I tested out a buzzy new text-to-video AI model from China

You may not be familiar with Kuaishou, but this Chinese company just hit a major milestone: It’s released the first ever text-to-video generative AI model that’s freely available for the public to test.

The short-video platform, which has over 600 million active users, announced the new tool, called Kling, on June 6. Like OpenAI’s Sora model, Kling is able to generate videos up to two minutes long from prompts.

But unlike Sora, which still remains inaccessible to the public four months after OpenAI debuted it, Kling has already started letting people try the model themselves. Zeyi Yang, our China reporter, has been putting it through its paces. Here’s what he made of it.

This story is from China Report, our weekly newsletter covering tech in China. Sign up to receive it in your inbox every Tuesday.

Meta has created a way to watermark AI-generated speech

The news: Meta has created a system that can embed hidden signals, known as watermarks, in AI-generated audio clips, which could help in detecting AI-generated content online. 

Why it matters: The tool, called AudioSeal, is the first that can pinpoint which bits of audio in, for example, a full hour-long podcast might have been generated by AI. It could help to tackle the growing problem of misinformation and scams using voice cloning tools. Read the full story.

—Melissa Heikkilä

The return of pneumatic tubes

Pneumatic tubes were once touted as something that would revolutionize the world. In science fiction, they were envisioned as a fundamental part of the future—even in dystopias like George Orwell’s 1984, where they help to deliver orders for the main character, Winston Smith, in his job rewriting history to fit the ruling party’s changing narrative. 

In real life, the tubes were expected to transform several industries in the late 19th century through the mid-20th. The technology involves moving a cylindrical carrier or capsule through a series of tubes with the aid of a blower that pushes or pulls it into motion, and for a while, the United States took up the systems with gusto.

But by the mid to late 20th century, use of the technology had largely fallen by the wayside, and pneumatic tube technology became virtually obsolete. Except in hospitals. Read the full story.

—Vanessa Armstrong

This story is from the forthcoming print issue of MIT Technology Review, which explores the theme of Play. It’s set to go live on Wednesday June 26, so if you don’t already, subscribe now to get a copy when it lands.

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 Nvidia has become the world’s most valuable company 
Leapfrogging Microsoft and Apple thanks to the AI boom. (BBC)
+ Nvidia’s meteoric rise echoes the dot com boom. (WSJ $)
+ CEO Jensen Huang is now one of the richest people in the world. (Forbes)
+ The firm is worth more than China’s entire agricultural industry. (NY Mag $)
+ What’s next in chips. (MIT Technology Review)

2 TikTok is introducing AI avatars for ads
Which seems like a slippery slope. (404 Media)
+ India’s farmers are getting their news from AI news anchors. (Bloomberg $)
+ Deepfakes of Chinese influencers are livestreaming 24/7. (MIT Technology Review)

3 Boeing’s Starliner spacecraft will stay in space for a little longer
Officials need to troubleshoot some issues before it can head back to Earth. (WP $)

4 STEM students are refusing to work at Amazon and Google
Until the companies end their involvement with Project Nimbus. (Wired $)

5 Google isn’t what it used to be
But is Reddit really a viable alternative? (WSJ $)
+ Why Google’s AI Overviews gets things wrong. (MIT Technology Review)

6 A security bug allows anyone to impersonate Microsoft corporate email accounts
It’s making it harder to spot phishing attacks. (TechCrunch)

7 How deep sea exploration has changed since the Titan disaster
Robots are taking humans’ place to plumb the depths. (NYT $)
+ Meet the divers trying to figure out how deep humans can go. (MIT Technology Review)

8 How the free streaming service Tubi took over the US
Its secret weapon? Old movies.(The Guardian)

9 A new AI video tool instantly started ripping off Disney
Raising some serious questions about what the model had been trained on. (The Verge)
+ What’s next for generative video. (MIT Technology Review)

10 Apple appears to have paused work on the next Vision Pro
Things aren’t looking too bright for the high-end headset. (The Information $)
+ These minuscule pixels are poised to take augmented reality by storm. (MIT Technology Review)

Quote of the day

“He’s like Taylor Swift, but for tech.”

—Mark Zuckerberg is suitably dazzled by Nvidia CEO Jensen Huang’s starpower, the Information reports.

The big story

How sounds can turn us on to the wonders of the universe

June 2023

Astronomy should, in principle, be a welcoming field for blind researchers. But across the board, science is full of charts, graphs, databases, and images that are designed to be seen.

So researcher Sarah Kane, who is legally blind, was thrilled three years ago when she encountered a technology known as sonification, designed to transform information into sound. Since then she’s been working with a project called Astronify, which presents astronomical information in audio form.

For millions of blind and visually impaired people, sonification could be transformative—opening access to education, to once unimaginable careers, and even to the secrets of the universe. Read the full story.

—Corey S. Powell

We can still have nice things

A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or tweet ’em at me.)

+ Clearing a pool table in 28 seconds? Don’t mind if I do.
+ As summer gets truly underway, it’s time to reorganize your closet.
+ Check out the winner’s of this year’s Food Photographer of the Year awards.
+ If you’re obsessed with the viral Steam game Banana, you’re far from alone.

ALAMY

In real life, the tubes were expected to transform several industries in the late 19th century through the mid-20th. “The possibilities of compressed air are not fully realized in this country,” declared an 1890 article in the New York Tribune. “The pneumatic tube system of communication is, of course, in use in many of the downtown stores, in newspaper offices […] but there exists a great deal of ignorance about the use of compressed air, even among engineering experts.”

Pneumatic tube technology involves moving a cylindrical carrier or capsule through a series of tubes with the aid of a blower that pushes or pulls it into motion. For a while, the United States took up the systems with gusto. Retail stores and banks were especially interested in their potential to move money more efficiently: “Besides this saving of time to the customer the store is relieved of all the annoying bustle and confusion of boys running for cash on the various retail floors,” one 1882 article in the Boston Globe reported. The benefit to the owner, of course, was reduced labor costs, with tube manufacturers claiming that stores would see a return on their investment within a year.  

“The motto of the company is to substitute machines for men and for children as carriers, in every possible way,” a 1914 Boston Globe article said about Lamson Service, one of the largest proprietors of tubes at the time, adding, “[President] Emeritus Charles W. Eliot of Harvard says: ‘No man should be employed at a task which a machine can perform,’ and the Lamson Company supplements that statement by this: ‘Because it doesn’t pay.’”

By 1912, Lamson had over 60,000 customers globally in sectors including retail, banks, insurance offices, courtrooms, libraries, hotels, and industrial plants. The postal service in cities such as Boston, Philadelphia, Chicago, and New York also used tubes to deliver the mail, with at least 45 miles of Lamson tubing in place by 1912.

On the transportation front, New York City’s first attempt at a subway system, in 1870, also ran on a pneumatic system, and the idea of using tubes to move people continues to beguile innovators to this day. (See Elon Musk’s largely abandoned Hyperloop concept of the 2010s.)

But by the mid to late 20th century, use of the technology had largely fallen by the wayside. It had become cheaper to transport mail by truck than by tube, and as transactions moved to credit cards, there was less demand to make change for cash payments. Electrical rail won out over compressed air, paper records and files disappeared in the wake of digitization, and tubes at bank drive-throughs started being replaced by ATMs, while only a fraction of pharmacies used them for their own such services. Pneumatic tube technology became virtually obsolete.

Except in hospitals. 

“A pneumatic tube system today for a new hospital that’s being built is ubiquitous. It’s like putting a washing machine or a central AC system in a new home. It just makes too much sense to not do it,” says Cory Kwarta, CEO of Swisslog Healthcare, a corporation that—under its TransLogic company—has provided pneumatic tube systems in health-care facilities for over 50 years. And while the sophistication of these systems has changed over time, the fundamental technology of using pneumatic force to move a capsule from one destination to another has remained the same. 

By the turn of the 20th century, health care had become a more scientific endeavor, and different spaces within a hospital were designated for new technologies (like x-rays) or specific procedures (like surgeries). “Instead of having patients in one place, with the doctors and the nurses and everything coming to them, and it’s all happening in the ward, [hospitals] became a bunch of different parts that each had a role,” explains Jeanne Kisacky, an architectural historian who wrote Rise of the Modern Hospital: An Architectural History of Health and Healing, 1870–1940. 

Designating different parts of a building for different medical specialties and services, like specimen analysis, also increased the physical footprint of health-care facilities. The result was that nurses and doctors had to spend much of their days moving from one department to another, which was an inefficient use of their time. Pneumatic tube technology provided a solution.

