This story first appeared in China Report, MIT Technology Review’s newsletter about technology in China. Sign up to receive it in your inbox every Tuesday.

Whether you’ve flown a drone before or not, you’ve probably heard of DJI, or at least seen its logo. With more than a 90% share of the global consumer market, this Shenzhen-based company’s drones are used by hobbyists and businesses alike for photography and surveillance, as well as for spraying pesticides, moving parcels, and many other purposes around the world.  

But on June 14, the US House of Representatives passed a bill that would completely ban DJI’s drones from being sold in the US. The bill is now being discussed in the Senate as part of the annual defense budget negotiations. 

The reason? While its market dominance has attracted scrutiny for years, it’s increasingly clear that DJI’s commercial products are so good and affordable they are also being used on active battlefields to scout out the enemy or carry bombs. As the US worries about the potential for conflict between China and Taiwan, the military implications of DJI’s commercial drones are becoming a top policy concern.

DJI has managed to set the gold standard for commercial drones because it is built on decades of electronic manufacturing prowess and policy support in Shenzhen. It is an example of how China’s manufacturing advantage can turn into a technological one.

“I’ve been to the DJI factory many times … and mainly, China’s industrial base is so deep that every component ends up being a fraction of the cost,” Sam Schmitz, the mechanical engineering lead at Neuralink, wrote on X. Shenzhen and surrounding towns have had a robust factory scene for decades, providing an indispensable supply chain for a hardware industry like drones. “This factory made almost everything, and it’s surrounded by thousands of factories that make everything else … nowhere else in the world can you run out of some weird screw and just walk down the street until you find someone selling thousands of them,” he wrote.

But Shenzhen’s municipal government has also significantly contributed to the industry. For example, it has granted companies more permission for potentially risky experiments and set up subsidies and policy support. Last year, I visited Shenzhen to experience how it’s already incorporating drones in everyday food delivery, but the city is also working with companies to use drones for bigger and bigger jobs—carrying everything from packages to passengers. All of these go into a plan to build up the “low-altitude economy” in Shenzhen that keeps the city on the leading edge of drone technology.

As a result, the supply chain in Shenzhen has become so competitive that the world can’t really use drones without it. Chinese drones are simply the most accessible and affordable out there. 

Most recently, DJI’s drones have been used by both sides in the Ukraine-Russia conflict for reconnaissance and bombing. Some American companies tried to replace DJI’s role, but their drones were more expensive and their performance unsatisfactory. And even as DJI publicly suspended its businesses in Russia and Ukraine and said it would terminate any reseller relationship if its products were found to be used for military purposes, the Ukrainian army is still assembling its own drones with parts sourced from China.

This reliance on one Chinese company and the supply chain behind it is what worries US politicians, but the danger would be more pronounced in any conflict between China and Taiwan, a prospect that is a huge security concern in the US and globally.

Last week, my colleague James O’Donnell wrote about a report by the think tank Center for a New American Security (CNAS) that analyzed the role of drones in a potential war in the Taiwan Strait. Right now, both Ukraine and Russia are still finding ways to source drones or drone parts from Chinese companies, but it’d be much harder for Taiwan to do so, since it would be in China’s interest to block its opponent’s supply. “So Taiwan is effectively cut off from the world’s foremost commercial drone supplier and must either make its own drones or find alternative manufacturers, likely in the US,” James wrote.

If the ban on DJI sales in the US is eventually passed, it will hit the company hard for sure, as the US drone market is currently worth an estimated $6 billion, the majority of which is going to DJI. But undercutting DJI’s advantage won’t magically grow an alternative drone industry outside China. 

“The actions taken against DJI suggest protectionism and undermine the principles of fair competition and an open market. The Countering CCP Drones Act risks setting a dangerous precedent, where unfounded allegations dictate public policy, potentially jeopardizing the economic well-being of the US,” DJI told MIT Technology Review in an emailed statement.

The Taiwanese government is aware of the risks of relying too much on China’s drone industry, and it’s looking to change. In March, Taiwan’s newly elected president, Lai Ching-te, said that Taiwan wants to become the “Asian center for the democratic drone supply chain.” 

Already the hub of global semiconductor production, Taiwan seems well positioned to grow another hardware industry like drones, but it will probably still take years or even decades to build the economies of scale seen in Shenzhen. With support from the US, can Taiwanese companies really grow fast enough to meaningfully sway China’s control of the industry? That’s a very open question.

A housekeeping note: I’m currently visiting London, and the newsletter will take a break next week. If you are based in the UK and would like to meet up, let me know by writing to zeyi@technologyreview.com.

---

Now read the rest of China Report

Catch up with China

1. ByteDance is working with the US chip design company Broadcom to develop a five-nanometer AI chip. This US-China collaboration, which should be compliant with US export restrictions, is rare these days given the political climate. (Reuters $)

2. After both the European Union and China announced new tariffs against each other, the two sides agreed to chat about how to resolve the dispute. (New York Times $)

• Canada is preparing to announce its own tariffs on Chinese-made electric vehicles. (Bloomberg $)

3. A NASA leader says the US is “on schedule” to send astronauts to the moon within a few years. There’s currently a heated race between the US and China on moon exploration. (Washington Post $)

4. A new cybersecurity report says RedJuliett, a China-backed hacker group, has intensified attacks on Taiwanese organizations this year. (Al Jazeera $)

5. The Canadian government is blocking a rare earth mine from being sold to a Chinese company. Instead, the government will buy the stockpiled rare earth materials for $2.2 million. (Bloomberg $)

6. Economic hardship at home has pushed some Chinese small investors to enter the US marijuana industry. They have been buying lands in the States, setting up marijuana farms, and hiring other new Chinese immigrants. (NPR)

Lost in translation

In the past week, the most talked-about person in China has been a 17-year-old girl named Jiang Ping, according to the Chinese publication Southern Metropolis Daily. Every year since 2018, the Chinese company Alibaba has been hosting a global mathematics contest that attracts students from prestigious universities around the world to compete for a generous prize. But to everyone’s surprise, Jiang, who’s studying fashion design at a vocational high school in a poor town in eastern China, ended up ranking 12th in the qualifying round this year, beating scores of college undergraduate or even master’s students. Other than reading college mathematics textbooks under her math teacher’s guidance, Jiang has received no professional training, as many of her competitors have.

Jiang’s story, highlighted by Alibaba following the announcement of the first-round results, immediately went viral in China. While some saw it as a tale of buried talents and how personal endeavor can overcome unfavorable circumstances, others questioned the legitimacy of her results. She became so famous that people, including social media influencers, kept visiting her home, turning her hometown into an unlikely tourist destination. The town had to hide Jiang from public attention while she prepared for the final round of the competition.

