Sep 13 2025
How do AI models generate videos?

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

It’s been a big year for video generation. In the last nine months OpenAI made Sora public, Google DeepMind launched Veo 3, the video startup Runway launched Gen-4. All can produce video clips that are (almost) impossible to distinguish from actual filmed footage or CGI animation. This year also saw Netflix debut an AI visual effect in its show The Eternaut, the first time video generation has been used to make mass-market TV.

Sure, the clips you see in demo reels are cherry-picked to showcase a company’s models at the top of their game. But with the technology in the hands of more users than ever before—Sora and Veo 3 are available in the ChatGPT and Gemini apps for paying subscribers—even the most casual filmmaker can now knock out something remarkable. 

The downside is that creators are competing with AI slop, and social media feeds are filling up with faked news footage. Video generation also uses up a huge amount of energy, many times more than text or image generation. 

With AI-generated videos everywhere, let’s take a moment to talk about the tech that makes them work.

How do you generate a video?

Let’s assume you’re a casual user. There are now a range of high-end tools that allow pro video makers to insert video generation models into their workflows. But most people will use this technology in an app or via a website. You know the drill: “Hey, Gemini, make me a video of a unicorn eating spaghetti. Now make its horn take off like a rocket.” What you get back will be hit or miss, and you’ll typically need to ask the model to take another pass or 10 before you get more or less what you wanted. 

So what’s going on under the hood? Why is it hit or miss—and why does it take so much energy? The latest wave of video generation models are what’s known as latent diffusion transformers. Yes, that’s quite a mouthful. Let’s unpack each part in turn, starting with diffusion. 

What’s a diffusion model?

Imagine taking an image and adding a random spattering of pixels to it. Take that pixel-spattered image and spatter it again and then again. Do that enough times and you will have turned the initial image into a random mess of pixels, like static on an old TV set. 

A diffusion model is a neural network trained to reverse that process, turning random static into images. During training, it gets shown millions of images in various stages of pixelation. It learns how those images change each time new pixels are thrown at them and, thus, how to undo those changes. 

The upshot is that when you ask a diffusion model to generate an image, it will start off with a random mess of pixels and step by step turn that mess into an image that is more or less similar to images in its training set. 

But you don’t want any image—you want the image you specified, typically with a text prompt. And so the diffusion model is paired with a second model—such as a large language model (LLM) trained to match images with text descriptions—that guides each step of the cleanup process, pushing the diffusion model toward images that the large language model considers a good match to the prompt. 

An aside: This LLM isn’t pulling the links between text and images out of thin air. Most text-to-image and text-to-video models today are trained on large data sets that contain billions of pairings of text and images or text and video scraped from the internet (a practice many creators are very unhappy about). This means that what you get from such models is a distillation of the world as it’s represented online, distorted by prejudice (and pornography).

It’s easiest to imagine diffusion models working with images. But the technique can be used with many kinds of data, including audio and video. To generate movie clips, a diffusion model must clean up sequences of images—the consecutive frames of a video—instead of just one image. 

What’s a latent diffusion model? 

All this takes a huge amount of compute (read: energy). That’s why most diffusion models used for video generation use a technique called latent diffusion. Instead of processing raw data—the millions of pixels in each video frame—the model works in what’s known as a latent space, in which the video frames (and text prompt) are compressed into a mathematical code that captures just the essential features of the data and throws out the rest. 

A similar thing happens whenever you stream a video over the internet: A video is sent from a server to your screen in a compressed format to make it get to you faster, and when it arrives, your computer or TV will convert it back into a watchable video. 

And so the final step is to decompress what the latent diffusion process has come up with. Once the compressed frames of random static have been turned into the compressed frames of a video that the LLM guide considers a good match for the user’s prompt, the compressed video gets converted into something you can watch.  

With latent diffusion, the diffusion process works more or less the way it would for an image. The difference is that the pixelated video frames are now mathematical encodings of those frames rather than the frames themselves. This makes latent diffusion far more efficient than a typical diffusion model. (Even so, video generation still uses more energy than image or text generation. There’s just an eye-popping amount of computation involved.) 

What’s a latent diffusion transformer?

Still with me? There’s one more piece to the puzzle—and that’s how to make sure the diffusion process produces a sequence of frames that are consistent, maintaining objects and lighting and so on from one frame to the next. OpenAI did this with Sora by combining its diffusion model with another kind of model called a transformer. This has now become standard in generative video. 

Transformers are great at processing long sequences of data, like words. That has made them the special sauce inside large language models such as OpenAI’s GPT-5 and Google DeepMind’s Gemini, which can generate long sequences of words that make sense, maintaining consistency across many dozens of sentences. 

But videos are not made of words. Instead, videos get cut into chunks that can be treated as if they were. The approach that OpenAI came up with was to dice videos up across both space and time. “It’s like if you were to have a stack of all the video frames and you cut little cubes from it,” says Tim Brooks, a lead researcher on Sora.

A selection of videos generated with Veo 3 and Midjourney. The clips have been enhanced in postproduction with Topaz, an AI video-editing tool. Credit: VaigueMan

Using transformers alongside diffusion models brings several advantages. Because they are designed to process sequences of data, transformers also help the diffusion model maintain consistency across frames as it generates them. This makes it possible to produce videos in which objects don’t pop in and out of existence, for example. 

And because the videos are diced up, their size and orientation do not matter. This means that the latest wave of video generation models can be trained on a wide range of example videos, from short vertical clips shot with a phone to wide-screen cinematic films. The greater variety of training data has made video generation far better than it was just two years ago. It also means that video generation models can now be asked to produce videos in a variety of formats. 

What about the audio? 

A big advance with Veo 3 is that it generates video with audio, from lip-synched dialogue to sound effects to background noise. That’s a first for video generation models. As Google DeepMind CEO Demis Hassabis put it at this year’s Google I/O: “We’re emerging from the silent era of video generation.” 

The challenge was to find a way to line up video and audio data so that the diffusion process would work on both at the same time. Google DeepMind’s breakthrough was a new way to compress audio and video into a single piece of data inside the diffusion model. When Veo 3 generates a video, its diffusion model produces audio and video together in a lockstep process, ensuring that the sound and images are synched.  

You said that diffusion models can generate different kinds of data. Is this how LLMs work too? 

No—or at least not yet. Diffusion models are most often used to generate images, video, and audio. Large language models—which generate text (including computer code)—are built using transformers. But the lines are blurring. We’ve seen how transformers are now being combined with diffusion models to generate videos. And this summer Google DeepMind revealed that it was building an experimental large language model that used a diffusion model instead of a transformer to generate text. 

Here’s where things start to get confusing: Though video generation (which uses diffusion models) consumes a lot of energy, diffusion models themselves are in fact more efficient than transformers. Thus, by using a diffusion model instead of a transformer to generate text, Google DeepMind’s new LLM could be a lot more efficient than existing LLMs. Expect to see more from diffusion models in the near future!

Ecommerce MGMT 0 Comments
Mar 28 2025
What is Signal? The messaging app, explained.

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

With the recent news that the Atlantic’s editor in chief was accidentally added to a group Signal chat for American leaders planning a bombing in Yemen, many people are wondering: What is Signal? Is it secure? If government officials aren’t supposed to use it for military planning, does that mean I shouldn’t use it either?

The answer is: Yes, you should use Signal, but government officials having top-secret conversations shouldn’t use Signal.

Read on to find out why.

What is Signal?

Signal is an app you can install on your iPhone or Android phone, or on your computer. It lets you send secure texts, images, and phone or video chats with other people or groups of people, just like iMessage, Google Messages, WhatsApp, and other chat apps.

Installing Signal is a two-minute process—again, it’s designed to work just like other popular texting apps.

Why is it a problem for government officials to use Signal?

Signal is very secure—as we’ll see below, it’s the best option out there for having private conversations with your friends on your cell phone.

But you shouldn’t use it if you have a legal obligation to preserve your messages, such as while doing government business, because Signal prioritizes privacy over ability to preserve data. It’s designed to securely delete data when you’re done with it, not to keep it. This makes it uniquely unsuited for following public record laws.

You also shouldn’t use it if your phone might be a target of sophisticated hackers, because Signal can only do its job if the phone it is running on is secure. If your phone has been hacked, then the hacker can read your messages regardless of what software you are running.