By the 1920s, more and more hospitals started installing tube systems. At first, the capsules primarily moved medical records, prescription orders, and items like money and receipts—similar cargo to what was moved around in banks and retail stores at the time. As early as 1927, however, the systems were also marketed to hospitals as a way to transfer specimens to a central laboratory for analysis. 

By the 1960s, pneumatic tubes were becoming standard in health care. As a hospital administrator explained in the January 1960 issue of Modern Hospital, “We are now getting eight hours’ worth of service per day from each nurse, where previously we had been getting about six hours of nursing plus two hours of errand running.”

As computers and credit cards started to become more prevalent in the 1980s, reducing paperwork significantly, the systems shifted to mostly carrying lab specimens, pharmaceuticals, and blood products. Today, lab specimens are roughly 60% of what hospital tube systems carry; pharmaceuticals account for 30%, and blood products for phlebotomy make up 5%.

The carriers or capsules, which can hold up to five pounds, move through piping six inches in diameter—just big enough to hold a 2,000-milliliter IV bag—at speeds of 18 to 24 feet per second, or roughly 12 to 16 miles per hour. The carriers are limited to those speeds to maintain specimen integrity. If blood samples move faster, for example, blood cells can be destroyed.

The pneumatic systems have also gone through major changes in structure in recent years, evolving from fixed routes to networked systems. “It’s like a train system, and you’re on one track and now you have to go to another track,” says Steve Dahl, an executive vice president at Pevco, a manufacturer of these systems.

GETTY IMAGES

Manufacturers try to get involved early in the hospital design process, says Swisslog’s Kwarta, so “we can talk to the clinical users and say, ‘Hey, what kind of contents do you anticipate sending through this pneumatic tube system, based on your bed count, based on your patient census, and from where and to where do these specimens or materials need to go?’”

Penn Medicine’s University City Medical District in Philadelphia opened up the state-of-the-art Pavilion in 2021. It has three pneumatic systems: the main one is for items directly related to health care, like specimens, and two separate ones handle linen and trash. The main system runs over 12 miles of pipe and completes more than 6,000 transactions on an average day. Sending a capsule between the two farthest points of the system—a distance of multiple city blocks—takes just under five minutes. Walking that distance would take around 20 minutes, not including getting to the floor where the item needs to go. 

Michigan Medicine has a system dedicated solely for use in nuclear medicine, which relies on radioactive materials for treatment. Getting the materials where they need to go is a five- to eight-minute walk—too long given their short shelf life. With the tubes, it gets there—in a lead-lined capsule—in less than a minute. 

Steven Fox, who leads the electrical engineering team for the pneumatic tubes at Michigan Medicine, describes the scale of the materials his system moves in terms of African elephants, which weigh about six tons. “We try to keep [a carrier’s] load to five pounds apiece,” he says. “So we could probably transport about 30,000 pounds per day. That’s two and a half African elephants that we transport from one side of the hospital to the other every day.”

The equipment to maintain these labyrinthian highways is vast. Michigan and Penn have between 150 and 200 stations where doctors, nurses, and technicians can pick up a capsule or send one off. Keeping those systems moving also requires around 30 blowers and over 150 transfer units to shift carriers to different tube lines as needed. At Michigan Medicine, moving an item from one end of the system to another requires 20 to 25 pieces of equipment.

Before the turn of the century, triggering the blower to move a capsule from point A to point B would be accomplished by someone turning or pressing an electronic or magnetic switch. In the 2000s, technicians managed the systems on DOS; these days, the latest systems run on programs that monitor every capsule in real time and allow adjustments based on the level of traffic, the priority level of a capsule, and the demand for additional carriers. The systems run 24 hours a day, every day. 

“We treat [the tube system] no different than electricity, steam, water, gas. It’s a utility,” says Frank Connelly, an assistant hospital director at Penn. “Without that, you can’t provide services to people that need it in a hospital.”

“You’re nervous—you just got blood taken,” he continues. “‘How long is it going to be before I get my results back?’ Imagine if they had to wait all that extra time because you’re not sending one person for every vial—they’re going to wait awhile until they get a basket full and then walk to the lab. Nowadays they fill up the tube and send it to the lab. And I think that helps patient care.” 

Vanessa Armstrong is a freelance writer whose work has appeared in the New York Times, Atlas Obscura, Travel + Leisure, and elsewhere. 

You may not be familiar with Kuaishou, but this Chinese company just hit a major milestone: It’s released the first text-to-video generative AI model that’s freely available for the public to test.

The short-video platform, which has over 600 million active users, announced the new tool on June 6. It’s called Kling. Like OpenAI’s Sora model, Kling is able to generate videos “up to two minutes long with a frame rate of 30fps and video resolution up to 1080p,” the company says on its website.

But unlike Sora, which still remains inaccessible to the public four months after OpenAI trialed it, Kling soon started letting people try the model themselves. 

I was one of them. I got access to it after downloading Kuaishou’s video-editing tool, signing up with a Chinese number, getting on a waitlist, and filling out an additional form through Kuaishou’s user feedback groups. The model can’t process prompts written entirely in English, but you can get around that by either translating the phrase you want to use into Chinese or including one or two Chinese words.

So, first things first. Here are a few results I generated with Kling to show you what it’s like. Remember Sora’s impressive demo video of Tokyo’s street scenes or the cat darting through a garden? Here are Kling’s takes:

Remember the image of Dall-E’s horse-riding astronaut? I asked Kling to generate a video version too. 

There are a few things worth applauding here. None of these videos deviates from the prompt much, and the physics seem right—the panning of the camera, the ruffling leaves, and the way the horse and astronaut turn, showing Earth behind them. The generation process took around three minutes for each of them. Not the fastest, but totally acceptable. 

But there are obvious shortcomings, too. The videos, while 720p in format, seem blurry and grainy; sometimes Kling ignores a major request in the prompt; and most important, all videos generated now are capped at five seconds long, which makes them far less dynamic or complex.

However, it’s not really fair to compare these results with things like Sora’s demos, which are hand-picked by OpenAI to release to the public and probably represent better-than-average results. These Kling videos are from the first attempts I had with each prompt, and I rarely included prompt-engineering keywords like “8k, photorealism” to fine-tune the results. 

If you want to see more Kling-generated videos, check out this handy collection put together by an open-source AI community in China, which includes both impressive results and all kinds of failures.

Kling’s general capabilities are good enough, says Guizang, an AI artist in Beijing who has been testing out the model since its release and has compiled a series of direct comparisons between Sora and Kling. Kling’s disadvantage lies in the aesthetics of the results, he says, like the composition or the color grading. “But that’s not a big issue. That can be fixed quickly,” Guizang, who wished to be identified only by his online alias, tells MIT Technology Review. 

“The core capability of a model is in how it simulates physics and real natural environments,” and he says Kling does well in that regard.

Kling works in a similar way to Sora: it combines the diffusion models traditionally used in video-generation AIs with a transformer architecture, which helps it understand larger video data files and generate results more efficiently.

But Kling may have a key advantage over Sora: Kuaishou, the most prominent rival to Douyin in China, has a massive video platform with hundreds of millions of users who have collectively uploaded an incredibly big trove of video data that could be used to train it. Kuaishou told MIT Technology Review in a statement that “Kling uses publicly available data from the global internet for model training, in accordance with industry standards.” However, the company didn’t elaborate on the specifics of the training data(neither did OpenAI about Sora, which has led to concerns about intellectual-property protections).

After testing the model, I feel the biggest limitation to Kling’s usefulness is that it only generates five-second-long videos.

“The longer a video is, the more likely it will hallucinate or generate inconsistent results,” says Shen Yang, a professor studying AI and media at Tsinghua University in Beijing. That limitation means the technology will leave a larger impact on the short-video industry than it does on the movie industry, he says. 

Short, vertical videos (those designed for viewing on phones) usually grab the attention of viewers in a few seconds. Shen says Chinese TikTok-like platforms often assess whether a video is successful by how many people would watch through the first three or five seconds before they scroll away—so an AI-generated high-quality video clip that’s just five seconds long could be a game-changer for short-video creators. 

Guizang agrees that AI could disrupt the content-creating scene for short-form videos. It will benefit creators in the short term as a productivity tool; but in the long run, he worries that platforms like Kuaishou and Douyin could take over the production of videos and directly generate content customized for users, reducing the platforms’ reliance on star creators.