One more thing

After I wrote about the new Chinese generative video model Kling last week, the AI tool added a new feature that can turn a static photo into a short video clip. Well, what better way to test its performance than feeding it the iconic “distracted boyfriend” meme and watching what the model predicts will happen after that moment?

可灵上线图生视频了，演绎效果很到位！ pic.twitter.com/MgcO3CCl9o

— Gorden Sun (@Gorden_Sun) June 21, 2024

Update: The story has been updated to include a statement from DJI.

The United States has an official web design system and a custom typeface. This public design system aims to make government websites not only good-looking but accessible and functional for all.

Before the internet, Americans may have interacted with the federal government by stepping into grand buildings adorned with impressive stone columns and gleaming marble floors. Today, the neoclassical architecture of those physical spaces has been (at least partially) replaced by the digital architecture of website design—HTML code, tables, forms, and buttons. 

While people visiting a government website to apply for student loans, research veterans’ benefits, or enroll in Medicare might not notice these digital elements, they play a crucial role. If a website is buggy or doesn’t work on a phone, taxpayers may not be able to access the services they have paid for—which can create a negative impression of the government itself.  

There are about 26,000 federal websites in the US. Early on, each site had its own designs, fonts, and log-in systems, creating frustration for the public and wasting government resources. The troubled launch of Healthcare.gov in 2013 highlighted the need for a better way to build government digital services. In 2014, President Obama created two new teams to help improve government tech.

Within the General Services Administration (GSA), a new team called 18F (named for its office at 1800 F Street in Washington, DC) was created to “collaborate with other agencies to fix technical problems, build products, and improve public service through technology.” The team was built to move at the speed of tech startups rather than lumbering bureaucratic agencies. 

The US Digital Service (USDS) was set up “to deliver better government services to the American people through technology and design.” In 2015, the two teams collaborated to build the US Web Design System (USWDS), a style guide and collection of user interface components and design patterns intended to ensure accessibility and a consistent user experience across government websites. “Inconsistency is felt, even if not always precisely articulated in usability research findings,” Dan Williams, the USWDS program lead, said in an email. 

Today, the system defines 47 user interface components such as buttons, alerts, search boxes, and forms, each with design examples, sample code, and guidelines such as “Be polite” and “Don’t overdo it.” Now in its third iteration, it is used in 160 government websites. “As of September 2023, 94 agencies use USWDS code, and it powers about 1.1 billion page views on federal websites,” says Williams.

To ensure clear and consistent typography, the free and open-source typeface Public Sans was created for the US government in 2019. “It started as a design experiment,” says Williams, who designed the typeface. “We were interested in trying to establish an open-source solution space for a typeface, just like we had for the other design elements in the design system.”

The teams behind Public Sans and the USWDS embrace transparency and collaboration with government agencies and the public.

And to ensure that the hard-learned lessons aren’t forgotten, the projects embrace continuous improvement. One of the design principles behind Public Sans offers key guidance in this area: “Strive to be better, not necessarily perfect.”

Jon Keegan writes Beautiful Public Data, a newsletter that curates visually interesting data sets collected by local, state, and federal government agencies
(beautifulpublicdata.com).

The philosopher Karl Popper once argued that there are two kinds of problems in the world: clock problems and cloud problems. As the metaphor suggests, clock problems obey a certain logic. They are orderly and can be broken down and analyzed piece by piece. When a clock stops working, you’re able to take it apart, look for what’s wrong, and fix it. The fix may not be easy, but it’s achievable. Crucially, you know when you’ve solved the issue because the clock starts telling the time again. 

Cloud problems offer no such assurances. They are inherently complex and unpredictable, and they usually have social, psychological, or political dimensions. Because of their dynamic, shape-shifting nature, trying to “fix” a cloud problem often ends up creating several new problems. For this reason, they don’t have a definitive “solved” state—only good and bad (or better and worse) outcomes. Trying to repair a broken-down car is a clock problem. Trying to solve traffic is a cloud problem.  

Engineers are renowned clock-problem solvers. They’re also notorious for treating every problem like a clock. Increasing specialization and cultural expectations play a role in this tendency. But so do engineers themselves, who are typically the ones who get to frame the problems they’re trying to solve in the first place. 

In his latest book, Wicked Problems, Guru Madhavan argues that the growing number of cloudy problems in our world demands a broader, more civic-minded approach to engineering. “Wickedness” is Madhavan’s way of characterizing what he calls “the cloudiest of problems.” It’s a nod to a now-famous coinage by Horst Rittel and Melvin Webber, professors at the University of California, Berkeley, who used the term “wicked” to describe complex social problems that resisted the rote scientific and engineering-based (i.e., clock-like) approaches that were invading their fields of design and urban planning back in the 1970s. 

Madhavan, who’s the senior director of programs at the National Academy of Engineering, is no stranger to wicked problems himself. He’s tackled such daunting examples as trying to make prescription drugs more affordable in the US and prioritizing development of new vaccines. But the book isn’t about his own work. Instead, Wicked Problems weaves together the story of a largely forgotten aviation engineer and inventor, Edwin A. Link, with case studies of man-made and natural disasters that Madhavan uses to explain how wicked problems take shape in society and how they might be tamed.

Link’s story, for those who don’t know it, is fascinating—he was responsible for building the first mechanical flight trainer, using parts from his family’s organ factory—and Madhavan gives a rich and detailed accounting. The challenges this inventor faced in the 1920s and ’30s—which included figuring out how tens of thousands of pilots could quickly and effectively be trained to fly without putting all of them up in the air (and in danger), as well as how to instill trust in “instrument flying” when pilots’ instincts frequently told them their instruments were wrong—were among the quintessential wicked problems of his time. 

To address a world full of wicked problems, we’re going to need a more expansive and inclusive idea of what engineering is and who gets to participate in it.

Unfortunately, while Link’s biography and many of the interstitial chapters on disasters, like Boston’s Great Molasses Flood of 1919, are interesting and deeply researched, Wicked Problems suffers from some wicked structural choices. 

The book’s elaborate conceptual framework and hodgepodge of narratives feel both fussy and unnecessary, making a complex and nuanced topic even more difficult to grasp at times. In the prologue alone, readers must bounce from the concept of cloud problems to that of wicked problems, which get broken down into hard, soft, and messy problems, which are then reconstituted in different ways and linked to six attributes—efficiency, vagueness, vulnerability, safety, maintenance, and resilience—that, together, form what Madhavan calls a “concept of operations,” which is the primary organizational tool he uses to examine wicked problems.

It’s a lot—or at least enough to make you wonder whether a “systems engineering” approach was the correct lens through which to examine wickedness. It’s also unfortunate because Madhavan’s ultimate argument is an important one, particularly in an age of rampant solutionism and “one neat trick” approaches to complex problems. To effectively address a world full of wicked problems, he says, we’re going to need a more expansive and inclusive idea of what engineering is and who gets to participate in it.  