This is why you shouldn’t use Signal to discuss classified material or military plans. For military communication your civilian phone is always considered hacked by adversaries, so you should instead use communication equipment that is safer—equipment that is physically guarded and designed to do only one job, making it harder to hack.

What about everyone else?

Signal is designed from bottom to top as a very private space for conversation. Cryptographers are very sure that as long as your phone is otherwise secure, no one can read your messages.

Why should you want that? Because private spaces for conversation are very important. In the US, the First Amendment recognizes, in the right to freedom of assembly, that we all need private conversations among our own selected groups in order to function.

And you don’t need the First Amendment to tell you that. You know, just like everyone else, that you can have important conversations in your living room, bedroom, church coffee hour, or meeting hall that you could never have on a public stage. Signal gives us the digital equivalent of that—it’s a space where we can talk, among groups of our choice, about the private things that matter to us, free of corporate or government surveillance. Our mental health and social functioning require that.

So if you’re not legally required to record your conversations, and not planning secret military operations, go ahead and use Signal—you deserve the privacy.

How do we know Signal is secure?

People often give up on finding digital privacy and end up censoring themselves out of caution. So are there really private ways to talk on our phones, or should we just assume that everything is being read anyway?

The good news is: For most of us who aren’t individually targeted by hackers, we really can still have private conversations.

Signal is designed to ensure that if you know your phone and the phones of other people in your group haven’t been hacked (more on that later), you don’t have to trust anything else. It uses many techniques from the cryptography community to make that possible.

Most important and well-known is “end-to-end encryption,” which means that messages can be read only on the devices involved in the conversation and not by servers passing the messages back and forth.

But Signal uses other techniques to keep your messages private and safe as well. For example, it goes to great lengths to make it hard for the Signal server itself to know who else you are talking to (a feature known as “sealed sender”), or for an attacker who records traffic between phones to later decrypt the traffic by seizing one of the phones (“perfect forward secrecy”).

These are only a few of many security properties built into the protocol, which is well enough designed and vetted for other messaging apps, such as WhatsApp and Google Messages, to use the same one.

Signal is also designed so we don’t have to trust the people who make it. The source code for the app is available online and, because of its popularity as a security tool, is frequently audited by experts.

And even though its security does not rely on our trust in the publisher, it does come from a respected source: the Signal Technology Foundation, a nonprofit whose mission is to “protect free expression and enable secure global communication through open-source privacy technology.” The app itself, and the foundation, grew out of a community of prominent privacy advocates. The foundation was started by Moxie Marlinspike, a cryptographer and longtime advocate of secure private communication, and Brian Acton, a cofounder of WhatsApp.

Why do people use Signal over other text apps? Are other ones secure?

Many apps offer end-to-end encryption, and it’s not a bad idea to use them for a measure of privacy. But Signal is a gold standard for private communication because it is secure by default: Unless you add someone you didn’t mean to, it’s very hard for a chat to accidentally become less secure than you intended.

That’s not necessarily the case for other apps. For example, iMessage conversations are sometimes end-to-end encrypted, but only if your chat has “blue bubbles,” and they aren’t encrypted in iCloud backups by default. Google Messages are sometimes end-to-end encrypted, but only if the chat shows a lock icon. WhatsApp is end-to-end encrypted but logs your activity, including “how you interact with others using our Services.”

Signal is careful not to record who you are talking with, to offer ways to reliably delete messages, and to keep messages secure even in online phone backups. This focus demonstrates the benefits of an app coming from a nonprofit focused on privacy rather than a company that sees security as a “nice to have” feature alongside other goals.

(Conversely, and as a warning, using Signal makes it rather easier to accidentally lose messages! Again, it is not a good choice if you are legally required to record your communication.)

Applications like WhatsApp, iMessage, and Google Messages do offer end-to-end encryption and can offer much better security than nothing. The worst option of all is regular SMS text messages (“green bubbles” on iOS)—those are sent unencrypted and are likely collected by mass government surveillance.

Wait, how do I know that my phone is secure?

Signal is an excellent choice for privacy if you know that the phones of everyone you’re talking with are secure. But how do you know that? It’s easy to give up on a feeling of privacy if you never feel good about trusting your phone anyway.

One good place to start for most of us is simply to make sure your phone is up to date. Governments often do have ways of hacking phones, but hacking up-to-date phones is expensive and risky and reserved for high-value targets. For most people, simply having your software up to date will remove you from a category that hackers target.

If you’re a potential target of sophisticated hacking, then don’t stop there. You’ll need extra security measures, and guides from the Freedom of the Press Foundation and the Electronic Frontier Foundation are a good place to start.

But you don’t have to be a high-value target to value privacy. The rest of us can do our part to re-create that private living room, bedroom, church, or meeting hall simply by using an up-to-date phone with an app that respects our privacy.

Jack Cushman is a fellow of the Berkman Klein Center for Internet and Society and directs the Library Innovation Lab at Harvard Law School Library. He is an appellate lawyer, computer programmer, and former board member of the ACLU of Massachusetts.

Ecommerce MGMT 0 Comments
Feb 5 2025
How the Rubin Observatory will help us understand dark matter and dark energy

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

We can put a good figure on how much we know about the universe: 5%. That’s how much of what’s floating about in the cosmos is ordinary matter—planets and stars and galaxies and the dust and gas between them. The other 95% is dark matter and dark energy, two mysterious entities aptly named for our inability to shed light on their true nature. 

Cosmologists have cast dark matter as the hidden glue binding galaxies together. Dark energy plays an opposite role, ripping the fabric of space apart. Neither emits, absorbs, or reflects light, rendering them effectively invisible. So rather than directly observing either of them, astronomers must carefully trace the imprint they leave behind. 

Previous work has begun pulling apart these dueling forces, but dark matter and dark energy remain shrouded in a blanket of questions—critically, what exactly are they?

Enter the Vera C. Rubin Observatory, one of our 10 breakthrough technologies for 2025. Boasting the largest digital camera ever created, Rubin is expected to study the cosmos in the highest resolution yet once it begins observations later this year. And with a better window on the cosmic battle between dark matter and dark energy, Rubin might narrow down existing theories on what they are made of. Here’s a look at how.

Untangling dark matter’s web

In the 1930s, the Swiss astronomer Fritz Zwicky proposed the existence of an unseen force named dunkle Materie—in English, dark matter—after studying a group of galaxies called the Coma Cluster. Zwicky found that the galaxies were traveling too quickly to be contained by their joint gravity and decided there must be a missing, unobservable mass holding the cluster together.

Zwicky’s theory was initially met with much skepticism. But in the 1970s an American astronomer, Vera Rubin, obtained evidence that significantly strengthened the idea. Rubin studied the rotation rates of 60 individual galaxies and found that if a galaxy had only the mass we’re able to observe, that wouldn’t be enough to contain its structure; its spinning motion would send it ripping apart and sailing into space. 

Rubin’s results helped sell the idea of dark matter to the scientific community, since an unseen force seemed to be the only explanation for these spiraling galaxies’ breakneck spin speeds. “It wasn’t necessarily a smoking-gun discovery,” says Marc Kamionkowski, a theoretical physicist at Johns Hopkins University. “But she saw a need for dark matter. And other people began seeing it too.”

Evidence for dark matter only grew stronger in the ensuing decades. But sorting out what might be behind its effects proved tricky. Various subatomic particles were proposed. Some scientists posited that the phenomena supposedly generated by dark matter could also be explained by modifications to our theory of gravity. But so far the hunt, which has employed telescopes, particle colliders, and underground detectors, has failed to identify the culprit. 

The Rubin observatory’s main tool for investigating dark matter will be gravitational lensing, an observational technique that’s been used since the late ’70s. As light from distant galaxies travels to Earth, intervening dark matter distorts its image—like a cosmic magnifying glass. By measuring how the light is bent, astronomers can reverse-engineer a map of dark matter’s distribution. 

Other observatories, like the Hubble Space Telescope and the James Webb Space Telescope, have already begun stitching together this map from their images of galaxies. But Rubin plans to do so with exceptional precision and scale, analyzing the shapes of billions of galaxies rather than the hundreds of millions that current telescopes observe, according to Andrés Alejandro Plazas Malagón, Rubin operations scientist at SLAC National Laboratory. “We’re going to have the widest galaxy survey so far,” Plazas Malagón says.