It might still take quite some time for the technology to advance to that level, but the field of text-to-video tools is getting much more buzzy now. One week after Kling’s release, a California-based startup called Luma AI also released a similar model for public usage. Runway, a celebrity startup in video generation, has teased a significant update that will make its model much more powerful. ByteDance, Kuaishou’s biggest rival, is also reportedly working on the release of its generative video tool soon. “By the end of this year, we will have a lot of options available to us,” Guizang says.

I asked Kling to generate what society looks like when “anyone can quickly generate a video clip based on their own needs.” And here’s what it gave me. Impressive hands, but you didn’t answer the question—sorry.

Do you have a prompt you want to see generated with Kling? Send it to zeyi@technologyreview.com and I’ll send you back the result. The prompt has to be less than 200 characters long, and preferably written in Chinese.

Now read the rest of China Report

Catch up with China

1. A new investigation revealed that the US military secretly ran a campaign to post anti-vaccine propaganda on social media in 2020 and 2021, aiming to sow distrust in the Chinese-made covid vaccines in Southeast Asian countries. (Reuters $)

2. A Chinese court sentenced Huang Xueqin, the journalist who helped launch the #MeToo movement in China, to five years in prison for “inciting subversion of state power.” (Washington Post $)

3. A Shein executive said the company’s corporate values basically make it an American company, but the company is now trying to hide that remark to avoid upsetting Beijing. (Financial Times $)

4. China is getting close to building the world’s largest particle collider, potentially starting in 2027. (Nature)

5. To retaliate for the European Union’s raising tariffs on electric vehicles, the Chinese government has opened an investigation into allegedly unfair subsidies for Europe’s pork exports. (New York Times $)

• On a related note about food: China’s exploding demand for durian fruit in recent years has created a $6 billion business in Southeast Asia, leading some farmers to cut down jungles and coffee plants to make way for durian plantations. (New York Times $)

Lost in translation

In 2012, Jiumei, a Chinese woman in her 20s, began selling a service where she sends “good night” text messages to people online at the price of 1 RMB per text (that’s about $0.14). 

Twelve years, three mobile phones, four different numbers, and over 50,000 messages later, she’s still doing it, according to the Chinese online publication Personage. Some of her clients are buying the service for themselves, hoping to talk to someone regularly at their most lonely or desperate times. Others are buying it to send anonymous messages—to a friend going through a hard time, or an ex-lover who has cut off communications. 

The business isn’t very profitable. Jiumei earns around 3,000 RMB ($410) annually from it on top of her day job, and even less in recent years. But she’s persisted because the act of sending these messages has become a nightly ritual—not just for her customers but also for Jiumei herself, offering her solace in her own times of loneliness and hardship.

One more thing

Globally, Kuaishou has been much less successful than its nemesis ByteDance, except in one country: Brazil. Kwai, the overseas version of Kuaishou, has been so popular in Brazil that even the Marubo people, a tribal group in the remote Amazonian rainforests and one of the last communities to be connected online, have begun using the app, according to the New York Times.

The tool, called AudioSeal, is the first that can pinpoint which bits of audio in, for example, a full hourlong podcast might have been generated by AI. It could help to tackle the growing problem of misinformation and scams using voice cloning tools, says Hady Elsahar, a research scientist at Meta. Malicious actors have used generative AI to create audio deepfakes of President Joe Biden, and scammers have used deepfakes to blackmail their victims. Watermarks could in theory help social media companies detect and remove unwanted content. 

However, there are some big caveats. Meta says it has no plans yet to apply the watermarks to AI-generated audio created using its tools. Audio watermarks are not yet adopted widely, and there is no single agreed industry standard for them. And watermarks for AI-generated content tend to be easy to tamper with—for example, by removing or forging them. 

Fast detection, and the ability to pinpoint which elements of an audio file are AI-generated, will be critical to making the system useful, says Elsahar. He says the team achieved between 90% and 100% accuracy in detecting the watermarks, much better results than in previous attempts at watermarking audio. 

AudioSeal is available on GitHub for free. Anyone can download it and use it to add watermarks to AI-generated audio clips. It could eventually be overlaid on top of AI audio generation models, so that it is automatically applied to any speech generated using them. The researchers who created it will present their work at the International Conference on Machine Learning in Vienna, Austria, in July.  

AudioSeal is created using two neural networks. One generates watermarking signals that can be embedded into audio tracks. These signals are imperceptible to the human ear but can be detected quickly using the other neural network. Currently, if you want to try to spot AI-generated audio in a longer clip, you have to comb through the entire thing in second-long chunks to see if any of them contain a watermark. This is a slow and laborious process, and not practical on social media platforms with millions of minutes of speech.  

AudioSeal works differently: by embedding a watermark throughout each section of the entire audio track. This allows the watermark to be “localized,” which means it can still be detected even if the audio is cropped or edited. 

Ben Zhao, a computer science professor at the University of Chicago, says this ability, and the near-perfect detection accuracy, makes AudioSeal better than any previous audio watermarking system he’s come across. 

“It’s meaningful to explore research improving the state of the art in watermarking, especially across mediums like speech that are often harder to mark and detect than visual content,” says Claire Leibowicz, head of AI and media integrity at the nonprofit  Partnership on AI. 

But there are some major flaws that need to be overcome before these sorts of audio watermarks can be adopted en masse. Meta’s researchers tested different attacks to remove the watermarks and found that the more information is disclosed about the watermarking algorithm, the more vulnerable it is. The system also requires people to voluntarily add the watermark to their audio files.  

This places some fundamental limitations on the tool, says Zhao. “Where the attacker has some access to the [watermark] detector, it’s pretty fragile,” he says. And this means only Meta will be able to verify whether audio content is AI-generated or not. 

Leibowicz says she remains unconvinced that watermarks will actually further public trust in the information they’re seeing or hearing, despite their popularity as a solution in the tech sector. That’s partly because they are themselves so open to abuse. 

“I’m skeptical that any watermark will be robust to adversarial stripping and forgery,” she adds. 

“Green hydrogen is the key driver to advance decarbonization,” says Dr. Christoph Noeres, head of green hydrogen at global electrolysis specialist thyssenkrupp nucera. This promising low-carbon-intensity technology has the potential to transform entire industries by providing a clean, renewable fuel source, moving us toward a greener world aligned with industry climate goals.

Accelerating production of green hydrogen

Hydrogen is the most abundant element in the universe, and its availability is key to its appeal as a clean energy source. However, hydrogen does not occur naturally in its pure form; it is always bound to other elements in compounds like water (H₂O). Pure hydrogen is extracted and isolated from water through an energy-intensive process called conventional electrolysis.

Hydrogen is typically produced today via steam-methane reforming, in which high-temperature steam is used to produce hydrogen from natural gas. Emissions produced by this process have implications for hydrogen’s overall carbon footprint: worldwide hydrogen production is currently responsible for as many CO₂ emissions as the United Kingdom and Indonesia combined.

A solution lies in green hydrogen—hydrogen produced using electrolysis powered by renewable sources. This unlocks the benefits of hydrogen without the dirty fuels. Unfortunately, very little hydrogen is currently powered by renewables: less than 1% came from non-fossil fuel sources in 2022.

A massive scale-up is underway. According to McKinsey, an estimated 130 to 345 gigawatts (GW) of electrolyzer capacity will be necessary to meet the green hydrogen demand by 2030, with 246 GW of this capacity already announced. This stands in stark contrast to the current installed base of just 1.1 GW. Notably, to ensure that green hydrogen constitutes at least 14% of total energy consumption by 2050, a target that the International Renewable Energy Agency (IRENA) estimates is required to meet climate goals, 5,500 GW of cumulative installed electrolyzer capacity will be required.

However, scaling up green hydrogen production to these levels requires overcoming cost and infrastructure constraints. Becoming cost-competitive means improving and standardizing the technology, harnessing the scale efficiencies of larger projects, and encouraging government action to create market incentives. Moreover, the expansion of renewable energy in regions with significant solar, hydro, or wind energy potential is another crucial factor in lowering renewable power prices and, consequently, the costs of green hydrogen.

Electrolysis innovation

While electrolysis technologies have existed for decades, scaling them up to meet the demand for clean energy will be essential. Alkaline Water Electrolysis (AWE), the most dominant and developed electrolysis method, is poised for this transition. It has been utilized for decades, demonstrating efficiency and reliability in the chemical industry. Moreover, it is more cost effective than other electrolysis technologies and is well suited to be run directly with fluctuating renewable power input. Especially for large-scale applications, AWE demonstrates significant advantages in terms of investment and operating costs. “Transferring small-scale manufacturing and optimizing it towards mass manufacturing will need a high level of investment across the industry,” says Noeres.

