King Midas, a figure from Greek mythology, wished that everything he touched would turn to gold. His wish was granted, but his gift quickly became a curse as even food and his daughter were transformed.

A decade ago, many AI doomsday thought experiments involved King Midas scenarios, in which humans told an AI system what to do, and the AI, in doing so maximally and literally, caused a catastrophe. An AI system told to maximize the output of a paper-clip factory would turn all of the atoms on earth, including those that make up human bodies, into paper clips, for example.

Those who work on alignment—the problem of ensuring AI systems behave as their creators intend—no longer worry about this. Researchers can now train AI systems to iteratively learn difficult-to-articulate goals by having humans rank the responses an AI gives by how helpful they are, and having the AI system learn to produce results that it predicts will be rated as helpful as possible. With this method, humans don’t have to say what they want the AI to do, they can simply tell the AI if it has done what they wanted.

As siblings go, Dario and Daniela Amodei agree more than most. “Since we were kids, we’ve always felt very aligned,” Daniela says.

Alignment is top of mind for the brother-and-sister duo at the helm of Anthropic, one of the world’s leading AI labs. In industry lingo, the term means ensuring AI systems are “aligned” with human values. Dario, 40, and Daniela, 36—CEO and president of Anthropic, respectively—believe they are taking a safer and more responsible approach to AI alignment than other companies building cutting-edge AI systems.

Anthropic, which was founded in 2021, has carried out pioneering “mechanistic interpretability” research that aims to allow developers to carry out something analogous to a brain scan—to see what’s really going on inside an AI system, rather than relying on its text outputs alone, which don’t give a true representation of its inner workings. Anthropic has also developed Constitutional AI, a radical new method for aligning AI systems. It has embedded those approaches into its latest chatbot, Claude 2, a close competitor to GPT-4, OpenAI’s most powerful model.

Sam Altman gives humanity a lot of credit. The human race is smart and adaptable enough, he believes, to cope with the release of increasingly powerful AIs into the world—so long as those releases are safe and incremental. “Society is capable of adapting, as people are much smarter and savvier than a lot of the so-called experts think,” Altman, the 38-year-old CEO of OpenAI, told TIME in May. “We can manage this.”

That philosophy not only explains why OpenAI decided to release ChatGPT, its world-shaking chatbot, in November 2022. It’s also why the company doubled down a few months later and launched GPT-4, the most powerful large language model ever made available to the public.

When Demis Hassabis was a young man, he helped design Theme Park, a popular computer game that gave the player a God’s-eye view of a sprawling fairground business. Ever since then, Hassabis, who leads one of the top AI labs, has been trying to attain a God’s-eye view of the world.

As CEO of DeepMind, which was founded in 2010 and acquired by Google in 2014, Hassabis has led teams of computer scientists to AI breakthroughs including solving the vexatious protein-folding problem, and beating human professionals at the complex board game Go. In April 2023, as Google’s Sundar Pichai reshuffled his company’s AI teams after the success of OpenAI’s ChatGPT, Hassabis acquired even more power. The reorganization merged DeepMind and Google’s other AI lab, Google Brain, and put Hassabis at the helm. It was an attempt by Pichai to streamline Google’s efforts toward building powerful artificial intelligence, and ward off the growing competition.

As China’s foremost futurist, Robin Li has been riding the AI wave for a long time. Ever since he founded Baidu, China’s most popular search engine, in 2000, Li’s mission has been to better understand and anticipate human behavior—plowing tens of billions of dollars into AI research. Baidu already has its own equivalent of Amazon’s virtual assistant Alexa, dubbed Xiaodu, as well as fleets of driverless taxis operating in some of China’s biggest cities, including 200 in Wuhan alone. But the recent explosion in generative AI means it’s now “a very exciting time,” Li, 54, tells TIME. “AI now has the ability to do all kinds of logical reasoning that couldn’t be done before.”