Capturing the cosmos in such high definition requires Rubin’s 3.2-billion-pixel Large Synoptic Survey Telescope (LSST). The LSST boasts the largest focal plane ever built for astronomy, granting it access to large patches of the sky. 

The telescope is also designed to reorient its gaze every 34 seconds, meaning astronomers will be able to scan the entire sky every three nights. The LSST will revisit each galaxy about 800 times throughout its tenure, says Steven Ritz, a Rubin project scientist at the University of California, Santa Cruz. The repeat exposures will let Rubin team members more precisely measure how the galaxies are distorted, refining their map of dark matter’s web. “We’re going to see these galaxies deeply and frequently,” Ritz says. “That’s the power of Rubin: the sheer grasp of being able to see the universe in detail and on repeat.”

The ultimate goal is to overlay this map on different models of dark matter and examine the results. The leading idea, the cold dark matter model, suggests that dark matter moves slowly compared to the speed of light and interacts with ordinary matter only through gravity. Other models suggest different behavior. Each comes with its own picture of how dark matter should clump in halos surrounding galaxies. By plotting its chart of dark matter against what those models predict, Rubin might exclude some theories and favor others. 

A cosmic tug of war

If dark matter lies on one side of a magnet, pulling matter together, then you’ll flip it over to find dark energy, pushing it apart. “You can think of it as a cosmic tug of war,” Plazas Malagón says.

Dark energy was discovered in the late 1990s, when astronomers found that the universe was not only expanding, but doing so at an accelerating rate, with galaxies moving away from one another at higher and higher speeds. 

“The expectation was that the relative velocity between any two galaxies should have been decreasing,” Kamionkowski says. “This cosmological expansion requires something that acts like antigravity.” Astronomers quickly decided there must be another unseen factor inflating the fabric of space and pegged it as dark matter’s cosmic foil. 

So far, dark energy has been observed primarily through Type Ia supernovas, a special breed of explosion that occurs when a white dwarf star accumulates too much mass. Because these supernovas all tend to have the same peak in luminosity, astronomers can gauge how far away they are by measuring how bright they appear from Earth. Paired with a measure of how fast they are moving, this data clues astronomers in on the universe’s expansion rate. 

Rubin will continue studying dark energy with high-resolution glimpses of Type Ia supernovas. But it also plans to retell dark energy’s cosmic history through gravitational lensing. Because light doesn’t travel instantaneously, when we peer into distant galaxies, we’re really looking at relics from millions to billions of years ago—however long it takes for their light to make the lengthy trek to Earth. Astronomers can effectively use Rubin as a makeshift time machine to see how dark energy has carved out the shape of the universe. 

“These are the types of questions that we want to ask: Is dark energy a constant? If not, is it evolving with time? How is it changing the distribution of dark matter in the universe?” Plazas Malagón says.

If dark energy was weaker in the past, astronomers expect to see galaxies grouped even more densely into galaxy clusters. “It’s like urban sprawl—these huge conglomerates of matter,” Ritz says. Meanwhile, if dark energy was stronger, it would have pushed galaxies away from one another, creating a more “rural” landscape. 

Researchers will be able to use Rubin’s maps of dark matter and the 3D distribution of galaxies to plot out how the structure of the universe changed over time, unveiling the role of dark energy and, they hope, helping scientists evaluate the different theories to account for its behavior. 

Of course, Rubin has a lengthier list of goals to check off. Some top items entail tracing the structure of the Milky Way, cataloguing cosmic explosions, and observing asteroids and comets. But since the observatory was first conceptualized in the early ’90s, its core goal has been to explore this hidden branch of the universe. After all, before a 2019 act of Congress dedicated the observatory to Vera Rubin, it was simply called the Dark Matter Telescope. 

Rubin isn’t alone in the hunt, though. In 2023, the European Space Agency launched the Euclid telescope into space to study how dark matter and dark energy have shaped the structure of the cosmos. And NASA’s Nancy Grace Roman Space Telescope, which is scheduled to launch in 2027, has similar plans to measure the universe’s expansion rate and chart large-scale distributions of dark matter. Both also aim to tackle that looming question: What makes up this invisible empire?

Rubin will test its systems throughout most of 2025 and plans to begin the LSST survey late this year or in early 2026. Twelve to 14 months later, the team expects to reveal its first data set. Then we might finally begin to know exactly how Rubin will light up the dark universe. 

Ecommerce MGMT 0 Comments
Dec 8 2024
What China’s critical mineral ban means for the US

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

This week, China banned exports of several critical minerals to the US, marking the latest move in an escalating series of tit-for-tat trade restrictions between the world’s two largest economies.

In explicitly cutting off, rather than merely restricting, materials of strategic importance to the semiconductor, defense, and electric vehicle sectors, China has clearly crossed a new line in the long-simmering trade war. 

At the same time, it selected minerals that won’t cripple any industries—which leaves China plenty of ammunition to inflict greater economic pain in response to any further trade restrictions that the incoming Trump administration may impose. 

The president-elect recently pledged to impose an additional 10% tariff on all Chinese goods, and he floated tariff rates as high as 60% to 100% during his campaign. But China, which dominates the supply chains for numerous critical minerals essential to high-tech sectors, seems to be telegraphing that it’s prepared to hit back hard.

“It’s a sign of what China is capable of,” says Gracelin Baskaran, director of the Critical Minerals Security Program at the Center for Strategic and International Studies, a bipartisan research nonprofit in Washington, DC. “Shots have been fired.”

What drove the decision?

China’s announcement directly followed the Biden administration’s decision to further restrict exports of chips and other technologies that could help China develop advanced semiconductors used in cutting-edge weapon systems, artificial intelligence, and other applications.

Throughout his presidency, Biden has enacted a series of increasingly aggressive export controls aimed at curbing China’s military strength, technological development, and growing economic power. But the latest clampdown crossed a “clear line in the sand for China,” by threatening its ability to protect national security or shift toward production of more advanced technologies, says Cory Combs, associate director at Trivium China, a research firm.

“It is very much indicative of where Beijing feels its interests lie,” he says.

What exactly did China ban?

In response to the US’s new chip export restrictions, China immediately banned exports of gallium, germanium, antimony, and so called “superhard materials” used heavily in manufacturing, arguing that they have both military and civilian applications, according to the New York Times. China had already placed limits on the sale of most of these goods to the US.

The nation said it may also further restrict sales of graphite, which makes up most of the material in the lithium-ion battery anodes used in electric vehicles, grid storage plants, and consumer electronics. 

What will the bans do?

Experts say, for the most part, the bans won’t have major economic impacts. This is in part because China already restricted exports of these minerals months ago, and also because they are mostly used for niche categories within the semiconductor industry. US imports of these materials from China have already fallen as US companies figured out new sources or substitutes for the materials. 

But a recent US Geological Survey study found that outright bans on gallium and germanium by China could cut US gross domestic product by $3.4 billion. In addition, these are materials that US politicians will certainly take note of, because they “touch on many forms of security: economic, energy, and defense,” Baskaran says. 

Antimony, for example, is used in “armor-piercing ammunition, night-vision goggles, infrared sensors, bullets, and precision optics,” Baskaran and a colleague noted in a recent essay.

Companies rely on gallium to produce a variety of military and electronics components, including satellite systems, power converters, LEDs, and the high-powered chips used in electric vehicles. Germanium is used in fiber optics, infrared optics, and solar cells. 

Before it restricted the flow of these materials, China accounted for more than half of US imports of gallium and germanium, according to the US Geological Survey. Together, China and Russia control 50% of the worldwide reserves of antimony.

How does it affect climate tech?

Any tightened restrictions on graphite could have a pronounced economic impact on US battery and EV makers, in part because there are so few other sources for it. China controls about 80% of graphite output from mines and processes around 70% of the material, according to the International Energy Agency. 

“It would be very significant for batteries,” says Seaver Wang, co-director of the climate and energy team at the Breakthrough Institute, where his research is focused on minerals and manufacturing supply chains. “By weight, you need way more graphite per terawatt hour than nickel, cobalt, or lithium. And the US has essentially no operating production.”