Industries that already practice electrolysis, as well as those that already use hydrogen, such as fertilizer production, are well poised for conversion to green hydrogen. For example, thyssenkrupp nucera benefits from a decades-long heritage using electrolyzer technology in the chlor-alkali process, which produces chlorine and caustic soda for the chemical industry. The company “is able to use its existing supply chain to ramp up production quickly, a distinction that all providers don’t share,” says Noeres.

Alongside scaling up existing solutions, thyssenkrupp nucera is developing complementary techniques and technologies. Among these are solid oxide electrolysis cells (SOEC), which perform electrolysis at very high temperatures. While the need for high temperatures means this technique isn’t right for all customers, in industries where waste heat is readily available—such as chemicals—Noeres says SOEC offers up to 20% enhanced efficiency and reduces production costs.

Thyssenkrupp nucera has entered into a strategic partnership with the renowned German research institute Fraunhofer IKTS to move the technology toward applications in industrial manufacturing. The company envisages SOEC as a complement to AWE in the areas where it is cost effective to reduce overall energy consumption. “The combination of AWE and SOEC in thyssenkrupp nucera’s portfolio offers a unique product suite to the industry,” says Noeres.

While advancements in electrolysis technology and the diversification of its applications across various scales and industries are promising for green hydrogen production, a coordinated global ramp-up of renewable energy sources and clean power grids is also crucial. Although AWE electrolyzers are ready for deployment in large-scale, centralized green hydrogen production facilities, these must be integrated with renewable energy sources to truly harness their potential.

Making the green hydrogen market

Storage and transportation remain obstacles to a larger market for green hydrogen. While hydrogen can be compressed and stored, its low density presents a practical challenge. The volume of hydrogen is nearly four times greater than that of natural gas, and storage requires either ultra-high compression or costly refrigeration. Overcoming the economic and technical hurdles of high-volume hydrogen storage and transport will be critical to its potential as an exportable energy carrier.

In 2024, several high-profile green hydrogen projects launched in the U.S., advancing the growth of green hydrogen infrastructure and technology. The landmark Inflation Reduction Act (IRA) provides tax credits and government incentives for producing clean hydrogen and the renewable electricity used in its production. In October 2023, the Biden administration announced $7 billion for the country’s first clean hydrogen hubs, and the U.S. Department of Energy further allocated $750 million for 52 projects across 24 states to dramatically reduce the cost of clean hydrogen and establish American leadership in the industry. The potential economic impact from the IRA legislation is substantial: thyssenkrupp nucera expects the IRA to double or triple the U.S. green hydrogen market size.

“The IRA was a wake-up call for Europe, setting a benchmark for all the other countries on how to support the green hydrogen industry in this startup phase,” says Noeres. Germany’s H2Global scheme was one of the first European efforts to facilitate hydrogen imports with the help of subsidies, and it has since been followed up by the European Hydrogen Bank, which provided €720 million for green hydrogen projects in its pilot auction. “However, more investment is needed to push the green hydrogen industry forward,” says Noeres.

In the current green hydrogen market, China has installed more renewable power than any other country. With lower capital expenditure costs, China produces 40% of the world’s electrolyzers. Additionally, state-owned firms have pledged to build an extensive 6,000-kilometer network of pipelines for green hydrogen transportation by 2050.

Coordinated investment and supportive policies are crucial to ensure attractive incentives that can bring green hydrogen from a niche technology to a scalable solution globally. The Chinese green hydrogen market, along with that of other regions such as the Middle East and North Africa, has advanced significantly, garnering global attention for its competitive edge through large-scale projects. To compete effectively, the EU must create a global level playing field for European technologies through attractive investment incentives that can drive the transition of hydrogen from a niche to a global-scale solution. Supportive policies must be in place to also ensure that green products made with hydrogen, such as steel, are sufficiently incentivized and protected against carbon leakage.

A comprehensive strategy, combining investment incentives, open markets, and protection against market distortions and carbon leakage, is crucial for the EU and other countries to remain competitive in the rapidly evolving global green hydrogen market and achieve a decarbonized energy future. “To advance several gigawatt scale or multi-hundred megawatts projects forward,” says Noeres, “we need significantly more volume globally and comparable funding opportunities to make a real impact on global supply chains.”

This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff.

Why does AI hallucinate?

The World Health Organization’s new chatbot launched on April 2 with the best of intentions. The virtual avatar named SARAH, was designed to dispense health tips about how to eat well, quit smoking, de-stress, and more, for millions around the world. But like all chatbots, SARAH can flub its answers. It was quickly found to give out incorrect information. In one case, it came up with a list of fake names and addresses for nonexistent clinics in San Francisco.

Chatbot fails are now a familiar meme. Meta’s short-lived scientific chatbot Galactica made up academic papers and generated wiki articles about the history of bears in space. In February, Air Canada was ordered to honor a refund policy invented by its customer service chatbot. Last year, a lawyer was fined for submitting court documents filled with fake judicial opinions and legal citations made up by ChatGPT.

This tendency to make things up—known as hallucination—is one of the biggest obstacles holding chatbots back from more widespread adoption. Why do they do it? And why can’t we fix it? Read the full story.

—Will Douglas Heaven

Will’s article is the latest entry in MIT Technology Review Explains, our series explaining the complex, messy world of technology to help you understand what’s coming next. You can check out the rest of the series here.

The story is also from the forthcoming magazine issue of MIT Technology Review, which explores the theme of Play. It’s set to go live on Wednesday June 26, so if you don’t already, subscribe now to get a copy when it lands.

Why artists are becoming less scared of AI

Knock, knock. Who’s there? An AI with generic jokes. Researchers from Google DeepMind asked 20 professional comedians to use popular AI language models to write jokes and comedy performances. Their results were mixed. Although the tools helped them to produce initial drafts and structure their routines, AI was not able to produce anything that was original, stimulating, or, crucially, funny. 

The study is symptomatic of a broader trend: we’re realizing the limitations of what AI can do for artists. It can take on some of the boring, mundane, formulaic aspects of the creative process, but it can’t replace the magic and originality that humans bring. Read the full story.

—Melissa Heikkilä 

This story is from The Algorithm, our weekly AI newsletter. Sign up to receive it in your inbox every Monday.

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 The US government is suing Adobe over concealed fees
And for making it too difficult to cancel a Photoshop subscription. (The Verge)
+ Regulators are going after firms with hard-to-cancel accounts. (NYT $)
+ Adobe’s had an incredibly profitable few years. (Insider $)
+ The company recently announced its plans to safeguard artists against exploitative AI. (MIT Technology Review)

2 The year’s deadly heat waves have only just begun
But not everyone is at equal risk from extreme temperatures. (Vox)
+ Here’s what you need to know about this week’s US heat wave. (WP $)
+ Here’s how much heat your body can take. (MIT Technology Review)

3 Being an influencer isn’t as lucrative as it used to be
It’s getting tougher for content creators to earn a crust from social media alone. (WSJ $)
+ Beware the civilian creators offering to document your wedding. (The Guardian)+ Deepfakes of Chinese influencers are livestreaming 24/7. (MIT Technology Review)

4 How crypto cash could influence the US Presidential election 
‘Crypto voters’ have started mobilizing for Donald Trump, who has been making pro-crypto proclamations. (NYT $)

5 Europe is pumping money into defense tech startups
It’ll be a while until it catches up with the US though. (FT $)
+ Here’s the defense tech at the center of US aid to Israel, Ukraine, and Taiwan. (MIT Technology Review)

6 China’s solar industry is in serious trouble
Its rapid growth hasn’t translated into big profits. (Economist $)
+ Recycling solar panels is still a major environmental challenge, too. (IEEE Spectrum)
+ This solar giant is moving manufacturing from China back to the US. (MIT Technology Review)

7 Brace yourself for AI reading companions
The systems are trained on famous writers’ thoughts on seminal titles. (Wired $)

8 McDonalds is ditching AI chatbots at drive-thrus
The tech just proved too unreliable. (The Guardian)

9 How ice freezes is surprisingly mysterious
It’s not as simple as cooling water to zero degrees. (Quanta Magazine)

10 Keeping your phone cool in hot weather is tough
No direct sunlight, no case, no putting it in the fridge. (WP $)

Quote of the day

“My goal was to show that nature is just so fantastic and creative, and I don’t think any machine can beat that.”

—Photographer Miles Astray explains to the Washington Post why he entered a real photograph of a surreal-looking flamingo into a competition for AI art.