On Aug. 31, Baidu publicly released ERNIE Bot, its own large language model (LLM), which Li claims outperforms ChatGPT in several key metrics. Li is also a trusted voice as Beijing ponders appropriate AI regulations. In July, Baidu was appointed a leader of the Chinese government’s National Artificial Intelligence Standardization Group’s LLM task force, whose mood over the past few months has “morphed into a more build-than-regulate mindset,” says Li. “I’m quite hopeful that very soon we will be able to offer public services in a wide range of scenarios.”

AI learns on data, and when China’s regulatory brakes come off, Baidu’s 677 million monthly users would offer quantities that might outstrip those of any domestic competitor. Li believes the possibilities are a “paradigm shift of human-computer interaction,” comparing the current inflection point to the birth of mobile internet, which heralded apps like Uber, WeChat, and TikTok that would never have flourished on desktop computing. Li is predicting “millions and millions of new, AI-native applications that we probably cannot even imagine.”

The burning question is whether Baidu will be able to access the hardware to realize Li’s ambitions. Although the $48 billion firm has developed its own Kunlun microprocessors, it relies heavily on imports from California-based Nvidia, which the Biden Administration has barred from selling its most advanced chips to Chinese customers. Li says the restrictions are “a concern” but might also present an opportunity. “If the barrier becomes higher and higher for us to buy American chips, then domestic chips will become a viable option,” says Li. “There are lots of opportunities to innovate either way, so I’m optimistic about the future of AI.”

Clément Delangue is the CEO of Hugging Face, an open-source, for-profit machine-learning platform where researchers from around the world convene to share their AI models, datasets, and best practices. In an industry where the cutting edge is dominated by big tech companies, Delangue says efforts like Paris-based Hugging Face are an essential counterweight, helping distribute AI to a far broader base of users and developers. At the same time, Hugging Face can enforce a set of community standards that prevent harmful AI models from proliferating, he says. (Investors in Hugging Face include Salesforce, where TIME co-chair and owner Marc Benioff is CEO.)

The entrepreneur Reid Hoffman has successfully bridged multiple generations of technology breakthroughs. He worked at Apple in the ’90s, was a founding member of PayPal’s board of directors, co-founded LinkedIn, and played a crucial role in Facebook’s founding.

Recently, Hoffman, 58, has turned his full, undivided attention to what he believes is the next great technological revolution. Hoffman was one of the first investors in OpenAI, and his venture-capital firm, Greylock Partners, has invested hundreds of millions of dollars in dozens more AI companies. (All his investments in the past two years have been related to AI.) More recently, Hoffman co-founded Inflection AI, an AI chatbot startup, and wrote a book in collaboration with AI called Impromptu.

Greg Brockman, co-founder and president of OpenAI, works 60 to 100 hours per week, and spends around 80% of the time coding. Former colleagues have described him as the hardest-working person at OpenAI.

Brockman fits the profile of the “10x engineer,” Silicon Valley–speak for someone who does the work of 10 normal coders. He was a science prodigy who went to Harvard and transferred to MIT, before dropping out to join Stripe, a financial-technology startup. He was chief technology officer there for five years before leaving to found OpenAI in 2015.

Brockman, now 34, has grander ambitions. If you’re part of a 10-person team, “even if you are actually the mythical 10x, you’re still only going to double the output of that team,” he explains. “What you want to do is 10x the output of the company.” And so Brockman spends much of his time “sniffing around” for things he could do—blockages to address, projects to launch—that would dramatically accelerate OpenAI’s performance.

With the other 20% of his time, Brockman thinks about the big questions facing OpenAI. One such question is OpenAI’s approach to AI safety, disagreements about which reportedly led to the 2021 split in which a number of senior employees left to found Anthropic, now one of OpenAI’s main competitors. Asked about Anthropic, Brockman holds firm. “I’ll observe that Anthropic and us are pursuing very similar strategies,” Brockman says. “So I guess it tells you something.”

Part of OpenAI’s safety strategy is deciding whether and how OpenAI should make the AI models it develops accessible to customers. OpenAI has previously been criticized for decisions to deploy its AI models despite potential harms. But Brockman, a startup engineer to his core, argues the only way to ensure safety is to continue to deploy more powerful models as they are developed and learn from each deployment, addressing issues as they arise.