Anything that pushes up the costs of EVs threatens to slow the shift away from gas-guzzlers in the US, as their lofty price tags remain one of the biggest hurdles for many consumers.

How does this impact China’s economy? 

There are real economic risks in China’s decision to cut off the sale of materials it dominates, as it creates incentives for US companies to seek out new sources around the world, switch to substitute materials, and work to develop more domestic supplies where geology allows.

“The challenge China faces is that most of its techniques to increase pain by disrupting supply chains would also impact China, which itself is connected to these supply chains,” says Chris Miller, a professor at Tufts University and author of Chip War: The Fight for the World’s Most Critical Technology.

Notably, the latest announcement could compel US companies to develop their own sources of gallium and germanium, which can be extracted as by-products of zinc and aluminum mining. There are a number of zinc mines in Alaska and Tennessee, and limited extraction of bauxite, which produces aluminum, in Arkansas, Alabama, and Georgia.

Gallium can also be recycled from numerous electronics, providing another potential domestic path for US companies, Combs notes.

The US has already taken steps to counter China’s dominance over the raw ingredients of essential industries, including by issuing a $150 million loan to an Australian company, Syrah Resources, to accelerate the development of graphite mining in Mozambique.

In addition, the mining company Perpetua Resources has proposed reopening a gold mine near Yellow Pine, Idaho, in part to extract antimony trisulfide for use in military applications. The US Department of Defense has provided tens of millions of dollars to help the company conduct environmental studies, though it will still take years for the mine to come online, noted Baskaran and her colleague. 

Wang says that China’s ban might prove “shortsighted,” as any success in diversifying these global supply chains will weaken the nation’s grip in the areas it now dominates. 

What happens next?

The US is also likely to pay very high economic costs in an escalating trade war with China. 

Should the nation decide to enact even stricter trade restrictions, Combs says China could opt to inflict greater economic pain on the US through a variety of means. These could include further restricting or fully banning graphite, as well other crucial battery materials like lithium; cutting off supplies of tungsten, which is used heavily in the aerospace, military, and nuclear power sectors; and halting the sale of copper, which is used in power transmission lines, solar panels, wind turbines, EVs, and many other products. 

China may also decide to take further steps to prevent US firms from selling their goods into the massive market of Chinese consumers and industries, Miller adds. Or it might respond to stricter export restrictions by turning to the US’s economic rivals for advanced technologies.

In the end, it’s not clear either nation wins in a protracted and increasingly combative trade war. But it’s also not apparent that mutually assured economic damage will prove to be an effective deterrent. Indeed, China may well feel the need to impose stricter measures in the coming months or years, as there are few signs that President-elect Trump intends to tone down his hawkish stance toward China.

“It’s hard to see a Trump 2.0 de-escalating with China,” Baskaran says. “We’re on a one-way trajectory toward continued escalation; the question is the pace and the form. It’s not really an ‘if” question.”

Ecommerce MGMT 0 Comments
Aug 19 2024
What Japan’s “megaquake” warning really tells us

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

On August 8, at 16:42 local time, a magnitude-7.1 earthquake shook southern Japan. The temblor, originating off the shores of mainland island of Kyūshū, was felt by nearly a million people across the region, and initially, the threat of a tsunami emerged. But only a diminutive wave swept ashore, buildings remained upright, and nobody died. The crisis was over as quickly as it began.

But then, something new happened. The Japan Meteorological Agency, a government organization, issued a ‘megaquake advisory’ for the first time. This pair of words may appear disquieting—and to some extent, they are. There is a ticking bomb below Japanese waters, a giant crevasse where one tectonic plate dives below another. Stress has been accumulating across this boundary for quite some time, and inevitably, it will do what it has repeatedly done in the past: part of it will violently rupture, generating a devastating earthquake and a potentially huge tsunami.

The advisory was in part issued because it is possible that the magnitude-7.1 quake is a foreshock – a precursory quake – to a far larger one, a tsunami-making monster that could kill a quarter of a million people.

The good news, for now, is that scientists think it is very unlikely that that magnitude-7.1 quake is a prelude to a cataclysm. Nothing is certain, but “the chances that this actually is a foreshock are really quite low,” says Harold Tobin, the director of the Pacific Northwest Seismic Network.

The advisory, ultimately, isn’t prophetic. Its primary purpose is to let the public know that scientists are aware of what’s going on, that they are cognizant of the worst-case scenario—and that everyone else should be mindful of that grim possibility too. Evacuation routes should be memorized, and emergency supplies should be obtained, just in case.

“Even if the probability is low, the consequences are so high,” says Judith Hubbard, an earthquake scientist at Cornell University. “It makes sense to worry about some of these low probabilities.”

Japan, which sits atop a tectonic jigsaw, is no stranger to large earthquakes. Just this past New Year’s Day, a magnitude-7.6 temblor convulsed the Noto Peninsula, killing 230 people. But special attention is paid to certain quakes even when they cause no direct harm.

The August 8 event took place on the Nankai subduction zone: here, the Philippine Sea plate creeps below Japan, which is attached to the Eurasian plate. This type of plate boundary is the sort capable of producing ‘megaquakes’, those of a magnitude-8.0 and higher. (The numerical difference may seem small, but the scale is logarithmic: a magnitude-8.0 quake unleashes 32 times more energy than a magnitude-7.0 quake.)

Consequently, the Nankai subduction zone (or Nankai Trough) has created several historical tragedies. A magnitude-7.9 quake in 1944 was followed by a magnitude-8.0 quake in 1946; both events were caused by part of the submarine trench jolting. The magnitude-8.6 quake of 1707, however, involved the rupture of the entire Nankai Trough. Thousands died on each occasion.

Predicting disaster

Predicting when and where the next major quake will happen anywhere on Earth is currently impossible. Nankai is no different: as recently noted by Hubbard on her blog Earthquake Insights – co-authored with geoscientist Kyle Bradley – there isn’t a set time between Nankai’s major quakes, which range from days to several centuries.

But as stress is continually accumulating on that plate boundary, it’s certain that, one day, the Nankai Trough will let loose another great quake, one which could push a vast volume of seawater toward a large swath of western and central Japan, making a tsunami 100 feet tall. The darkest scenario suggests that 230,000 could perish, two million buildings would be damaged or destroyed, and the country would be left with a $1.4 trillion bill.

Naturally, a magnitude-7.1 quake on that Trough worries scientists. Aftershocks (a series of smaller magnitude quakes) are a guaranteed feature of potent quakes. But there is a small chance that a large quake will be followed by an even larger quake, retrospectively making the first a foreshock.

“The earthquake changes the stress in the surrounding crust a little bit,” says Hubbard. Using the energy released during the August 8 rupture, and decoding the seismic waves created during the quake, scientists can estimate how much stress gets shifted to surrounding faults.

The worry is that some of the stress released by one quake gets transferred to a big fault that hasn’t ruptured in a very long time but is ready to fold like an explosive house of cards. “You never know which increment of stress is gonna be the one that pushes it over the edge.”

Scientists cannot tell whether a large quake is a foreshock until a larger quake occurs. But the possibility remains that the August 8 temblor is a foreshock to something considerably worse. Statistically, it’s unlikely. But there is additional context to why that megaquake advisory was issued: the specter of 2011’s magnitude-9.1 Tōhoku earthquake and tsunami, which killed 18,000 people, still haunts the Japanese government and the nation’s geoscientists. 

Hubbard explains that, two days before that quake struck off Japan’s eastern seaboard, there was a magnitude-7.2 event in the same area—now known to be a foreshock to the catastrophe. Reportedly, authorities in Japan regretted not highlighting that possibility in advance, which may have meant people on the eastern seaboard would have been more prepared, and more capable, of escaping their fate.

A sign to get prepared

In response, Japan’s government created new protocols for signaling that foreshock possibility. Most magnitude-7.0-or-so quakes would not be followed by a ‘megaquake advisory’. Only those happening in tectonic settings able to trigger truly gigantic quakes will—and that includes the Nankai Trough.