The big story

The Atlantic’s vital currents could collapse. Scientists are racing to understand the dangers.

December 2021

Scientists are searching for clues about one of the most important forces in the planet’s climate system: a network of ocean currents known as the Atlantic Meridional Overturning Circulation. They want to better understand how global warming is changing it, and how much more it could shift, or even collapse.

The problem is the Atlantic circulation seems to be weakening, transporting less water and heat. Because of climate change, melting ice sheets are pouring fresh water into the ocean at the higher latitudes, and the surface waters are retaining more of their heat. Warmer and fresher waters are less dense and thus not as prone to sink, which may be undermining one of the currents’ core driving forces. Read the full story.

—James Temple

We can still have nice things

A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or tweet ’em at me.)

+ This cookie is the perfect replica of those frustrating maze games.
+ Each year, the Roland Garros tennis tournament commissions an artist to create a poster. This collection is remarkable
+ Sesame Street is the best.
+ If your plants aren’t flourishing, these tips might help to get them looking their best.

Knock, knock. 

Who’s there? 

An AI with generic jokes. Researchers from Google DeepMind asked 20 professional comedians to use popular AI language models to write jokes and comedy performances. Their results were mixed. 

The comedians said that the tools were useful in helping them produce an initial “vomit draft” that they could iterate on, and helped them structure their routines. But the AI was not able to produce anything that was original, stimulating, or, crucially, funny. My colleague Rhiannon Williams has the full story.

As Tuhin Chakrabarty, a computer science researcher at Columbia University who specializes in AI and creativity, told Rhiannon, humor often relies on being surprising and incongruous. Creative writing requires its creator to deviate from the norm, whereas LLMs can only mimic it.

And that is becoming pretty clear in the way artists are approaching AI today. I’ve just come back from Hamburg, which hosted one of the largest events for creatives in Europe, and the message I got from those I spoke to was that AI is too glitchy and unreliable to fully replace humans and is best used instead as a tool to augment human creativity. 

Right now, we are in a moment where we are deciding how much creative power we are comfortable giving AI companies and tools. After the boom first started in 2022, when DALL-E 2 and Stable Diffusion first entered the scene, many artists raised concerns that AI companies were scraping their copyrighted work without consent or compensation. Tech companies argue that anything on the public internet falls under fair use, a legal doctrine that allows the reuse of copyrighted-protected material in certain circumstances. Artists, writers, image companies, and the New York Times have filed lawsuits against these companies, and it will likely take years until we have a clear-cut answer as to who is right. 

Meanwhile, the court of public opinion has shifted a lot in the past two years. Artists I have interviewed recently say they were harassed and ridiculed for protesting AI companies’ data-scraping practices two years ago. Now, the general public is more aware of the harms associated with AI. In just two years, the public has gone from being blown away by AI-generated images to sharing viral social media posts about how to opt out of AI scraping—a concept that was alien to most laypeople until very recently. Companies have benefited from this shift too. Adobe has been successful in pitching its AI offerings as an “ethical” way to use the technology without having to worry about copyright infringement. 

There are also several grassroots efforts to shift the power structures of AI and give artists more agency over their data. I’ve written about Nightshade, a tool created by researchers at the University of Chicago, which lets users add an invisible poison attack to their images so that they break AI models when scraped. The same team is behind Glaze, a tool that lets artists mask their personal style from AI copycats. Glaze has been integrated into Cara, a buzzy new art portfolio site and social media platform, which has seen a surge of interest from artists. Cara pitches itself as a platform for art created by people; it filters out AI-generated content. It got nearly a million new users in a few days. 

This all should be reassuring news for any creative people worried that they could lose their job to a computer program. And the DeepMind study is a great example of how AI can actually be helpful for creatives. It can take on some of the boring, mundane, formulaic aspects of the creative process, but it can’t replace the magic and originality that humans bring. AI models are limited to their training data and will forever only reflect the zeitgeist at the moment of their training. That gets old pretty quickly.

Now read the rest of The Algorithm

Deeper Learning

Apple is promising personalized AI in a private cloud. Here’s how that will work.

Last week, Apple unveiled its vision for supercharging its product lineup with artificial intelligence. The key feature, which will run across virtually all of its product line, is Apple Intelligence, a suite of AI-based capabilities that promises to deliver personalized AI services while keeping sensitive data secure. 

Why this matters: Apple says its privacy-focused system will first attempt to fulfill AI tasks locally on the device itself. If any data is exchanged with cloud services, it will be encrypted and then deleted afterward. It’s a pitch that offers an implicit contrast with the likes of Alphabet, Amazon, or Meta, which collect and store enormous amounts of personal data. Read more from James O’Donnell here. 

Bits and Bytes

How to opt out of Meta’s AI training
If you post or interact with chatbots on Facebook, Instagram, Threads, or WhatsApp, Meta can use your data to train its generative AI models. Even if you don’t use any of Meta’s platforms, it can still scrape data such as photos of you if someone else posts them. Here’s our quick guide on how to opt out. (MIT Technology Review) 

Microsoft’s Satya Nadella is building an AI empire
Nadella is going all in on AI. His $13 billion investment in OpenAI was just the beginning. Microsoft has become an “the world’s most aggressive amasser of AI talent, tools, and technology” and has started building an in-house OpenAI competitor. (The Wall Street Journal)

OpenAI has hired an army of lobbyists
As countries around the world mull AI legislation, OpenAI is on a lobbyist hiring spree to protect its interests. The AI company has expanded its global affairs team from three lobbyists at the start of 2023 to 35 and intends to have up to 50 by the end of this year. (Financial Times)  

UK rolls out Amazon-powered emotion recognition AI cameras on trains
People traveling through some of the UK’s biggest train stations have likely had their faces scanned by Amazon software without their knowledge during an AI trial. London stations such as Euston and Waterloo have tested CCTV cameras with AI to reduce crime and detect people’s emotions. Emotion recognition technology is extremely controversial. Experts say it is unreliable and simply does not work. 
(Wired) 

Clearview AI used your face. Now you may get a stake in the company.
The facial recognition company, which has been under fire for scraping images of people’s faces from the web and social media without their permission, has agreed to an unusual settlement in a class action against it. Instead of paying cash, it is offering a 23% stake in the company for Americans whose faces are in its data sets. (The New York Times) 

Elephants call each other by their names
This is so cool! Researchers used AI to analyze the calls of two herds of African savanna elephants in Kenya. They found that elephants use specific vocalizations for each individual and recognize when they are being addressed by other elephants. (The Guardian) 

The World Health Organization’s new chatbot launched on April 2 with the best of intentions. 

A fresh-faced virtual avatar backed by GPT-3.5, SARAH (Smart AI Resource Assistant for Health) dispenses health tips in eight different languages, 24/7, about how to eat well, quit smoking, de-stress, and more, for millions around the world.

But like all chatbots, SARAH can flub its answers. It was quickly found to give out incorrect information. In one case, it came up with a list of fake names and addresses for nonexistent clinics in San Francisco. The World Health Organization warns on its website that SARAH may not always be accurate.

Here we go again. Chatbot fails are now a familiar meme. Meta’s short-lived scientific chatbot Galactica made up academic papers and generated wiki articles about the history of bears in space. In February, Air Canada was ordered to honor a refund policy invented by its customer service chatbot. Last year, a lawyer was fined for submitting court documents filled with fake judicial opinions and legal citations made up by ChatGPT. 

The problem is, large language models are so good at what they do that what they make up looks right most of the time. And that makes trusting them hard.

This tendency to make things up—known as hallucination—is one of the biggest obstacles holding chatbots back from more widespread adoption. Why do they do it? And why can’t we fix it?

Magic 8 Ball

To understand why large language models hallucinate, we need to look at how they work. The first thing to note is that making stuff up is exactly what these models are designed to do. When you ask a chatbot a question, it draws its response from the large language model that underpins it. But it’s not like looking up information in a database or using a search engine on the web. 

Peel open a large language model and you won’t see ready-made information waiting to be retrieved. Instead, you’ll find billions and billions of numbers. It uses these numbers to calculate its responses from scratch, producing new sequences of words on the fly. A lot of the text that a large language model generates looks as if it could have been copy-pasted from a database or a real web page. But as in most works of fiction, the resemblances are coincidental. A large language model is more like an infinite Magic 8 Ball than an encyclopedia. 