While John Downer would likely agree with that sentiment, his new book, Rational Accidents, makes a strong argument that there are hard limits to even the best and broadest engineering approaches. Similarly set in the world of aviation, Downer’s book explores a fundamental paradox at the heart of today’s civil aviation industry: the fact that flying is safer and more reliable than should technically be possible.

Jetliners are an example of what Downer calls a “catastrophic technology.” These are “complex technological systems that require extraordinary, and historically unprecedented, failure rates—of the order of hundreds of millions, or even billions, of operational hours between catastrophic failures.”

Take the average modern jetliner, with its 7 million components and 170 miles’ worth of wiring—an immensely complex system in and of itself. There were over 25,000 jetliners in regular service in 2014, according to Downer. Together, they averaged 100,000 flights every single day. Now consider that in 2017, no passenger-carrying commercial jetliner was involved in a fatal accident. Zero. That year, passenger totals reached 4 billion on close to 37 million flights. Yes, it was a record-setting year for the airline industry, safety-wise, but flying remains an almost unfathomably safe and reliable mode of transportation—even with Boeing’s deadly 737 Max crashes in 2018 and 2019 and the company’s ongoing troubles. 

Downer, a professor of science and technology studies at the University of Bristol, does an excellent job in the first half of the book dismantling the idea that we can objectively recognize, understand, and therefore control all risk involved in such complex technologies. Using examples from well-known jetliner crashes, as well as from the Fukushima nuclear plant meltdown, he shows why there are simply too many scenarios and permutations of failure for us to assess or foresee such risks, even with today’s sophisticated modeling techniques and algorithmic assistance.

So how does the airline industry achieve its seemingly unachievable record of safety and reliability? It’s not regulation, Downer says. Instead, he points to three unique factors. First is the massive service experience the industry has amassed. Over the course of 70 years, manufacturers have built tens of thousands of jetliners, which have failed (and continue to fail) in all sorts of unpredictable ways. 

This deep and constantly growing data set, combined with the industry’s commitment to thoroughly investigating each and every failure, lets it generalize the lessons learned across the entire industry—the second key to understanding jetliner reliability. 

Finally is what might be the most interesting and counterintuitive factor: Downer argues that the lack of innovation in jetliner design is an essential but overlooked part of the reliability record. The fact that the industry has been building what are essentially iterations of the same jetliner for 70 years ensures that lessons learned from failures are perpetually relevant as well as generalizable, he says. 

That extremely cautious relationship to change flies in the face of the innovate-or-die ethos that drives most technology companies today. And yet it allows the airline industry to learn from decades of failures and continue to chip away at the future “failure performance” of jetliners.

The bad news is that the lessons in jetliner reliability aren’t transferable to other catastrophic technologies. “It is an irony of modernity that the only catastrophic technology with which we have real experience, the jetliner, is highly unrepresentative, and yet it reifies a misleading perception of mastery over catastrophic technologies in general,” writes Downer.

For instance, to make nuclear reactors as reliable as jetliners, that industry would need to commit to one common reactor design, build tens of thousands of reactors, operate them for decades, suffer through thousands of catastrophes, slowly accumulate lessons and insights from those catastrophes, and then use them to refine that common reactor design.  

This obviously won’t happen. And yet “because we remain entranced by the promise of implausible reliability, and implausible certainty about that reliability, our appetite for innovation has outpaced our insight and humility,” writes Downer. With the age of catastrophic technologies still in its infancy, our continued survival may very well hinge not on innovating our way out of cloudy or wicked problems, but rather on recognizing, and respecting, what we don’t know and can probably never understand.  

If Wicked Problems and Rational Accidents are about the challenges and limits of trying to understand complex systems using objective science- and engineering-based methods, Georgina Voss’s new book, Systems Ultra, provides a refreshing alternative. Rather than dispassionately trying to map out or make sense of complex systems from the outside, Voss—a writer, artist, and researcher—uses her book to grapple with what they feel like, and ultimately what they mean, from the inside.

“There is something rather wonderful about simply feeling our way through these enormous structures,” she writes before taking readers on a whirlwind tour of systems visible and unseen, corrupt and benign, ancient and new. Stops include the halls of hype at Las Vegas’s annual Consumer Electronics Show (“a hot mess of a Friday casual hellscape”), the “memetic gold mine” that was the container ship Ever Given and the global supply chain it broke when it got stuck in the Suez Canal, and the payment systems that undergird the porn industry. 

For Voss, systems are both structure and behavior. They are relational technologies that are “defined by their ability to scale and, perhaps more importantly, their peculiar relationship to scale.” She’s also keenly aware of the pitfalls of using an “experiential” approach to make sense of these large-scale systems. “Verbal attempts to neatly encapsulate what a system is can feel like a stoner monologue with pointed hand gestures (‘Have you ever thought about how electricity is, like, really big?’),” she writes. 

Nevertheless, her written attempts are a delight to read. Voss manages to skillfully unpack the power structures that make up, and reinforce, the large-scale systems we live in. Along the way, she also dispels many of the stories we’re told about their inscrutability and inevitability. That she does all this with humor, intelligence, and a boundless sense of curiosity makes Systems Ultra both a shining example of the “civic engagement as engineering” approach that Madhavan argues for in Wicked Problems, and proof that his argument is spot on. 

Bryan Gardiner is a writer based in Oakland, California.

This story first appeared in China Report, MIT Technology Review’s newsletter about technology in China. Sign up to receive it in your inbox every Tuesday.

We finally know the result of a legal case I’ve been tracking in Hong Kong for almost a year. Last week, the Hong Kong Court of Appeal granted an injunction that permits the city government to go to Western platforms like YouTube and Spotify and demand they remove the protest anthem “Glory to Hong Kong,” because the government claims it has been used for sedition.

To read more about how this injunction is specifically designed for Western Big Tech platforms, and the impact it’s likely to have on internet freedom, you can read my story here.

Aside from the depressing implications for pro-democracy movements’ decline in Hong Kong, this lawsuit has also been an interesting case study of the local government’s complicated relationship with internet control and censorship.

I was following this case because it’s a perfect example of how censorship can be built brick by brick. Having reported on China for so long, I sometimes take for granted how powerful and all-encompassing its censorship regime is and need to be reminded that the same can’t be said for most other places in the world.

Hong Kong had a free internet in the past. And unlike mainland China, it remains relatively open: almost all Western platforms and services are still available there, and only a few websites have been censored in recent years. 

Since Hong Kong was returned to China from the UK in 1997, the Chinese central government has clashed several times with local pro-democracy movements asking for universal elections and less influence from Beijing. As a result, it started cementing tighter and tighter control over Hong Kong, and people have been worrying about whether its Great Firewall will eventually extend there. But actually, neither Beijing nor Hong Kong may want to see that happen. All the recent legal maneuverings are only necessary because the government doesn’t want a full-on ban of Western platforms.