Crucially, this advisory is not a warning that a megaquake is imminent. It means: “be ready for when the big earthquake comes,” says Hubbard. Nobody is mandated to evacuate, but they are asked to know their escape routes. Meanwhile, local news reports that nursing homes and hospitals in the region are tallying emergency supplies while moving immobile patients to higher floors or other locations. The high-speed Shinkansen railway trains are running at a reduced maximum speed, and certain flights are carrying more fuel than usual in case they need to divert.

Earthquake advisories aren’t new. “California has something similar, and has issued advisories before,” says Wendy Bohon, an independent earthquake geologist. In September 2016, for example, a swarm of hundreds of modest quakes caused the U.S. Geological Survey to publicly advise that, for a week, there was a 0.03 to 1% chance of a magnitude-7.0-or-greater quake rocking the Southern San Andreas Fault—an outcome that fortunately didn’t come to pass.

But this megaquake advisory is Japan’s first, and it will have both pros and cons. “There are economic and social consequences to this,” says Bohon. Some confusion about how to respond has been reported, and widespread cancellations of travel to the region will come with a price tag. 

But calm reactions to the advisory seem to be the norm, and (ideally) this advisory will result in an increased understanding of the threat of the Nankai Trough. “It really is about raising awareness,” says Adam Pascale, chief scientist at the Seismology Research Centre in Melbourne, Australia. “It’s got everyone talking. And that’s the point.”

Geoscientists are also increasingly optimistic that the August 8 quake isn’t a harbinger of a seismic pandemonium. “This thing is way off to the extreme margin of the actual Nankai rupture zone,” says Tobin—meaning it may not even count as being in the zone of tectonic concern. 

A blog post co-authored by Shinji Toda, a seismologist at Tōhoku University in Sendai, Japan, also estimates that any stress transferal to the dangerous parts of the Trough is negligible. There is no clear evidence that the plate boundary is acting weirdly. And with each day that goes by, the odds of the August 8 quake being a foreshock drop even further.

Tech defenses

But if a megaquake did suddenly emerge, Japan has a technological shield that may mitigate a decent portion of the disaster. 

Buildings are commonly fitted with dampeners that allow them to withstand dramatic quake-triggered shaking. And like America’s West Coast, the entire archipelago has a sophisticated earthquake early-warning system: seismometers close to the quake’s origins listen to its seismic screams, and software makes a quick estimate of the magnitude and shaking intensity of the rupture, before beaming it to people’s various devices, giving them invaluable seconds to get to cover. Automatic countermeasures also slow trains down, control machinery in factories, hospitals, and office buildings, to minimize damage from the incoming shaking.

A tsunami early-warning system also kicks into gear if activated, beaming evacuation notices to phones, televisions, radios, sirens, and myriad specialized receivers in buildings in the afflicted region—giving people several minutes to flee. A megaquake advisory may be new, but for a population highly knowledgeable about earthquake and tsunami defense, it’s just another layer of protection.

The advisory has had other effects too: it’s caused those in another imperiled part of the world to take notice. The Cascadia Subduction Zone offshore from the US Pacific Northwest is also capable of producing both titanic quakes and prodigious tsunamis. Its last grand performance, in 1700, created a tsunami that not only inundated large sections of the North American coast, but it also swamped parts of Japan, all the way across the ocean.

Japan’s megaquake advisory has got Tobin thinking: “What would we do if our subduction zone starts acting weird?” he says—which includes a magnitude-7.0 quake in the Cascadian depths. “There is not a protocol in place the way there is in Japan.” Tobin speculates that a panel of experts would quickly assemble, and a statement – perhaps one not too dissimilar to Japan’s own advisory – would emerge from the U.S. Geological Survey. Like Japan, “we would have to be very forthright about the uncertainty,” he says.

Whether it’s Japan or the US or anywhere else, such advisories aren’t meant to engender panic. “You don’t want people to live their lives in fear,” says Hubbard. But it’s no bad thing to draw attention to the fact that Earth can sometimes be an unforgiving place to live.

Robin George Andrews is an award-winning science journalist and doctor of volcanoes based in London. He regularly writes about the Earth, space, and planetary sciences, and is the author of two critically acclaimed books: Super Volcanoes (2021) and How To Kill An Asteroid (October 2024).

Ecommerce MGMT 0 Comments
Jul 20 2024
How to fix a Windows PC affected by the global outage

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more here.

Windows PCs have crashed in a major IT outage around the world, bringing airlines, major banks, TV broadcasters, health-care providers, and other businesses to a standstill.

Airlines including United, Delta, and American have been forced to ground and delay flights, stranding passengers in airports, while the UK broadcaster Sky News was temporarily pulled off air. Meanwhile, banking customers in Europe, Australia, and India have been unable to access their online accounts. Doctor’s offices and hospitals in the UK have lost access to patient records and appointment scheduling systems. 

The problem stems from a defect in a single content update for Windows machines from the cybersecurity provider CrowdStrike. George Kurtz, CrowdStrike’s CEO, says that the company is actively working with customers affected.

“This is not a security incident or cyberattack,” he said in a statement on X. “The issue has been identified, isolated and a fix has been deployed. We refer customers to the support portal for the latest updates and will continue to provide complete and continuous updates on our website.” CrowdStrike pointed MIT Technology Review to its blog with additional updates for customers.

What caused the issue?

The issue originates from a faulty update from CrowdStrike, which has knocked affected servers and PCs offline and caused some Windows workstations to display the “blue screen of death” when users attempt to boot them. Mac and Linux hosts are not affected.

The update was intended for CrowdStrike’s Falcon software, which is “endpoint detection and response” software designed to protect companies’ computer systems from cyberattacks and malware. But instead of working as expected, the update caused computers running Windows software to crash and fail to reboot. Home PCs running Windows are less likely to have been affected, because CrowdStrike is predominantly used by large organizations. Microsoft did not immediately respond to a request for comment.

“The CrowdStrike software works at the low-level operating system layer. Issues at this level make the OS not bootable,” says Lukasz Olejnik, an independent cybersecurity researcher and consultant, and author of Philosophy of Cybersecurity.

Not all computers running Windows were affected in the same way, he says, pointing out that if a machine’s systems had been turned off at the time CrowdStrike pushed out the update (which has since been withdrawn), it wouldn’t have received it.

For the machines running systems that received the mangled update and were rebooted, an automated update from CloudStrike’s server management infrastructure should suffice, he says.

“But in thousands or millions of cases, this may require manual human intervention,” he adds. “That means a really bad weekend ahead for plenty of IT staff.”

How to manually fix your affected computer

There is a known workaround for Windows computers that requires administrative access to its systems. If you’re affected and have that high level of access, CrowdStrike has recommended the following steps:

1. Boot Windows into safe mode or the Windows Recovery Environment.

2. Navigate to the C:WindowsSystem32driversCrowdStrike directory.

3. Locate the file matching “C-00000291*.sys” and delete it.

4. Boot the machine normally.

Sounds simple, right? But while the above fix is fairly easy to administer, it requires someone to enter it physically, meaning IT teams will need to track down remote machines that have been affected, says Andrew Dwyer of the Department of Information Security at Royal Holloway, University of London.

“We’ve been quite lucky that this is an outage and not an exploitation by a criminal gang or another state,” he says. “It also shows how easy it is to inflict quite significant global damage if you get into the right part of the IT supply chain.”

While fixing the problem is going to cause headaches for IT teams for the next week or so, it’s highly unlikely to cause significant long-term damage to the affected systems—which would not have been the case if it had been ransomware rather than a bungled update, he says.

“If this was a piece of ransomware, there could have been significant outages for months,” he adds. “Without endpoint detection software, many organizations would be in a much more vulnerable place. But they’re critical nodes in the system that have a lot of access to the computer systems that we use.”

Ecommerce MGMT 0 Comments
Jul 19 2024
Google, Amazon and the problem with Big Tech’s climate claims

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

Last week, Amazon trumpeted that it had purchased enough clean electricity to cover the energy demands of all the offices, data centers, grocery stores, and warehouses across its global operations, seven years ahead of its sustainability target. 

That news closely followed Google’s acknowledgment that the soaring energy demands of its AI operations helped ratchet up its corporate emissions by 13% last year—and that it had backed away from claims that it was already carbon neutral.

If you were to take the announcements at face value, you’d be forgiven for believing that Google is stumbling while Amazon is speeding ahead in the race to clean up climate pollution. 