Large language models generate text by predicting the next word in a sequence. If a model sees “the cat sat,” it may guess “on.” That new sequence is fed back into the model, which may now guess “the.” Go around again and it may guess “mat”—and so on. That one trick is enough to generate almost any kind of text you can think of, from Amazon listings to haiku to fan fiction to computer code to magazine articles and so much more. As Andrej Karpathy, a computer scientist and cofounder of OpenAI, likes to put it: large language models learn to dream internet documents. 

Think of the billions of numbers inside a large language model as a vast spreadsheet that captures the statistical likelihood that certain words will appear alongside certain other words. The values in the spreadsheet get set when the model is trained, a process that adjusts those values over and over again until the model’s guesses mirror the linguistic patterns found across terabytes of text taken from the internet. 

To guess a word, the model simply runs its numbers. It calculates a score for each word in its vocabulary that reflects how likely that word is to come next in the sequence in play. The word with the best score wins. In short, large language models are statistical slot machines. Crank the handle and out pops a word. 

It’s all hallucination

The takeaway here? It’s all hallucination, but we only call it that when we notice it’s wrong. The problem is, large language models are so good at what they do that what they make up looks right most of the time. And that makes trusting them hard. 

Can we control what large language models generate so they produce text that’s guaranteed to be accurate? These models are far too complicated for their numbers to be tinkered with by hand. But some researchers believe that training them on even more text will continue to reduce their error rate. This is a trend we’ve seen as large language models have gotten bigger and better. 

Another approach involves asking models to check their work as they go, breaking responses down step by step. Known as chain-of-thought prompting, this has been shown to increase the accuracy of a chatbot’s output. It’s not possible yet, but future large language models may be able to fact-check the text they are producing and even rewind when they start to go off the rails.

But none of these techniques will stop hallucinations fully. As long as large language models are probabilistic, there is an element of chance in what they produce. Roll 100 dice and you’ll get a pattern. Roll them again and you’ll get another. Even if the dice are, like large language models, weighted to produce some patterns far more often than others, the results still won’t be identical every time. Even one error in 1,000—or 100,000—adds up to a lot of errors when you consider how many times a day this technology gets used. 

The more accurate these models become, the more we will let our guard down. Studies show that the better chatbots get, the more likely people are to miss an error when it happens.  

Perhaps the best fix for hallucination is to manage our expectations about what these tools are for. When the lawyer who used ChatGPT to generate fake documents was asked to explain himself, he sounded as surprised as anyone by what had happened. “I heard about this new site, which I falsely assumed was, like, a super search engine,” he told a judge. “I did not comprehend that ChatGPT could fabricate cases.” 

The cost of building the perfect wave

For nearly as long as surfing has existed, surfers have been obsessed with the search for the perfect wave. 

While this hunt has taken surfers from tropical coastlines to icebergs, these days that search may take place closer to home. That is, at least, the vision presented by developers and boosters in the growing industry of surf pools, spurred by advances in wave-­generating technology that have finally created artificial waves surfers actually want to ride.

But there’s a problem: some of these pools are in drought-ridden areas, and face fierce local opposition. At the core of these fights is a question that’s also at the heart of the sport: What is the cost of finding, or now creating, the perfect wave—and who will have to bear it? Read the full story.

—Eileen Guo

This story is from the forthcoming print issue of MIT Technology Review, which explores the theme of Play. It’s set to go live on Wednesday June 26, so if you don’t already, subscribe now to get a copy when it lands.

What happened when 20 comedians got AI to write their routines

AI is good at lots of things: spotting patterns in data, creating fantastical images, and condensing thousands of words into just a few paragraphs. But can it be a useful tool for writing comedy?

New research from Google DeepMind suggests that it can, but only to a very limited extent. It’s an intriguing finding that hints at the ways AI can—and cannot—assist with creative endeavors more generally. Read the full story.

—Rhiannon Williams

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 Meta has paused plans to train AI on European user data
Data regulators rebuffed its claims it had “legitimate interests” in doing so. (Ars Technica)
+ Meta claims it sent more than two billion warning notifications. (TechCrunch)
+ How to opt out of Meta’s AI training. (MIT Technology Review)

2 AI assistants and chatbots can’t say who won the 2020 US election
And that’s a major problem as we get closer to the 2024 polls opening. (WP $)
+ Online conspiracy theorists are targeting political abuse researchers. (The Atlantic $)
+ Asking Meta AI how to disable it triggers some interesting conversations. (Insider $)
+ Meta says AI-generated election content is not happening at a “systemic level.” (MIT Technology Review)

3 A smartphone battery maker claims to have made a breakthrough
Japanese firm TDK says its new material could revolutionize its solid-state batteries. (FT $)
+ And it’s not just phones that could stand to benefit. (CNBC)
+ Meet the new batteries unlocking cheaper electric vehicles. (MIT Technology Review)

4 What should AI logos look like?
Simple, abstract and non-threatening, if these are anything to go by. (TechCrunch)

5 Radiopharmaceuticals fight cancer with molecular precision
Their accuracy can lead to fewer side effects for patients. (Knowable Magazine)

6 UK rail passengers’ emotions were assessed by AI cameras 
Major stations tested surveillance cameras designed to predict travelers’ emotions. (Wired $)
+ The movement to limit face recognition tech might finally get a win. (MIT Technology Review)

7 The James Webb Space Telescope has spotted dozens of new supernovae
Dating back to the early universe. (New Scientist $)

8 Rice farming in Vietnam has had a hi-tech makeover
Drones and AI systems are making the laborious work a bit simpler. (Hakai Magazine)
+ How one vineyard is using AI to improve its winemaking. (MIT Technology Review)

9 Meet the researchers working to cool down city parks
Using water misters, cool tubes, and other novel techniques. (Bloomberg $)
+ Here’s how much heat your body can take. (MIT Technology Review)

10 The latest generative AI viral trend? Pregnant male celebrities.
The stupider and weirder the image, the better. (Insider $)

Quote of the day

“It’s really easy to get people addicted to things like social media or mobile games. Learning is really hard.”

—Liz Nagler, senior director of product management at language app Duolingo, tells the Wall Street Journal it’s far trickier to get people to go back to the app every day than you might think.

The big story

The big new idea for making self-driving cars that can go anywhere

May 2022

When Alex Kendall sat in a car on a small road in the British countryside and took his hands off the wheel back in 2016, it was a small step in a new direction—one that a new bunch of startups bet might be the breakthrough that makes driverless cars an everyday reality.

This was the first time that reinforcement learning—an AI technique that trains a neural network to perform a task via trial and error—had been used to teach a car to drive from scratch on a real road. It took less than 20 minutes for the car to learn to stay on the road by itself, Kendall claims.

These startups are betting that smarter, cheaper tech will let them overtake current market leaders. But is this yet more hype from an industry that’s been drinking its own Kool-Aid for years? Read the full story.

—Will Douglas Heaven

We can still have nice things

A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or tweet ’em at me.)

+ Twin Peaks meets Sylvanian Families: what’s not to love?
+ You heard it here first: Brat is the album of the summer.
+ Chilis can be pretty painful to eat, but we love them anyway.
+ How people have been crafting artificial eyes for thousands of years.

For nearly as long as surfing has existed, surfers have been obsessed with the search for the perfect wave. It’s not just a question of size, but also of shape, surface conditions, and duration—ideally in a beautiful natural environment. 

While this hunt has taken surfers from tropical coastlines reachable only by boat to swells breaking off icebergs, these days—as the sport goes mainstream—that search may take place closer to home. That is, at least, the vision presented by developers and boosters in the growing industry of surf pools, spurred by advances in wave-­generating technology that have finally created artificial waves surfers actually want to ride. 

Some surf evangelists think these pools will democratize the sport, making it accessible to more communities far from the coasts—while others are simply interested in cashing in. But a years-long fight over a planned surf pool in Thermal, California, shows that for many people who live in the places where they’re being built, the calculus isn’t about surf at all. 

Just some 30 miles from Palm Springs, on the southeastern edge of the Coachella Valley desert, Thermal is the future home of the 118-acre private, members-only Thermal Beach Club (TBC). The developers promise over 300 luxury homes with a dazzling array of amenities; the planned centerpiece is a 20-plus-acre artificial lagoon with a 3.8-acre surf pool offering waves up to seven feet high. According to an early version of the website, club memberships will start at $175,000 a year. (TBC’s developers did not respond to multiple emails asking for comment.)

That price tag makes it clear that the club is not meant for locals. Thermal, an unincorporated desert community, currently has a median family income of $32,340. Most of its residents are Latino; many are farmworkers. The community lacks much of the basic infrastructure that serves the western Coachella Valley, including public water service—leaving residents dependent on aging private wells for drinking water. 