When I visited Hong Kong last November, it was pretty clear that both Beijing and Hong Kong want to take advantage of the free flow of finance and business through the city. That’s why the Hong Kong government was given tacit permission in 2023 to explore government cryptocurrency projects, even though crypto trading and mining are illegal in China. Hong Kong officials have boasted on many occasions about the city’s value proposition: connecting untapped demand in the mainland to the wider crypto world by attracting mainland investors and crypto companies to set up shop in Hong Kong. 

But that wouldn’t be possible if Hong Kong closed off its internet. Imagine a “global” crypto industry that couldn’t access Twitter or Discord. Crypto is only one example, but the things that have made Hong Kong successful—the nonstop exchange of cargo, capital, ideas, and people—would cease to function if basic and universal tools like Google or Facebook became unavailable.

That’s why there are these calculated offenses on internet freedom in Hong Kong. It’s about seeking control but also leaving some breathing space; it’s as much about looking tough on the outside as negotiating with platforms down below; it’s about showing its determination to Beijing but also not showing too much aggression to the West. 

For example, the experts I’ve talked to don’t expect the government to request that YouTube remove the videos for everyone globally. More likely, they may ask for the content to be geo-blocked just for users in Hong Kong.

“As long as Hong Kong is still useful as a financial hub, I don’t think they would establish the Great Firewall [there],” says Chung Ching Kwong, a senior analyst at the Inter-Parliamentary Alliance on China, an advocacy organization that connects legislators from over 30 countries working on relations with China. 

It’s also the reason why the Hong Kong government has recently come out to say that it won’t outright ban platforms like Telegram and Signal, even though it said that it had received comments from the public asking it to do so.

But coming back to the court decision to restrict “Glory to Hong Kong,” even if the government doesn’t end up enforcing a full-blown ban of the song, as opposed to the more targeted injunction it’s imposed now, it may still result in significant harm to internet freedom.

We are still watching the responses roll in after the court decision last Wednesday. The Hong Kong government is anxiously waiting to hear how Google will react. Meanwhile, some videos have already been taken down, though it’s unclear whether they were pulled by the creators or by the platform. 

Michael Mo, a former district councilor in Hong Kong who’s now a postgraduate researcher at the University of Leeds in the UK, created a website right after the injunction was first initiated last June to embed all but one of the YouTube videos the government sought to ban. 

The domain name, “gloryto.hk,” was the first test of whether the Hong Kong domain registry would have trouble with it, but nothing has happened to it so far. The second test was seeing how soon the videos would be taken down on YouTube, which is now easy to tell by how many “video unavailable” gaps there are on the page. “Those videos were pretty much intact until the Court of Appeal overturned the rulings of the High Court. The first two have gone,” Mo says. 

The court case is having a chilling effect. Even entities that are not governed by the Hong Kong court are taking precautions. Some YouTube accounts owned by media based in Taiwan and the US proactively enabled geo-blocking to restrict people in Hong Kong from watching clips of the song they uploaded as soon as the injunction application was filed, Mo says. 

Are you optimistic or pessimistic about the future of internet freedom in Hong Kong? Let me know what you think at zeyi@technologyreview.com.

---

Now read the rest of China Report

Catch up with China

1. The Biden administration plans to raise tariffs on Chinese-made EVs, from 25% to 100%. Since few Chinese cars are currently sold in the US, this is mostly a move to deter future imports of Chinese EVs. But it could slow down the decarbonization timeline in the US.  (ABC News)

2. Government officials from the US and China met in Geneva today to discuss how to mitigate the risks of AI. It’s a notable event, given how rare it is for the two sides to find common ground in the highly politicized field of technology. (Reuters $)

3. It will be more expensive soon to ride the bullet trains in China. A 20% to 39% fare increase is causing controversy among Chinese people. (New York Times $)

4. From executive leadership to workplace culture, TikTok has more in common with its Chinese sister app Douyin than the company wants to admit. (Rest of World)

5. China’s most indebted local governments have started claiming troves of data as “intangible assets” on their accounting books. Given the insatiable appetite for AI training data, they may have a point. (South China Morning Post $)

6. A crypto company with Chinese roots purchased a piece of land in Wyoming for crypto mining. Now the Biden administration is blocking the deal for national security reasons. (Associated Press)

Lost in translation

Recently, following an order made by the government, hotels in many major Chinese cities stopped asking guests to submit to facial recognition during check-in. 

According to the Chinese publication TechSina, this has had a devastating impact on the industry of facial recognition hardware. 

As hotels around the country retire their facial recognition kiosks en masse, equipment made by major tech companies has flooded online secondhand markets at steep discounts. What was sold for thousands of dollars is now resold for as little as 1% of the original price. Alipay, the Alibaba-affiliated payment app, once invested hundreds of millions of dollars to research and roll out these kiosks. Now it’s one of the companies being hit the hardest by the policy change.

One more thing

I had to double-check that this is not a joke. It turns out that for the past 10 years, the Louvre museum has been giving visitors a Nintendo 3DS—a popular handheld gaming console—as an audio and visual guide. 

It feels weird seeing people holding a 3DS up to the Mona Lisa as if they were in their own private Pokémon Go–style gaming world rather than just enjoying the museum. But apparently it doesn’t work very well anyway. Oops.

and it was THE WORST at navigating bc a 3ds can’t tell which direction you’re facing + the floorplan isn’t updated to match ongoing renovations. kept tryna send me into a wall i almost chucked the thing i stg

— taylor (@taylorhansss) May 12, 2024

The global race for computational power is well underway, fueled by a worldwide boom in artificial intelligence. OpenAI’s Sam Altman is seeking to raise as much as $7 trillion for a chipmaking venture. Tech giants like Microsoft and Amazon are building AI chips of their own. The need for more computing horsepower to train and use AI models—fueling a quest for everything from cutting-edge chips to giant data sets—isn’t just a current source of geopolitical leverage (as with US curbs on chip exports to China). It is also shaping the way nations will grow and compete in the future, with governments from India to the UK developing national strategies and stockpiling Nvidia graphics processing units. 

I believe it’s high time for America to have its own national compute strategy: an Apollo program for the age of AI.

In January, under President Biden’s executive order on AI, the National Science Foundation launched a pilot program for the National AI Research Resource (NAIRR), envisioned as a “shared research infrastructure” to provide AI computing power, access to open government and nongovernment data sets, and training resources to students and AI researchers. 

The NAIRR pilot, while incredibly important, is just an initial step. The NAIRR Task Force’s final report, published last year, outlined an eventual $2.6 billion budget required to operate the NAIRR over six years. That’s far from enough—and even then, it remains to be seen if Congress will authorize the NAIRR beyond the pilot.

