he says, is incredibly permissive, allowing other groups to create their own modified versions of the code (a process known as “forking”).
“Someone could fork it if they needed to, if something went completely off the rails,” says Nalley. IBM’s Agent Communication Protocol was actually spun off of MCP.
Anthropic is still deciding exactly how to develop MCP. For now, it works with a steering committee of outside companies that help make decisions on MCP’s development, but Anthropic seems open to changing this approach. “We are looking to evolve how we think about both ownership and governance in the future,” says Chu.
Is natural language fast enough?
MCP and A2A work on the agents’ terms—they use words and phrases (termed natural language in AI), just as AI models do when they are responding to a person. This is part of the selling point for these protocols, because it means the model doesn’t have to be trained to talk in a way that is unnatural to it. “Allowing a natural-language interface to be used between agents and not just with humans unlocks sharing the intelligence that is built into these agents,” says Surapaneni.
But this choice does come with drawbacks. Natural-language interfaces lack the precision of APIs, and that could result in incorrect responses. And it creates inefficiencies.
Related Story
Are we ready to hand AI agents the keys?
We’re starting to give AI agents real autonomy, and we’re not prepared for what could happen next.
Usually, an AI model reads and responds to text by splitting words into tokens. The AI model will read a prompt, split it into input tokens, generate a response in the form of output tokens, and then put these tokens into words to send back. These tokens define in some sense how much work the AI model has to do—that’s why most AI platforms charge users according to the number of tokens used.
But the whole point of working in tokens is so that people can understand the output—it’s usually faster and more efficient for machine-to-machine communication to just work over code. MCP and A2A both work in natural language, so they require the model to spend tokens as the agent talks to other machines, like tools and other agents. The user never even sees these exchanges—all the effort of making everything human-readable doesn’t ever get read by a human. “You waste a lot of tokens if you want to use MCP,” says Chen.
Chen describes this process as potentially very costly. For example, suppose the user wants the agent to read a document and summarize it. If the agent uses another program to summarize here, it needs to read the document, write the document to the program, read back the summary, and write it back to the user. Since the agent needed to read and write everything, both the document and the summary get doubled up. According to Chen, “It’s actually a lot of tokens.”
As with so many aspects of MCP and A2A’s designs, their benefits also create new challenges. “There’s a long way to go if we want to scale up and actually make them useful,” says Chen.
Correction: This story was updated to clarify Nalley’s involvement with A2A.
Article link: https://www.technologyreview.com/2025/08/04/1120996/protocols-help-agents-navigate-lives-mcp-a2a?
healthcarereimagined 