“I think the most important decision we’ve made in OpenAI’s history was the decision to do iterative deployment,” he says. “Imagine you actually had a very powerful AI, you actually built an AGI [artificial general intelligence, a system that can match human performance on all cognitive tasks], and it’s your first time ever deploying. Are you gonna get that right?”

Marc Andreessen has called his shot before. In 2011, the billionaire venture capitalist penned a blog post titled “Why Software Is Eating the World,” which helped usher in a digital-first era at a time when many companies were still skeptical of its importance.

In June, Andreessen made another bold proclamation with a sequel post of sorts: “Why AI Will Save the World.” Artificial intelligence, Andreessen wrote, could be “a way to make everything we care about better.”

Unsurprisingly, Andreessen is selling a vision of the future in which he is heavily invested. His VC fund, Andreessen Horowitz, backed 18 AI startups in 2022 and at least 10 more in 2023, according to The Information. His bets, which tally in the hundreds of millions of dollars, have included massively successful companies like OpenAI and Character.AI as well as nascent startups. As Andreessen places these early bets, he has also used his prominence to stump against regulation that might constrict these fledgling AI companies.

Of course, Andreessen has also been wrong in the past. Andreessen Horowitz was one of the main drivers of a crypto bubble that popped last year. Bitcoin hasn’t replaced cash, and Coinbase, a crypto exchange that Andreessen has heavily invested in, hasn’t replaced banking—or at least, not yet.

None of this has slowed Andreessen in his quest to impose his tech-forward vision upon the world. If AI continues to race forward, be it toward utopia or doom, the result will be in part due to Andreessen’s unflagging conviction and deep pockets.

Sandra Rivera has worn many hats over the course of her 23-year career at Intel. Now, as chief of its Data Center and AI Group, she leads the company’s push to become one of the go-to makers of AI accelerator chips. The initiative is part of a turnaround effort after a series of what she calls “missteps” that set the company back in an increasingly competitive chip market.

Since she became chief of Intel’s Data Center, as well as its AI strategy and execution in 2021, Rivera has overseen the rollout of its Gaudi AI accelerator chips. The company hopes the Gaudi3, due to launch next year, will give rival Nvidia’s most powerful AI offering, the H100, a run for its money. Intel says the Gaudi2 chip, which launched in May 2022, outperforms Nvidia’s A100—widely seen as the most popular graphics processing unit (GPU) on the market. “The reception has been quite positive because the market wants alternatives to the market leader and they’re looking for something that has better price-performance,” Rivera says, referring to the amount of training that can be achieved with the accelerators per unit of cost.

Rivera must navigate a challenging geopolitical landscape as Intel looks to grow its market share around the world. While still the largest chipmaker in the U.S., it has been overtaken by rival Taiwan Semiconductor Manufacturing Co. on a global scale. Intel has also lagged behind the dominant maker of AI accelerators, Nvidia. In July, Rivera traveled to Beijing to launch a lower-grade version of the Gaudi2, tweaked for the Chinese market to comply with U.S. export restrictions rolled out in October 2022. “There’s a lot of interest in that product,” she says. “There’s just a lot of interest in AI in China, period.”

A daughter of Colombian immigrants who grew up in New Jersey, Rivera credits much of her success to having a “different perspective and experience” than her Intel colleagues. She says she told herself early in her career, “You are not going to be in a majority, you will be remembered [for] how you showed up in that room because you’re so different from most of the other participants. So be remembered for something good.”

Correction, Sept. 8

The original version of this story misstated the name of the group Sandra Rivera leads. It is the Data Center and AI Group, not the AI Group and Data Center.

In Silicon Valley, the specialized semiconductor chips required to train AI systems are a hot commodity. Wait times can be months—forever in startup land. One tech satirist posted a video of himself headed out to sea to intercept a container ship as a way to get his hands on the coveted chips quicker.