Meanwhile, much more needs to be done to expand the government’s access to computing power and to deploy AI in the nation’s service. Advanced computing is now core to the security and prosperity of our nation; we need it to optimize national intelligence, pursue scientific breakthroughs like fusion reactions, accelerate advanced materials discovery, ensure the cybersecurity of our financial markets and critical infrastructure, and more. The federal government played a pivotal role in enabling the last century’s major technological breakthroughs by providing the core research infrastructure, like particle accelerators for high-energy physics in the 1960s and supercomputing centers in the 1980s. 

Now, with other nations around the world devoting sustained, ambitious government investment to high-performance AI computing, we can’t risk falling behind. It’s a race to power the most world-altering technology in human history. 

First, more dedicated government AI supercomputers need to be built for an array of missions ranging from classified intelligence processing to advanced biological computing. In the modern era, computing capabilities and technical progress have proceeded in lockstep. 

Over the past decade, the US has successfully pushed classic scientific computing into the exascale era with the Frontier, Aurora, and soon-to-arrive El Capitan machines—massive computers that can perform over a quintillion (a billion billion) operations per second. Over the next decade, the power of AI models is projected to increase by a factor of 1,000 to 10,000, and leading compute architectures may be capable of training a 500-trillion-parameter AI model in a week (for comparison, GPT-3 has 175 billion parameters). Supporting research at this scale will require more powerful and dedicated AI research infrastructure, significantly better algorithms, and more investment. 

Although the US currently still has the lead in advanced computing, other countries are nearing parity and set on overtaking us. China, for example, aims to boost its aggregate computing power more than 50% by 2025, and it has been reported that the country plans to have 10 exascale systems by 2025. We cannot risk acting slowly. 

Second, while some may argue for using existing commercial cloud platforms instead of building a high-performance federal computing infrastructure, I believe a hybrid model is necessary. Studies have shown significant long-term cost savings from using federal computing instead of commercial cloud services. In the near term, scaling up cloud computing offers quick, streamlined base-level access for projects—that’s the approach the NAIRR pilot is embracing, with contributions from both industry and federal agencies. In the long run, however, procuring and operating powerful government-owned AI supercomputers with a dedicated mission of supporting US public-sector needs will set the stage for a time when AI is much more ubiquitous and central to our national security and prosperity. 

Such an expanded federal infrastructure can also benefit the public. The life cycle of the government’s computing clusters has traditionally been about seven years, after which new systems are built and old ones decommissioned. Inevitably, as newer cutting-edge GPUs emerge, hardware refreshes will phase out older supercomputers and chips, which can then be recycled for lower-intensity research and nonprofit use—thus adding cost-effective computing resources for civilian purposes. While universities and the private sector have driven most AI progress thus far, a fully distributed model will increasingly face computing constraints as demand soars. In a survey by MIT and the nonprofit US Council on Competitiveness of some of the biggest computing users in the country, 84% of respondents said they faced computation bottlenecks in running key programs. America will need big investments from the federal government to stay ahead.

Third, any national compute strategy must go hand in hand with a talent strategy. The government can better compete with the private sector for AI talent by offering workers an opportunity to tackle national security challenges using world-class computational infrastructure. To ensure that the nation has available a large and sophisticated workforce for these highly technical, specialized roles in developing and implementing AI, America must also recruit and retain the best global students. Crucial to this effort will be creating clear immigration pathways—for example, exempting PhD holders in relevant technical fields from the current H-1B visa cap. We’ll need the brightest minds to fundamentally reimagine how computation takes place and spearhead novel paradigms that can shape AI for the public good, push forward the technology’s boundaries, and deliver its gains to all.

America has long benefitted from its position as the global driver of innovation in advanced computing. Just as the Apollo program galvanized our country to win the space race, setting national ambitions for compute will not just bolster our AI competitiveness in the decades ahead but also drive R&D breakthroughs across practically all sectors with greater access. Advanced computing architecture can’t be erected overnight. Let’s start laying the groundwork now.

Eric Schmidt was the CEO of Google from 2001 to 2011. In 2024, Eric & Wendy co-founded Schmidt Sciences, a philanthropic venture to fund unconventional areas of exploration in science & tech. 

It wasn’t exactly surprising when on Wednesday, May 8, a Hong Kong appeals court sided with the city government to take down “Glory to Hong Kong” from the internet. The trial, in which no one represented the defense, was the culmination of a years-long battle over a song that has become the unofficial anthem for protesters fighting China’s tightening control and police brutality in the city. But it remains an open question how exactly Big Tech will respond. Even as the injunction is narrowly designed to make it easier for them to comply, these Western companies may be seen as aiding authoritarian control and obstructing internet freedom if they do so.  

Google, Apple, Meta, Spotify, and others have spent the last several years largely refusing to cooperate with previous efforts by the Hong Kong government to prevent the spread of the song, which the government has claimed is a threat to national security. But the government has also hesitated to leverage criminal law to force them to comply with requests for removal of content, which could risk international uproar and hurt the city’s economy. 

Now, the new ruling seemingly finds a third option: imposing a civil injunction that doesn’t invoke criminal prosecution, which is similar to how copyright violations are enforced. Theoretically, the platforms may face less reputational blowback when they comply with this court order.

“If you look closely at the judgment, it’s basically tailor-made for the tech companies at stake,” says Chung Ching Kwong, a senior analyst at the Inter-Parliamentary Alliance on China, an advocacy organization that connects legislators from over 30 countries working on relations with China. She believes the language in the judgment suggests the tech companies will now be ready to comply with the government’s request.

A Google spokesperson said the company is reviewing the court’s judgment and didn’t respond to specific questions sent by MIT Technology Review. A Meta spokesperson pointed to a statement from Jeff Paine, the managing director of the Asia Internet Coalition, a trade group representing many tech companies in the Asia-Pacific region: “[The AIC] is assessing the implications of the decision made today, including how the injunction will be implemented, to determine its impact on businesses. We believe that a free and open internet is fundamental to the city’s ambitions to become an international technology and innovation hub.” The AIC did not immediately reply to questions sent via email. Apple and Spotify didn’t immediately respond to requests for comment.

But no matter what these companies do next, the ruling is already having an effect. Just over 24 hours after the court order, some of the 32 YouTube videos that are explicitly targeted in the injunction were inaccessible for users worldwide, not just in Hong Kong. 

While it’s unclear whether the videos were removed by the platform or by their creators, experts say the court decision will almost certainly set a precedent for more content to be censored from Hong Kong’s internet in the future.

“Censorship of the song would be a clear violation of internet freedom and freedom of expression,” says Yaqiu Wang, the research director for China, Hong Kong, and Taiwan at Freedom House, a human rights advocacy group. “Google and other internet companies should use all available channels to challenge the decision.” 