Just a few blocks away from the TBC site is the 60-acre Oasis Mobile Home Park. A dilapidated development designed for some 1,500 people in about 300 mobile homes, Oasis has been plagued for decades by a lack of clean drinking water. The park owners have been cited numerous times by the Environmental Protection Agency for providing tap water contaminated with high levels of arsenic, and last year, the US Department of Justice filed a lawsuit against them for violating the Safe Drinking Water Act. Some residents have received assistance to relocate, but many of those who remain rely on weekly state-funded deliveries of bottled water and on the local high school for showers. 

Stephanie Ambriz, a 28-year-old special-needs teacher who grew up near Thermal, recalls feeling “a lot of rage” back in early 2020 when she first heard about plans for the TBC development. Ambriz and other locals organized a campaign against the proposed club, which she says the community doesn’t want and won’t be able to access. What residents do want, she tells me, is drinkable water, affordable housing, and clean air—and to have their concerns heard and taken seriously by local officials. 

Despite the grassroots pushback, which twice led to delays to allow more time for community feedback, the Riverside County Board of Supervisors unanimously approved the plans for the club in October 2020. It was, Ambriz says, “a shock to see that the county is willing to approve these luxurious developments when they’ve ignored community members” for decades. (A Riverside County representative did not respond to specific questions about TBC.) 

The desert may seem like a counterintuitive place to build a water-intensive surf pool, but the Coachella Valley is actually “the very best place to possibly put one of these things,” argues Doug Sheres, the developer behind DSRT Surf, another private pool planned for the area. It is “close to the largest [and] wealthiest surf population in the world,” he says, featuring “360 days a year of surfable weather” and mountain and lake views in “a beautiful resort setting” served by “a very robust aquifer.” 

In addition to the two planned projects, the Palm Springs Surf Club (PSSC) has already opened locally. The trifecta is turning the Coachella Valley into “the North Shore of wave pools,” as one aficionado described it to Surfer magazine. 

The effect is an acute cognitive dissonance—one that I experienced after spending a few recent days crisscrossing the valley and trying out the waves at PSSC. But as odd as this setting may seem, an analysis by MIT Technology Review reveals that the Coachella Valley is not the exception. Of an estimated 162 surf pools that have been built or announced around the world, as tracked by the industry publication Wave Pool Magazine, 54 are in areas considered by the nonprofit World Resources Institute (WRI) to face high or extremely high water stress, meaning that they regularly use a large portion of their available surface water supply annually. Regions in the “extremely high” category consume 80% or more of their water, while those in the “high” category use 40% to 80% of their supply. (Not all of Wave Pool Magazine’s listed pools will be built, but the publication tracks all projects that have been announced. Some have closed and over 60 are currently operational.)

Zoom in on the US and nearly half are in places with high or extremely high water stress, roughly 16 in areas served by the severely drought-stricken Colorado River. The greater Palm Springs area falls under the highest category of water stress, according to Samantha Kuzma, a WRI researcher (though she notes that WRI’s data on surface water does not reflect all water sources, including an area’s access to aquifers, or its water management plan).

Now, as TBC’s surf pool and other planned facilities move forward and contribute to what’s becoming a multibillion-dollar industry with proposed sites on every continent except Antarctica, inland waves are increasingly becoming a flash point for surfers, developers, and local communities. There are at least 29 organized movements in opposition to surf clubs around the world, according to an ongoing survey from a coalition called No to the Surf Park in Canéjan, which includes 35 organizations opposing a park in Bordeaux, France.  

While the specifics vary widely, at the core of all these fights is a question that’s also at the heart of the sport: What is the cost of finding, or now creating, the perfect wave—and who will have to bear it? 

Though wave pools have been around since the late 1800s, the first artificial surfing wave was built in 1969, and also in the desert—at Big Surf in Tempe, Arizona. But at that pool and its early successors, surfing was secondary; people who went to those parks were more interested in splashing around, and surfers themselves weren’t too excited by what they had to offer. The manufactured waves were too small and too soft, without the power, shape, or feel of the real thing. 

The tide really turned in 2015, when Kelly Slater, widely considered to be the greatest professional surfer of all time, was filmed riding a six-foot-tall, 50-second barreling wave. As the viral video showed, he was not in the wild but atop a wave generated in a pool in California’s Central Valley, some 100 miles from the coast.

Waves of that height, shape, and duration are a rarity even in the ocean, but “Kelly’s wave,” as it became known, showed that “you can make waves in the pool that are as good as or better than what you get in the ocean,” recalls Sheres, the developer whose company, Beach Street Development, is building mul­tiple surf pools around the country, including DSRT Surf. “That got a lot of folks excited—myself included.” 

In the ocean, a complex combination of factors—including wind direction, tide, and the shape and features of the seafloor—is required to generate a surfable wave. Re-creating them in an artificial environment required years of modeling, precise calculations, and simulations. 

Surf Ranch, Slater’s project in the Central Valley, built a mechanical system in which a 300-ton hydrofoil—which resembles a gigantic metal fin—is pulled along the length of a pool 700 yards long and 70 yards wide by a mechanical device the size of several train cars running on a track. The bottom of the pool is precisely contoured to mimic reefs and other features of the ocean floor; as the water hits those features, its movement creates the 50-second-long barreling wave. Once the foil reaches one end of the pool, it runs backwards, creating another wave that breaks in the opposite direction. 

While the result is impressive, the system is slow, producing just one wave every three to four minutes. 

Around the same time Slater’s team was tinkering with his wave, other companies were developing their own technologies to produce multiple waves, and to do so more rapidly and efficiently—key factors in commercial viability. 

Fundamentally, all the systems create waves by displacing water, but depending on the technology deployed, there are differences in the necessary pool size, the project’s water and energy requirements, the level of customization that’s possible, and the feel of the wave. 

Thomas Lochtefeld is a pioneer in the field and the CEO of Surf Loch, which powers PSSC’s waves. Surf Loch uses pneumatic technology, in which compressed air cycles water through chambers the size of bathroom stalls and lets operators create countless wave patterns.

One demo pool in Australia uses what looks like a giant mechanical doughnut that sends out waves the way a pebble dropped in water sends out ripples. Another proposed plan uses a design that spins out waves from a circular fan—a system that is mobile and can be placed in existing bodies of water. 

Of the two most popular techniques in commercial use, one relies on modular paddles attached to a pier that runs across a pool, which move in precise ways to generate waves. The other is pneumatic technology, which uses compressed air to push water through chambers the size of bathroom stalls, called caissons; the caissons pull in water and then push it back out into the pool. By choosing which modular paddles or caissons move first against the different pool bottoms, and with how much force at a time, operators can create a range of wave patterns. 

Regardless of the technique used, the design and engineering of most modern wave pools are first planned out on a computer. Waves are precisely calculated, designed, simulated, and finally tested in the pool with real surfers before they are set as options on a “wave menu” in proprietary software that surf-pool technologists say offers a theoretically endless number and variety of waves. 

On a Tuesday afternoon in early April, I am the lucky tester at the Palm Springs Surf Club, which uses pneumatic technology, as the team tries out a shoulder-high right-breaking wave. 

I have the pool to myself as the club prepares to reopen; it had closed to rebuild its concrete “beach” just 10 days after its initial launch because the original beach had not been designed to withstand the force of the larger waves that Surf Loch, the club’s wave technology provider, had added to the menu at the last minute. (Weeks after reopening in April, the surf pool closed again as the result of “a third-party equipment supplier’s failure,” according to Thomas Lochtefeld, Surf Loch’s CEO.)

SPENCER LOWELL

In some ways, these pneumatic waves are better than what I typically ride around Los Angeles—more powerful, more consistent, and (on this day, at least) uncrowded. But the edge of the pool and the control tower behind it are almost always in my line of sight. And behind me are the PSSC employees (young men, incredible surfers, who keep an eye on my safety and provide much-needed tips) and then, behind them, the snow-capped San Jacinto Mountains. At the far end of the pool, behind the recently rebuilt concrete beach, is a restaurant patio full of diners who I can’t help but imagine are judging my every move. Still, for the few glorious seconds that I ride each wave, I am in the same flow state I experience in the ocean itself.  

Then I fall and sheepishly paddle back to PSSC’s encouraging surfer-employees to restart the whole process. I would be having a lot of fun—if I could just forget my self-consciousness, and the jarring feeling that I shouldn’t be riding waves in the middle of the desert at all.  