But while both companies are coming up short in their own ways, Google’s approach to driving down greenhouse-gas emissions is now arguably more defensible. 

In fact, there’s a growing consensus that how a company gets to net zero is more important than how fast it does so. And a new school of thought is emerging that moves beyond the net-zero model of corporate climate action, arguing that companies should focus on achieving broader climate impacts rather than trying to balance out every ton of carbon dioxide they emit. 

But to understand why, let’s first examine how the two tech giants’ approaches stack up, and where company climate strategies often go wrong.

Perverse incentives

The core problem is that the costs and complexity of net-zero emissions plans, which require companies to cut or cancel out every ton of climate pollution across their supply chains, can create perverse incentives. Corporate sustainability officers often end up pursuing the quickest, cheapest ways of cleaning up a company’s pollution on paper, rather than the most reliable ways of reducing its emissions in the real world. 

That may mean buying inexpensive carbon credits to offset ongoing pollution from their direct operations or that of their suppliers, rather than undertaking the tougher task of slashing those emissions at the source. Those programs can involve paying other parties to plant trees, restore coastal ecosystems, or alter agriculture practices in ways that purport to reduce emissions or pull carbon dioxide out of the air. The snag is, numerous studies and investigative stories have shown that such efforts often overstate the climate benefits, sometimes wildly.  

Net-zero goals can also compel companies to buy what are known as renewable energy credits (RECs), which ostensibly support additional generation of renewable electricity but raise similar concerns that the climate gains are overstated.

The argument for RECs is that companies often can’t purchase a pure stream of clean electricity to power their operations, since grid operators rely on a mix of natural gas, coal, solar, wind, and other sources. But if those businesses provide money or an indication of demand that spurs developers to build new renewables projects and generate more clean electricity than they would have otherwise, the companies can then claim this cancels out ongoing pollution from the electricity they use.

Experts, however, are less and less convinced of the value of RECs at this stage.

The claim that clean-energy projects wouldn’t have been built without that added support is increasingly unconvincing in a world where those facilities can easily compete in the marketplace on their own, Emily Grubert, an associate professor at Notre Dame, previously told me. And if a company’s purchase of such credits doesn’t bring about changes that reduce the emissions in the atmosphere, it can’t balance out the company’s ongoing pollution. 

‘Creative accounting’

For its part, Amazon is relying on both carbon credits and RECs. 

In its sustainability report, the company says that it reached its clean-electricity targets and drove down emissions by improving energy efficiency, buying more carbon-free power, building renewables projects at its facilities, and supporting such projects around the world. It did this in part by “purchasing additional environmental attributes (such as renewable energy credits) to signal our support for renewable energy in the grids where we operate, in line with the expected generation of the projects we have contracted.”

But there’s yet another issue that can arise when a company pays for clean power that it’s not directly consuming, whether through RECs or through power purchase agreements made before a project is built: Merely paying for renewable electricity generation that occurred at some point, somewhere in the world, isn’t the same as procuring the amount of electricity that the company consumed in the specific places and times that it did so. As you may have heard, the sun stops shining and the wind stops blowing, even as Amazon workers and operations keep grinding around the world and around the clock. 

Paying a solar-farm operator some additional money for producing electricity it was already going to generate in the middle of the day doesn’t in any meaningful way reverse the emissions that an Amazon fulfillment center or server farm produces by, say, drawing electricity from a natural-gas power plant two states away in the middle of the night. 

“The reality on the ground is that its data centers are driving up demand for fossil fuels,” argued a report last week from Amazon Employees for Climate Justice, a group of workers that has been pushing the company to take more aggressive action on climate change. 

The organization said that a significant share of Amazon’s RECs aren’t driving development of new projects. It also stressed that those payments and projects often aren’t generating electricity in the same areas and at the same times that Amazon is consuming power.

The employee group estimates that 78% of Amazon’s US energy comes from nonrenewable sources and accuses the company of using “creative accounting” to claim it’s reached its clean-electricity goals.

To its credit, Amazon is investing billions of dollars in renewables, electrifying its fleet of delivery vehicles, and otherwise making real strides in reducing its waste and emissions. In addition, it’s lobbying US legislators to make it easier to permit electric transmission projects, funding more reliable forms of carbon removal, and working to diversify its mix of electricity sources. The company also insists it’s being careful and selective about the types of carbon offsets it supports, investing only in “additional, quantifiable, real, permanent, and socially beneficial” projects.

“Amazon is focused on making the grid cleaner and more reliable for everyone,” the company said in response to an inquiry from MIT Technology Review. “An emissions-first approach is the fastest, most cost-effective and scalable way to leverage corporate clean-energy procurement to help decarbonize global power grids. This includes procuring renewable energy in locations and countries that still rely heavily on fossil fuels to power their grids, and where energy projects can have the biggest impact on carbon reduction.”

The company has adopted what’s known as a “carbon matching” approach (which it lays out further here), stressing that it wants to be sure the emissions reduced through its investments in renewables equal or exceed the emissions it continues to produce. 

But a recent study led by Princeton researchers found that carbon matching had a “minimal impact” on long-term power system emissions, because it rarely helps get projects built or clean energy generated where those things wouldn’t have happened anyway.

“It’s an offsetting scheme at its core,” Wilson Ricks, an author of the study and an energy systems researcher at Princeton, said of the method, without commenting on Amazon specifically. 

(Meta, Salesforce, and General Motors have also embraced this model, the study notes.)

The problem in asserting that a company is effectively running entirely on clean electricity, when it’s not doing so directly and may not be doing so completely, is that it takes off any pressure to finish the job for real. 

Backing off claims of carbon neutrality

Google has made its own questionable climate claims over the years as well, and it faces growing challenges as the energy it uses for artificial intelligence soars. 

But it is striving to address its power consumption in arguably more defensible ways and now appears to be taking some notable course-correcting steps, according to its recent sustainability report. 

Google says that it’s no longer buying carbon credits that purport to prevent emissions. With this change, it has also backed away from the claim that it had already achieved carbon neutrality across its operations years ago.

“We’re no longer procuring carbon avoidance credits year-over-year to compensate for our annual operational emissions,” the company told MIT Technology Review in a statement. “We’re instead focusing on accelerating an array of carbon solutions and partnerships that will help us work toward our net-zero goal, while simultaneously helping develop broader solutions to mitigate climate change.”

Notably, that includes funding the development of more expensive but possibly more reliable ways of pulling greenhouse gas out of the atmosphere through direct air capture machines or other methods. The company pledged $200 million to Frontier, an effort to pay in advance for one billion tons of carbon dioxide that startups will eventually draw down and store. 

Those commitments may not allow the company to make any assertions about its own emissions today, and some of the early-stage approaches it funds might not work at all. But the hope is that these sorts of investments could help stand up a carbon removal industry, which studies find may be essential for keeping warming in check over the coming decades. 

Clean power around the clock

In addition, for several years now Google has worked to purchase or otherwise support generation of clean power in the areas where it operates and across every hour that it consumes electricity—an increasingly popular approach known as 24/7 carbon-free energy.

The idea is that this will stimulate greater development of what grid operators increasingly need: forms of carbon-free energy that can run at all hours of the day (commonly called “firm generation”), matching up with the actual hour-by-hour energy demands of corporations. That can include geothermal plants, nuclear reactors, hydroelectric plants, and more.

More than 150 organizations and governments have now signed the 24/7 Carbon-Free Energy Compact, a pledge to ensure that clean-electricity purchases match up hourly with their consumption. Those include Google, Microsoft, SAP, and Rivian.

The Princeton study notes that hourly matching is more expensive than other approaches but finds that it drives “significant reductions in system-level CO2 emissions” while “incentivizing advanced clean firm generation and long-duration storage technologies that would not otherwise see market uptake.”

In Google’s case, pursuing 24/7 matching has steered the company to support more renewables projects in the areas where it operates and to invest in more energy storage projects. It has also entered into purchase agreements with power plants that can deliver carbon-free electricity around the clock. These include several deals with Fervo Energy, an enhanced-geothermal startup.

The company says its goal is to achieve net-zero emissions across its supply chains by 2030, with all its electricity use synced up, hour by hour, with clean sources across every grid it operates on.

Energy-hungry AI

Which brings us back to the growing problem of AI energy consumption.