Erasing a song from the internet

Since “Glory to Hong Kong” was first uploaded to YouTube in August 2019 by an anonymous group called Dgx Music, it’s been adored by protesters and applauded as their anthem. Its popularity only grew after China passed the harsh Hong Kong national security law in 2020. 

With lyrics like “Liberate Hong Kong, revolution of our times,” it’s no surprise that it became a major flash point. The city and national Chinese governments were wary of its spread. 

Their fears escalated when the song was repeatedly mistaken for China’s national anthem at international events and was broadcast at sporting events after Hong Kong athletes won. By mid-2023 the mistake, intentional or not, had happened 887 times, according to the Hong Kong government’s request for the content’s removal, which cites YouTube videos and Google search results referring to the song as the “Hong Kong National Anthem” as the reason. 

The government has been arresting people for performing the song on the ground in Hong Kong, but it has been harder to prosecute the online activity since most of the videos and music were uploaded anonymously, and Hong Kong, unlike mainland China, has historically had a free internet. This meant officials needed to explore new approaches to content removal. 

To comply or not to comply

Using the controversial 2020 national security law as legal justification to make requests for removal of certain content that it deems threatening, the Hong Kong government has been able to exert pressure on local companies, like internet service providers. “In Hong Kong, all the major internet service providers are locally owned or Chinese-owned. For business reasons, probably within the last 20 years, most of the foreign investors like Verizon left on their own,” says Charles Mok, a researcher at Stanford University’s Cyber Policy Center and a former legislator in Hong Kong. “So right now, the government is focusing on telling the customer-facing internet service providers to do the blocking.” And it seems to have been somewhat effective, with a few websites for human rights organizations becoming inaccessible locally.

But the city government can’t get its way as easily when the content is on foreign-owned platforms like YouTube or Facebook. Back in 2020, most major Western companies declared they would pause processing data requests from the Hong Kong government while they assessed the law. Over time, some of them have started answering government requests again. But they’ve largely remained firm: over the first six months of 2023, for example, Meta received 41 requests from the Hong Kong government to obtain user data and answered none; during the same period, Google received requests to remove 164 items from Google services and ended up removing 82 of them, according to both companies’ transparency reports. Google specifically mentioned that it chose to not remove two YouTube videos and one Google Drive file related to “Glory to Hong Kong.”

Both sides are in tight spots. Tech companies don’t want to lose the Hong Kong market or endanger their local staff, but they are also worried about being seen as complying with authoritarian government actions. And the Hong Kong government doesn’t want to be seen as openly fighting Western platforms while trust in the region’s financial markets is already in decline. In particular, officials fear international headlines if the government invokes criminal law to force tech companies to remove certain content. 

“I think both sides are navigating this balancing act. So the government finally figured out a way that they thought might be able to solve the impasse: by going to the court and narrowly seeking an injunction,” Mok says.

That happened in June 2023, when Hong Kong’s government requested a court injunction to ban the distribution of the song online with the purpose of “inciting others to commit secession.” It named 32 YouTube videos explicitly, including the original version and live performances, translations into other languages, instrumental and opera versions, and an interview with the original creators. But the order would also cover “any adaptation of the song, the melody and/or lyrics of which are substantially the same as the song,” according to court documents. 

The injunction went through a year of back-and-forth hearings, including a lower court ruling that briefly swatted down the ban. But now, the Court of Appeal has granted the government approval. The case can theoretically be appealed one last time, but with no defendants present, that’s unlikely to happen.

The key difference between this action and previous attempts to remove content is that this is a civil injunction, not a criminal prosecution—meaning it is, at least legally speaking, closer to a copyright takedown request. A platform could arguably be less likely to take a reputational hit if it removes the content upon request. 

Kwong believes this will indeed make platforms more likely to cooperate, and there have already been pretty clear signs to that effect. In one hearing in December, the government was asked by the court to consult online platforms as to the feasibility of the injunction. The final judgment this week says that while the platforms “have not taken part in these proceedings, they have indicated that they are ready to accede to the Government’s request if there is a court order.”

“The actual targets in this case, mainly the tech giants, may have less hesitation to comply with a civil court order than a national security order because if it’s the latter, they may also face backfire from the US,” says Eric Yan-Ho Lai, a research fellow at Georgetown Center for Asian Law. 

Lai also says now that the injunction is granted, it will be easier to prosecute an individual based on violation of a civil injunction rather than prosecuting someone for criminal offenses, since the government won’t need to prove criminal intent.

The chilling effect

Immediately after the injunction, human rights advocates called on tech companies to remain committed to their values. “Companies like Google and Apple have repeatedly claimed that they stand by the universal right to freedom of expression. They should put their ideals into practice,” says Freedom House’s Wang. “Google and other tech companies should thoroughly document government demands, and publish detailed transparency reports on content takedowns, both for those initiated by the authorities and those done by the companies themselves.”

Without making their plans clear, it’s too early to know just how tech companies will react. But right after the injunction was granted, the song largely remained available for Hong Kong users on most platforms, including YouTube, iTunes, and Spotify, according to the South China Morning Post. On iTunes, the song even returned to the top of the download rankings a few hours after the injunction.

One key factor that may still determine corporate cooperation is how far the content removal requests go. There will surely be more videos of the song that are uploaded to YouTube, not to mention independent websites hosting the videos and music for more people to access. Will the government go after each of them too?

The Hong Kong government has previously said in court hearings that it seeks only local restriction of the online content, meaning content will be inaccessible only to users physically in the city. Large platforms like YouTube can do that without difficulty. 

Theoretically, this allows local residents to circumvent the ban by using VPN software, but not everyone is technologically savvy enough to do so. And that wouldn’t do much to minimize the larger chilling effect on free speech, says Kwong from the Inter-Parliamentary Alliance on China. 

“As a Hong Konger living abroad, I do rely on Hong Kong services or international services based in Hong Kong to get ahold of what’s happening in the city. I do use YouTube Hong Kong to see certain things, and I do use Spotify Hong Kong or Apple Music because I want access to Cantopop,” she says. “At the same time, you worry about what you can share with friends in Hong Kong and whatnot. We don’t want to put them into trouble by sharing things that they are not supposed to see, which they should be able to see.”

The court made at least two explicit exemptions to the song’s ban, for “lawful activities conducted in connection with the song, such as those for the purpose of academic activity and news activity.” But even the implementation of these could be incredibly complex and confusing in practice. “In the current political context in Hong Kong, I don’t see anyone willing to take the risk,” Kwong says. 

The government has already arrested prominent journalists on accusations of endangering national security, and a new law passed in 2024 has expanded the crimes that can be prosecuted on national security grounds. As with all efforts to suppress free speech, the impact of vague boundaries that encourage self-censorship on potentially sensitive topics is often sprawling and hard to measure. 