Though long inhabited by Cahuilla Indians, the Coachella Valley was sparsely populated until 1876, when the Southern Pacific Railroad added a new line out to the middle of the arid expanse. Shortly after, the first non-native settlers came to the valley and realized that its artesian wells, which flow naturally without the need to be pumped, provided ideal conditions for farming.  

Agricultural production exploded, and by the early 1900s, these once freely producing wells were putting out significantly less, leading residents to look for alternative water sources. In 1918, they created the Coachella Valley Water District (CVWD) to import water from the Colorado River via a series of canals. This water was used to supply the region’s farms and recharge the Coachella Aquifer, the region’s main source of drinking water. 

SPENCER LOWELL

The water imports continue to this day—though the seven states that draw on the river are currently renegotiating their water rights amid a decades-long megadrought in the region. 

The imported water, along with CVWD’s water management plan, has allowed Coachella’s aquifer to maintain relatively steady levels “going back to 1970, even though most development and population has occurred since,” Scott Burritt, a CVWD spokesperson, told MIT Technology Review in an email. 

This has sustained not only agriculture but also tourism in the valley, most notably its world-class—and water-intensive—golf courses. In 2020, the 120 golf courses under the jurisdiction of the CVWD consumed 105,000 acre-feet of water per year (AFY); that’s an average of 875 AFY, or 285 million gallons per year per course. 

Surf pools’ proponents frequently point to the far larger amount of water golf courses consume to argue that opposing the pools on grounds of their water use is misguided. 

PSSC, the first of the area’s three planned surf clubs to open, requires an estimated 3 million gallons per year to fill its pool; the proposed DSRT Surf holds 7 million gallons and estimates that it will use 24 million gallons per year, which includes maintenance and filtration, and accounts for evaporation. TBC’s planned 20-acre recreational lake, 3.8 acres of which will contain the surf pool, will use 51 million gallons per year, according to Riverside County documents. Unlike standard swimming pools, none of these pools need to be drained and refilled annually for maintenance, saving on potential water use. DSRT Surf also boasts about plans to offset its water use by replacing 1 million square feet of grass from an adjacent golf course with drought-tolerant plants. 

SPENCER LOWELL

With surf parks, “you can see the water,” says Jess Ponting, a cofounder of Surf Park Central, the main industry association, and Stoke, a nonprofit that aims to certify surf and ski resorts—and, now, surf pools—for sustainability. “Even though it’s a fraction of what a golf course is using, it’s right there in your face, so it looks bad.”

But even if it were just an issue of appearance, public perception is important when residents are being urged to reduce their water use, says Mehdi Nemati, an associate professor of environmental economics and policy at the University of California, Riverside. It’s hard to demand such efforts from people who see these pools and luxury developments being built around them, he says. “The questions come: Why do we conserve when there are golf courses or surfing … in the desert?” 

(Burritt, the CVWD representative, notes that the water district “encourages all customers, not just residents, to use water responsibly” and adds that CVWD’s strategic plans project that there should be enough water to serve both the district’s golf courses and its surf pools.)  

Locals opposing these projects, meanwhile, argue that developers are grossly underestimating their water use, and various engineering firms and some county officials have in fact offered projections that differ from the developers’ estimates. Opponents are specifically concerned about the effects of spray, evaporation, and other factors, which increase with higher temperatures, bigger waves, and larger pool sizes. 

As a rough point of reference, Slater’s 14-acre wave pool in Lemoore, California, can lose up to 250,000 gallons of water per day to evaporation, according to Adam Fincham, the engineer who designed the technology. That’s roughly half an Olympic swimming pool.

More fundamentally, critics take issue with even debating whether surf clubs or golf courses are worse. “We push back against all of it,” says Ambriz, who organized opposition to TBC and argues that neither the pool nor an exclusive new golf course in Thermal benefits the local community. Comparing them, she says, obscures greater priorities, like the water needs of households. 

SPENCER LOWELL

The “primary beneficiary” of the area’s water, says Mark Johnson, who served as CVWD’s director of engineering from 2004 to 2016, “should be human consumption.”

Studies have shown that just one AFY, or nearly 326,000 gallons, is generally enough to support all household water needs of three California families every year. In Thermal, the gap between the demands of the surf pool and the needs of the community is even more stark: each year for the past three years, nearly 36,000 gallons of water have been delivered, in packages of 16-ounce plastic water bottles, to residents of the Oasis Mobile Home Park—some 108,000 gallons in all. Compare that with the 51 million gallons that will be used annually by TBC’s lake: it would be enough to provide drinking water to its neighbors at Oasis for the next 472 years.

Furthermore, as Nemati notes, “not all water is the same.” CVWD has provided incentives for golf courses to move toward recycled water and replace grass with less water-­intensive landscaping. But while recycled water and even rainwater have been proposed as options for some surf pools elsewhere in the world, including France and Australia, this is unrealistic in Coachella, which receives just three to four inches of rain per year. 

Instead, the Coachella Valley surf pools will depend on a mix of imported water and nonpotable well water from Coachella’s aquifer. 

But any use of the aquifer worries Johnson. Further drawing down the water, especially in an underground aquifer, “can actually create water quality problems,” he says, by concentrating “naturally occurring minerals … like chromium and arsenic.” In other words, TBC could worsen the existing problem of arsenic contamination in local well water. 

When I describe to Ponting MIT Technology Review’s analysis showing how many surf pools are being built in desert regions, he seems to concede it’s an issue. “If 50% of the surf parks in development are in water-stressed areas,” he says, “then the developers are not thinking about the right things.” 

Before visiting the future site of Thermal Beach Club, I stopped in La Quinta, a wealthy town where, back in 2022, community opposition successfully stopped plans for a fourth pool planned for the Coachella Valley. This one was developed by the Kelly Slater Wave Company, which was acquired by the World Surf League in 2016. 

Alena Callimanis, a longtime resident who was a member of the community group that helped defeat the project, says that for a year and a half, she and other volunteers often spent close to eight hours a day researching everything they could about surf pools—and how to fight them. “We knew nothing when we started,” she recalls. But the group learned quickly, poring over planning documents, consulting hydrologists, putting together presentations, providing comments at city council hearings, and even conducting their own citizen science experiments to test the developers’ assertions about the light and noise pollution the project could create. (After the council rejected the proposal for the surf club, the developers pivoted to previously approved plans for a golf course. Callimanis’s group also opposes the golf course, raising similar concerns about water use, but since plans have already been approved, she says, there is little they can do to fight back.) 

It was a different story in Thermal, where three young activists juggled jobs and graduate programs as they tried to mobilize an under-resourced community. “Folks in Thermal lack housing, lack transportation, and they don’t have the ability to take a day off from work to drive up and provide public comment,” says Ambriz. 

But the local pushback did lead to certain promises, including a community benefit payment of $2,300 per luxury housing unit, totaling $749,800. In the meeting approving the project, Riverside County supervisor Manuel Perez called this “unprecedented” and credited the efforts of Ambriz and her peers. (Ambriz remains unconvinced. “None of that has happened,” she says, and payments to the community don’t solve the underlying water issues that the project could exacerbate.) 

That affluent La Quinta managed to keep a surf pool out of its community where working-class Thermal failed is even more jarring in light of industry rhetoric about how surf pools could democratize the sport. For Bryan Dickerson, the editor in chief of Wave Pool Magazine, the collective vision for the future is that instead of “the local YMCA … putting in a skate park, they put in a wave pool.” Other proponents, like Ponting, describe how wave pools can provide surf therapy or opportunities for underrepresented groups. A design firm in New York City, for example, has proposed to the city a plan for an indoor wave pool in a low-income, primarily black and Latino neighborhood in Queens—for $30 million. 

For its part, PSSC cost an estimated $80 million to build. On top of a $40 general admission fee, a private session like the one I had would cost $3,500 to $5,000 per hour, while a public session would be at least $100 to $200, depending on the surfer’s skill level and the types of waves requested. 

In my two days traversing the 45-mile Coachella Valley, I kept thinking about how this whole area was an artificial oasis made possible only by innovations that changed the very nature of the desert, from the railroad stop that spurred development to the irrigation canals and, later, the recharge basins that stopped the wells from running out. 

In this transformed environment, I can see how the cognitive dissonance of surfing a desert wave begins to shrink, tempting us to believe that technology can once again override the reality of living (or simply playing) in the desert in a warming and drying world. 

But the tension over surf pools shows that when it comes to how we use water, maybe there’s no collective “us” here at all.