Jonathan Koomey, an independent researcher studying the energy demands of computing, argues that the hue and cry over rising electricity use for AI is overblown. He notes that AI accounts for only a sliver of overall energy consumption from information technology, which produces about 1.4% of global emissions.

But major data center companies like Google, Amazon, and others will need to make significant changes to ensure that they stay ahead of rising AI-driven energy use while keeping on track with their climate goals.

They will have to improve overall energy efficiency, procure more clean energy, and use their clout as major employers to push utilities to increase carbon-free generation in the areas where they operate, he says. But the clear focus must be on directly cutting corporate climate pollution, not mucking around with RECs and offsets.

“Reduce your emissions; that’s it,” Koomey says. “We need actual, real, meaningful emissions reductions, not trading around credits that have, at best, an ambiguous effect.”

Google says it’s already making progress on its AI footprint, while stressing that it’s leveraging artificial intelligence to find ways to drive down climate pollution across sectors. Those include efforts like Tapestry, a project within the company’s X “moonshot factory” to create more efficient and reliable electricity grids, as well as a Google Research collaboration to determine airline flight paths that produce fewer heat-trapping cirrus clouds. 

“AI holds immense promise to drive climate action,” the company said in its report.

The contribution model

The contrasting approaches of Google and Amazon call to mind an instructive hypothetical that a team of carbon market researchers sketched out in a paper this January. They noted that one company could do the hard, expensive work of directly eliminating nearly every ton of its emissions, while another could simply buy cheap offsets to purportedly address all of its own. In that case the first company would have done more actual good for the climate, but only the latter would be able to say it had reached its net-zero target.

Given these challenges and the perverse incentives driving companies toward cheap offsets, the authors have begun arguing for a different approach, known as the “contribution model.”

Like Koomey and others, they stress that companies should dedicate most of their money and energy to directly cutting their emissions as much as possible. But they assert that companies should adopt a new way of dealing with what’s left over (either because that remaining pollution is occurring outside their direct operations or because there are not yet affordable, emissions-free alternatives).

Instead of trying to cancel out every ongoing ton of emissions, a company might pick a percentage of its revenue or set a defensible carbon price on those tons, and then dedicate all that money toward achieving the maximum climate benefit the money can buy, says Libby Blanchard, a research scholar at the University of Cambridge. (She coauthored the paper on the contribution model with Barbara Haya of the University of California, Berkeley, and Bill Anderegg at the University of Utah.)

That could mean funding well-managed forestry projects that help trap carbon dioxide, protect biodiversity, and improve air and water quality. It could mean supporting research and development on the technologies still needed to slow global warming and efforts to scale them up, as Google seems to be doing. Or it could even mean lobbying for stricter climate laws, since few things can drive change as quickly as public policy. 

But the key difference is that the company won’t be able to claim that those actions canceled out every ton of remaining emissions—only that it took real, responsible steps to “contribute” to addressing the problem of climate change. 

The hope is that this approach frees companies to focus on the quality of the projects it funds, not the quantity of cheap offsets it buys, Blanchard says.

It could “replace this race to the bottom with a race to the top,” she says.

As with any approach put before profit-motivated companies that employ ranks of savvy accountants and attorneys, there will surely be ways to abuse this method in the absence of appropriate safeguards and oversight.

And plenty of companies may refuse to adopt it, since they won’t be able to claim they’ve achieved net-zero emissions, which has become the de facto standard for corporate climate action.

But Blanchard says there’s one obvious incentive for them to move away from that goal.

“There’s way less risk that they’ll be sued or accused of greenwashing,” she says.

Ecommerce MGMT 0 Comments
Jul 7 2024
What are AI agents? 

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

When ChatGPT was first released, everyone in AI was talking about the new generation of AI assistants. But over the past year, that excitement has turned to a new target: AI agents. 

Agents featured prominently in Google’s annual I/O conference in May, when the company unveiled its new AI agent called Astra, which allows users to interact with it using audio and video. OpenAI’s new GPT-4o model has also been called an AI agent.  

And it’s not just hype, although there is definitely some of that too. Tech companies are plowing vast sums into creating AI agents, and their research efforts could usher in the kind of useful AI we have been dreaming about for decades. Many experts, including Sam Altman, say they are the next big thing.   

But what are they? And how can we use them? 

How are they defined? 

It is still early days for research into AI agents, and the field does not have a definitive definition for them. But simply, they are AI models and algorithms that can autonomously make decisions in a dynamic world, says Jim Fan, a senior research scientist at Nvidia who leads the company’s AI agents initiative. 

The grand vision for AI agents is a system that can execute a vast range of tasks, much like a human assistant. In the future, it could help you book your vacation, but it will also remember if you prefer swanky hotels, so it will only suggest hotels that have four stars or more and then go ahead and book the one you pick from the range of options it offers you. It will then also suggest flights that work best with your calendar, and plan the itinerary for your trip according to your preferences. It could make a list of things to pack based on that plan and the weather forecast. It might even send your itinerary to any friends it knows live in your destination and invite them along. In the workplace, it  could analyze your to-do list and execute tasks from it, such as sending calendar invites, memos, or emails. 

One vision for agents is that they are multimodal, meaning they can process language, audio, and video. For example, in Google’s Astra demo, users could point a smartphone camera at things and ask the agent questions. The agent could respond to text, audio, and video inputs. 

These agents could also make processes smoother for businesses and public organizations, says David Barber, the director of the University College London Centre for Artificial Intelligence. For example, an AI agent might be able to function as a more sophisticated customer service bot. The current generation of language-model-based assistants can only generate the next likely word in a sentence. But an AI agent would have the ability to act on natural-language commands autonomously and process customer service tasks without supervision. For example, the agent would be able to analyze customer complaint emails and then know to check the customer’s reference number, access databases such as customer relationship management and delivery systems to see whether the complaint is legitimate, and process it according to the company’s policies, Barber says. 

Broadly speaking, there are two different categories of agents, says Fan: software agents and embodied agents. 

Software agents run on computers or mobile phones and use apps, much as in the travel agent example above. “Those agents are very useful for office work or sending emails or having this chain of events going on,” he says. 

Embodied agents are agents that are situated in a 3D world such as a video game, or in a robot. These kinds of agents might make video games more engaging by letting people play with nonplayer characters controlled by AI. These sorts of agents could also help build more useful robots that could help us with everyday tasks at home, such as folding laundry and cooking meals. 

Fan was part of a team that built an embodied AI agent called MineDojo in the popular computer game Minecraft. Using a vast trove of data collected from the internet, Fan’s AI agent was able to learn new skills and tasks that allowed it to freely explore the virtual 3D world and complete complex tasks such as encircling llamas with fences or scooping lava into a bucket. Video games are good proxies for the real world, because they require agents to understand physics, reasoning, and common sense. 

In a new paper, which has not yet been peer-reviewed, researchers at Princeton say that AI agents tend to have three different characteristics. AI systems are considered “agentic” if they can pursue difficult goals without being instructed in complex environments. They also qualify if they can be instructed in natural language and act autonomously without supervision. And finally, the term “agent” can also apply to systems that are able to use tools, such as web search or programming, or are capable of planning. 

Are they a new thing?

The term “AI agents” has been around for years and has meant different things at different times, says Chirag Shah, a computer science professor at the University of Washington. 

There have been two waves of agents, says Fan. The current wave is thanks to the language model boom and the rise of systems such as ChatGPT. 

The previous wave was in 2016, when Google DeepMind introduced AlphaGo, its AI system that can play—and win—the game Go. AlphaGo was able to make decisions and plan strategies. This relied on reinforcement learning, a technique that rewards AI algorithms for desirable behaviors. 

“But these agents were not general,” says Oriol Vinyals, vice president of research at Google DeepMind. They were created for very specific tasks—in this case, playing Go. The new generation of foundation-model-based AI makes agents more universal, as they can learn from the world humans interact with. 

“You feel much more that the model is interacting with the world and then giving back to you better answers or better assisted assistance or whatnot,” says Vinyals. 

What are the limitations? 