“Nobody knows where the actual red line is,” Kwong says.

This story first appeared in China Report, MIT Technology Review’s newsletter about technology in China. Sign up to receive it in your inbox every Tuesday.

Allow me to indulge in a little reflection this week. Last week, the divest-or-ban TikTok bill was passed in Congress and signed into law. Four years ago, when I was just starting to report on the world of Chinese technologies, one of my first stories was about very similar news: President Donald Trump announcing he’d ban TikTok. 

That 2020 executive order came to nothing in the end—it was blocked in the courts, put aside after the presidency changed hands, and eventually withdrawn by the Biden administration. Yet the idea—that the US government should ban TikTok in some way—never went away. It would repeatedly be suggested in different forms and shapes. And eventually, on April 24, 2024, things came full circle.

A lot has changed in the four years between these two news cycles. Back then, TikTok was a rising sensation that many people didn’t understand; now, it’s one of the biggest social media platforms, the originator of a generation-defining content medium, and a music-industry juggernaut. 

What has also changed is my outlook on the issue. For a long time, I thought TikTok would find a way out of the political tensions, but I’m increasingly pessimistic about its future. And I have even less hope for other Chinese tech companies trying to go global. If the TikTok saga tells us anything, it’s that their Chinese roots will be scrutinized forever, no matter what they do.

I don’t believe TikTok has become a larger security threat now than it was in 2020. There have always been issues with the app, like potential operational influence by the Chinese government, the black-box algorithms that produce unpredictable results, and the fact that parent company ByteDance never managed to separate the US side and the China side cleanly, despite efforts (one called Project Texas) to store and process American data locally. 

But none of those problems got worse over the last four years. And interestingly, while discussions in 2020 still revolved around potential remedies like setting up data centers in the US to store American data or having an organization like Oracle audit operations, those kinds of fixes are not in the law passed this year. As long as it still has Chinese owners, the app is not permissible in the US. The only thing it can do to survive here is transfer ownership to a US entity. 

That’s the cold, hard truth not only for TikTok but for other Chinese companies too. In today’s political climate, any association with China and the Chinese government is seen as unacceptable. It’s a far cry from the 2010s, when Chinese companies could dream about developing a killer app and finding audiences and investors around the globe—something many did pull off. 

There’s something I wrote four years ago that still rings true today: TikTok is the bellwether for Chinese companies trying to go global. 

The majority of Chinese tech giants, like Alibaba, Tencent, and Baidu, operate primarily within China’s borders. TikTok was the first to gain mass popularity in lots of other countries across the world and become part of daily life for people outside China. To many Chinese startups, it showed that the hard work of trying to learn about foreign countries and users can eventually pay off, and it’s worth the time and investment to try.

On the other hand, if even TikTok can’t get itself out of trouble, with all the resources that ByteDance has, is there any hope for the smaller players?

When TikTok found itself in trouble, the initial reaction of these other Chinese companies was to conceal their roots, hoping they could avoid attention. During my reporting, I’ve encountered multiple companies that fret about being described as Chinese. “We are headquartered in Boston,” one would say, while everyone in China openly talked about its product as the overseas version of a Chinese app.

But with all the political back-and-forth about TikTok, I think these companies are also realizing that concealing their Chinese associations doesn’t work—and it may make them look even worse if it leaves users and regulators feeling deceived.

With the new divest-or-ban bill, I think these companies are getting a clear signal that it’s not the technical details that matter—only their national origin. The same worry is spreading to many other industries, as I wrote in this newsletter last week. Even in the climate and renewable power industries, the presence of Chinese companies is becoming increasingly politicized. They, too, are finding themselves scrutinized more for their Chinese roots than for the actual products they offer.

Obviously, none of this is good news to me. When they feel unwelcome in the US market, Chinese companies don’t feel the need to talk to international media anymore. Without these vital conversations, it’s even harder for people in other countries to figure out what’s going on with tech in China.

Instead of banning TikTok because it’s Chinese, maybe we should go back to focus on what TikTok did wrong: why certain sensitive political topics seem deprioritized on the platform; why Project Texas has stalled; how to make the algorithmic workings of the platform more transparent. These issues, instead of whether TikTok is still controlled by China, are the things that actually matter. It’s a harder path to take than just banning the app entirely, but I think it’s the right one.

Do you believe the TikTok ban will go through? Let me know your thoughts at zeyi@technologyreview.com.

---

Now read the rest of China Report

Catch up with China

1. Facing the possibility of a total ban on TikTok, influencers and creators are making contingency plans. (Wired $)

2. TSMC has brought hundreds of Taiwanese employees to Arizona to build its new chip factory. But the company is struggling to bridge cultural and professional differences between American and Taiwanese workers. (Rest of World)

3. The US secretary of state, Antony Blinken, met with Chinese president Xi Jinping during a visit to China this week. (New York Times $)

• Here’s the best way to describe these recent US-China diplomatic meetings: “The US and China talk past each other on most issues, but at least they’re still talking.” (Associated Press)

4. Half of Russian companies’ payments to China are made through middlemen in Hong Kong, Central Asia, or the Middle East to evade sanctions. (Reuters $)

5. A massive auto show is taking place in Beijing this week, with domestic electric vehicles unsurprisingly taking center stage. (Associated Press)

• Meanwhile, Elon Musk squeezed in a quick trip to China and met with his “old friend” the Chinese premier Li Qiang, who was believed to have facilitated establishing the Gigafactory in Shanghai. (BBC)
• Tesla may finally get a license to deploy its autopilot system, which it calls Full Self Driving, in China after agreeing to collaborate with Baidu. (Reuters $)

6. Beijing has hosted two rival Palestinian political groups, Hamas and Fatah, to talk about potential reconciliation. (Al Jazeera)

Lost in translation

The Chinese dubbing community is grappling with the impacts of new audio-generating AI tools. According to the Chinese publication ACGx, for a new audio drama, a music company licensed the voice of the famous dubbing actor Zhao Qianjing and used AI to transform it into multiple characters and voice the entire script. 

But online, this wasn’t really celebrated as an advancement for the industry. Beyond criticizing the quality of the audio drama (saying it still doesn’t sound like real humans), dubbers are worried about the replacement of human actors and increasingly limited opportunities for newcomers. Other than this new audio drama, there have been several examples in China where AI audio generation has been used to replace human dubbers in documentaries and games. E-book platforms have also allowed users to choose different audio-generated voices to read out the text. 

One more thing

While in Beijing, Antony Blinken visited a record store and bought two vinyl records—one by Taylor Swift and another by the Chinese rock star Dou Wei. Many Chinese (and American!) people learned for the first time that Blinken had previously been in a rock band.