There are still many open questions that need to be answered. Kanjun Qiu, CEO and founder of the AI startup Imbue, which is working on agents that can reason and code, likens the state of agents to where self-driving cars were just over a decade ago. They can do stuff, but they’re unreliable and still not really autonomous. For example, a coding agent can generate code, but it sometimes gets it wrong, and it doesn’t know how to test the code it’s creating, says Qiu. So humans still need to be actively involved in the process. AI systems still can’t fully reason, which is a critical step in operating in a complex and  ambiguous human world. 

“We’re nowhere close to having an agent that can just automate all of these chores for us,” says Fan. Current systems “hallucinate and they also don’t always follow instructions closely,” Fan says. “And that becomes annoying.”  

Another limitation is that after a while, AI agents lose track of what they are working on. AI systems are limited by their context windows, meaning the amount of data they can take into account at any given time. 

“ChatGPT can do coding, but it’s not able to do long-form content well. But for human developers, we look at an entire GitHub repository that has tens if not hundreds of lines of code, and we have no trouble navigating it,” says Fan. 

To tackle this problem, Google has increased its models’ capacity to process data, which allows users to have longer interactions with them in which they remember more about past interactions. The company said it is working on making its context windows infinite in the future.

For embodied agents such as robots, there are even more limitations. There is not enough training data to teach them, and researchers are only just starting to harness the power of foundation models in robotics. 

So amid all the hype and excitement, it’s worth bearing in mind that research into AI agents is still in its very early stages, and it will likely take years until we can experience their full potential. 

That sounds cool. Can I try an AI agent now? 

Sort of. You’ve most likely tried their early prototypes, such as OpenAI’s ChatGPT and GPT-4. “If you’re interacting with software that feels smart, that is kind of an agent,” says Qiu. 

Right now the best agents we have are systems with very narrow and specific use cases, such as coding assistants, customer service bots, or workflow automation software like Zapier, she says. But these are a far cry from a universal AI agent that can do complex tasks. 

“Today we have these computers and they’re really powerful, but we have to micromanage them,” says Qiu. 

OpenAI’s ChatGPT plug-ins, which allow people to create AI-powered assistants for web browsers, were an attempt at agents, says Qiu. But these systems are still clumsy, unreliable, and not capable of reasoning, she says. 

Despite that, these systems will one day change the way we interact with technology, Qiu believes, and it is a trend people need to pay attention to. 

“It’s not like, ‘Oh my God, all of a sudden we have AGI’ … but more like ‘Oh my God, my computer can do way more than it did five years ago,’” she says.

Ecommerce MGMT 0 Comments
Jun 19 2024
Why does AI hallucinate?

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

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.” 

Ecommerce MGMT 0 Comments
Apr 27 2024
Here’s the defense tech at the center of US aid to Israel, Ukraine, and Taiwan

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next. You can read more from the series here.

After weeks of drawn-out congressional debate over how much the United States should spend on conflicts abroad, President Joe Biden signed a $95.3 billion aid package into law on Wednesday.

The bill will send a significant quantity of supplies to Ukraine and Israel, while also supporting Taiwan with submarine technology to aid its defenses against China. It’s also sparked renewed calls for stronger crackdowns on Iranian-produced drones. 

Though much of the money will go toward replenishing fairly standard munitions and supplies, the spending bill provides a window into US strategies around four key defense technologies that continue to reshape how today’s major conflicts are being fought.

For a closer look at the military technology at the center of the aid package, I spoke with Andrew Metrick, a fellow with the defense program at the Center for a New American Security, a think tank.

Ukraine and the role of long-range missiles

Ukraine has long sought the Army Tactical Missile System (ATACMS), a long-range ballistic missile made by Lockheed Martin. First debuted in Operation Desert Storm in Iraq in 1990, it’s 13 feet high, two feet wide, and over 3,600 pounds. It can use GPS to accurately hit targets 190 miles away. 

Last year, President Biden was apprehensive about sending such missiles to Ukraine, as US stockpiles of the weapons were relatively low. In October, the administration changed tack. The US sent shipments of ATACMS, a move celebrated by President Volodymyr Zelensky of Ukraine, but they came with restrictions: the missiles were older models with a shorter range, and Ukraine was instructed not to fire them into Russian territory, only Ukrainian territory. 

This week, just hours before the new aid package was signed, multiple news outlets reported that the US had secretly sent more powerful long-range ATACMS to Ukraine several weeks before. They were used on Tuesday, April 23, to target a Russian airfield in Crimea and Russian troops in Berdiansk, 50 miles southwest of Mariupol.

The long range of the weapons has proved essential for Ukraine, says Metrick. “It allows the Ukrainians to strike Russian targets at ranges for which they have very few other options,” he says. That means being able to hit locations like supply depots, command centers, and airfields behind Russia’s front lines in Ukraine. This capacity has grown more important as Ukraine’s troop numbers have waned, Metrick says.

Replenishing Israel’s Iron Dome

On April 13, Iran launched its first-ever direct attack on Israeli soil. In the attack, which Iran says was retaliation for Israel’s airstrike on its embassy in Syria, hundreds of missiles were lobbed into Israeli airspace. Many of them were neutralized by the web of cutting-edge missile launchers dispersed throughout Israel that can automatically detonate incoming strikes before they hit land. 

One of those systems is Israel’s Iron Dome, in which radar systems detect projectiles and then signal units to launch defensive missiles that detonate the target high in the sky before it strikes populated areas. Israel’s other system, called David’s Sling, works a similar way but can identify rockets coming from a greater distance, upwards of 180 miles. 

Both systems are hugely costly to research and build, and the new US aid package allocates $15 billion to replenish their missile stockpile. The missiles can cost anywhere from $100,000 to $10 million each, and a system like Iron Dome might fire them daily during intense periods of conflict. 

The aid comes as funding for Israel has grown more contentious amid the dire conditions faced by displaced Palestinians in Gaza. While the spending bill worked its way through Congress, increasing numbers of Democrats sought to put conditions on the military aid to Israel, particularly after an Israeli air strike on April 1 killed seven aid workers from World Central Kitchen, an international food charity. The funding package does provide $9 billion in humanitarian assistance for the conflict, but the efforts to impose conditions for Israeli military aid failed. 

Taiwan and underwater defenses against China

A rising concern for the US defense community—and a subject of “wargaming” simulations that Metrick has carried out—is an amphibious invasion of Taiwan from China. The rising risk of that scenario has driven the US to build and deploy larger numbers of advanced submarines, Metrick says. A bigger fleet of these submarines would be more likely to keep attacks from China at bay, thereby protecting Taiwan.

The trouble is that the US shipbuilding effort, experts say, is too slow. It’s been hampered by budget cuts and labor shortages, but the new aid bill aims to jump-start it. It will provide $3.3 billion to do so, specifically for the production of Columbia-class submarines, which carry nuclear weapons, and Virginia-class submarines, which carry conventional weapons. 

Though these funds aim to support Taiwan by building up the US supply of submarines, the package also includes more direct support, like $2 billion to help it purchase weapons and defense equipment from the US. 

The US’s Iranian drone problem 

Shahed drones are used almost daily on the Russia-Ukraine battlefield, and Iran launched more than 100 against Israel earlier this month. Produced by Iran and resembling model planes, the drones are fast, cheap, and lightweight, capable of being launched from the back of a pickup truck. They’re used frequently for potent one-way attacks, where they detonate upon reaching their target. US experts say the technology is tipping the scales toward Russian and Iranian military groups and their allies. 

The trouble of combating them is partly one of cost. Shooting down the drones, which can be bought for as little as $40,000, can cost millions in ammunition.

“Shooting down Shaheds with an expensive missile is not, in the long term, a winning proposition,” Metrick says. “That’s what the Iranians, I think, are banking on. They can wear people down.”

This week’s aid package renewed White House calls for stronger sanctions aimed at curbing production of the drones. The United Nations previously passed rules restricting any drone-related material from entering or leaving Iran, but those expired in October. The US now wants them reinstated. 

Even if that happens, it’s unlikely the rules would do much to contain the Shahed’s dominance. The components of the drones are not all that complex or hard to obtain to begin with, but experts also say that Iran has built a sprawling global supply chain to acquire the materials needed to manufacture them and has worked with Russia to build factories. 

“Sanctions regimes are pretty dang leaky,” Metrick says. “They [Iran] have friends all around the world.”

Ecommerce MGMT 0 Comments