One of the formative memories of my youth took place on a camping trip at an Alabama state park. My dad’s friend brought an at-the-time gee-whiz gadget, a portable television, and we used it to watch the very first space shuttle launch from under the loblolly pines. It was thrilling. And it was hard not to believe, watching that shuttle go up (and, a few days later, land), that we were entering an era when travel into the near reaches of space would become common. 

But as it turns out, that’s not the future we built.

This is our Build issue, and although it’s certainly about creating the future we want, in many ways this issue is also about a future that never arrived. Interplanetary space stations. Friendly robots. Even (if you squint and accept a generous definition) terraforming an increasingly uninhabitable Earth. 

Building is a popular tech industry motif—especially in Silicon Valley, where “Time to build” has become something of a call to arms following an influential essay by Marc Andreessen that lamented America’s seeming inability to build just about anything. That essay was published four years ago, at the apex of the country’s disastrous response to covid-19, when masks, PPE, and even hospital beds were in short supply. (As were basic necessities of day-to-day life like eggs, flour, and toilet paper.) It’s an alluring argument. 

Yet the future is built brick by brick from the imperfect decisions we make in the present. We don’t often recognize that the seeming steps forward we are taking today could be seen as steps back in the years to come. This could very well be how we come to view some of the efforts we are making in terms of climate remediation. Xander Peters (accompanied by some incredible photography from Virginia Hanusik) writes about Louisiana’s attempts to protect communities against increased flooding—and wonders if perhaps a managed retreat might not be the better course of action.  

Sometimes the things we don’t do, or the steps we skip, have bigger implications than the actions we do take. For the space program, the decision to race to the moon rather than to first build a way station—as was originally envisioned by some of the pioneers of space travel—may have had the long-term effect of keeping us more earthbound than we might otherwise be. David W. Brown looks at the fallout of those skipped steps and recounts the race to build a new, privately operated space station before the International Space Station comes plummeting back to Earth around 2030. 

Other times, we’re just held back because we haven’t figured out how to do things yet. Simply put: the tech just isn’t quite there. For our cover story on home robots, Melissa Heikkilä looks at how the intersection of robotics and artificial intelligence, and especially large language models, could at last be ushering in the era of helper robots that we’ve been dreaming of since the days of The Jetsons. It’s such a fertile area of development, with action from both big industry incumbents like Google and highly specialized, sometimes secretive startups, that there is far more than we could get into in a single story.

“There was an entire interview with Meta that I didn’t end up using,” Melissa told me. “They have a team working on ‘embodied AI,’ which believes that true general intelligence needs a physical element to it, such as robots or glasses. They’ve built an entire mock apartment in one of their offices, including a full-size living room, kitchen, dining room, and so on, in which they conduct experiments with robots and virtual reality. It’s pretty cool!”

Look for us to keep that reporting going at technologyreview.com. 

And there’s much more, too—including a zinger of a story from Annalee Newitz that takes on the history of brainwashing, a feature on building accountability into police body cameras, and a wild report on designing vegan cheese with generative AI. We hope you find something to take away and build on. 

Thanks for reading,

Mat Honan

The ongoing revolution in artificial intelligence has the potential to dramatically improve our lives—from the way we work to what we do to stay healthy. Yet ensuring that America and other democracies can help shape the trajectory of this technology requires going beyond the tech development taking place at private companies. 

Research at universities drove the AI advances that laid the groundwork for the commercial boom we are experiencing today. Importantly, academia also produced the leaders of pioneering AI companies. 

But today, large foundational models, or LFMs, like ChatGPT, Claude, and Gemini require such vast computational power and such extensive data sets that private companies have replaced academia at the frontier of AI. Empowering our universities to remain alongside them at the forefront of AI research will be key to realizing the field’s long-term potential. This will require correcting the stark asymmetry between academia and industry in access to computing resources.  

Academia’s greatest strength lies in its ability to pursue long-term research projects and fundamental studies that push the boundaries of knowledge. The freedom to explore and experiment with bold, cutting-edge theories will lead to discoveries and innovations that serve as the foundation for future innovation. While tools enabled by LFMs are in everybody’s pocket, there are many questions that need to be answered about them, since they remain a “black box” in many ways. For example, we know AI models have a propensity to hallucinate, but we still don’t fully understand why. 

Because they are insulated from market forces, universities can chart a future where AI truly benefits the many. Expanding academia’s access to resources would foster more inclusive approaches to AI research and its applications. 

The pilot of the National Artificial Intelligence Research Resource (NAIRR), mandated in President Biden’s October 2023 executive order on AI, is a step in the right direction. Through partnerships with the private sector, the NAIRR will create a shared research infrastructure for AI. If it realizes its full potential, it will be an essential hub that helps academic researchers access GPU computational power more effectively. Yet even if the NAIRR is fully funded, its resources are likely to be spread thin. 

This problem could be mitigated if the NAIRR focused on a select number of discrete projects, as some have suggested. But we should also pursue additional creative solutions to get meaningful numbers of GPUs into the hands of academics. Here are a few ideas:

First, we should use large-scale GPU clusters to improve and leverage the supercomputer infrastructure the US government already funds. Academic researchers should be enabled to partner with the US National Labs on grand challenges in AI research. 

Second, the US government should explore ways to reduce the costs of high-end GPUs for academic institutions—for example, by offering financial assistance such as grants or R&D tax credits. Initiatives like New York’s, which make universities key partners with the state in AI development, are already playing an important role at a state level. This model should be emulated across the country. 

Lastly, recent export control restrictions could over time leave some US chipmakers with surplus inventory of leading-edge AI chips. In that case, the government could purchase this surplus and distribute it to universities and academic institutions nationwide.

Imagine the surge of academic AI research and innovation these actions would ignite. Ambitious researchers at universities have a wealth of diverse ideas that are too often stopped short for lack of resources. But supplying universities with adequate computing power will enable their work to complement the research carried out by private industry. Thus equipped, academia can serve as an indispensable hub for technological progress, driving interdisciplinary collaboration, pursuing long-term research, nurturing talent that produces the next generation of AI pioneers, and promoting ethical innovation. 

Historically, similar investments have yielded critical dividends in innovation. The United States of the postwar era cultivated a symbiotic relationship among government, academia, and industry that carried us to the moon, seeded Silicon Valley, and created the internet. 

We need to ensure that academia remains a strong pole in our innovation ecosystem. Investing in its compute capacity is a necessary first step. 

Ylli Bajraktari is CEO of the Special Competitive Studies Project (SCSP), a nonprofit initiative that seeks to strengthen the United States’ long-term competitiveness. 

Tom Mitchell is the Founders University Professor at Carnegie Mellon University. 

Daniela Rus is a professor of electrical engineering and computer science at MIT and director of its Computer Science and Artificial Intelligence Laboratory (CSAIL).