Daniel Gross, 32, saw this coming. This year, along with fellow investor Nat Friedman, Gross built Andromeda, a mountain of cutting-edge chips weighing 7,255 lb. and costing around $100 million (including electricity and cooling). The chips are wired together to form a giant “compute cluster,” to which Gross and Friedman trade access in exchange for equity in AI startups they judge to be promising, a move that other venture capitalists are considering aping.

“What these businesses really need that are getting started in AI today is effectively the equivalent of a white hot oven to run their pizza through,” explains Gross. “They need that oven just once or twice to train—to heat up—their basic model and prove to the world that they’re good at what they do.”

It is a characteristically bold bet. When he was just a teenager, Gross founded the search engine Greplin (later renamed Cue). It was acquired by Apple in 2013 for an undisclosed amount, reported to be between $40 million and $60 million. That acquisition led him to run AI and search projects at Apple until he left in 2017 to found the AI vertical at Y Combinator, Silicon Valley’s storied startup accelerator.

Since then, he’s been investing. He’s picked many winners, such as Uber, Instacart, and Coinbase. Where is the smart money in AI now? “Doing monotonous labor with text, extracting things out of PDFs, organizing information—all that stuff, I think, will get accelerated,” Gross predicts.

That’s just the start. Three years after the launch of the iPhone, the top apps were Facebook and a lot of games: Angry Birds, Bejeweled, Words With Friends. “Everyone thought that this was what the iPhone was for; sort of a gaming product with your friends,” says Gross. “The ideas of Uber and Instacart had not fully come around.”

As always, Gross is on the lookout for the next big thing. “The equivalent of looking at the iPhone and dreaming of Uber,” says Gross, “may be very hard to predict.”

Kai-fu Lee wasn’t sure he would see this moment. The Taipei native has been at the vanguard of computer engineering for over four decades, having created what he claims was the world’s first large-vocabulary speech-recognition model for his doctoral dissertation back in 1988, and later serving as a top executive for Apple and Microsoft, as well as head of Google in China.

Still, as recently as 2018, Lee, who is chairman and CEO of Beijing-based Sinovation Ventures, a venture-capital firm that manages $3 billion in Chinese high-tech assets, wrote that artificial general intelligence (AGI)—a hypothetical future technology that can perform most cognitive tasks better than a human—was still decades away. Yet in 2023 the rapid advances of large language model (LLM) applications like ChatGPT means that “by some measures, we’ve already achieved it,” Lee tells TIME. “By other measures, it’s within grasp.”

It’s this astonishing progress that spurred Lee, 61, in July to launch a new language startup, 01.AI, writing that LLM technology “is a historic opportunity that China cannot miss.”

Lee is more than an entrepreneur. He’s an avowed futurist and cancer survivor who has written prolifically about job displacement and the various social upheavals that the AI revolution is already bringing. And as AI capabilities advance faster than anyone thought possible, the timeline for that disruption has also been dramatically shortened, putting the onus on governments to make necessary preparations. The clock is ticking to enact regulation that protects people without stymieing the enormous benefits AI can bring. “A lot more needs to be done,” says Lee. “When AI is this powerful, able to come up with things that we didn’t know before, it might be used to come up with new ways of harming others, of creating weapons, [or] using misinformation to manipulate people for profit or evil intent.”

When Microsoft CEO Satya Nadella asked Jaime Teevan to become the company’s first-ever chief scientist in 2018, aiming to drive research-backed innovation, he was anticipating a period of disruption. What neither could have known was how a global pandemic would completely alter the way many of us work.

The sudden shift to remote and hybrid work required a rethink of the ways many workers meet, communicate, and collaborate, but also generated a wealth of fresh data that could be used to help inform Microsoft’s approach to using AI in its products. ”It’s really important for the current moment we’re in,” Teevan says. Microsoft has been preparing for this inflection point in AI innovation for many years. As technical adviser to Nadella in 2017 and 2018, Teevan focused on how to make AI research central to the company.

Around a year ago she was tasked with integrating GPT-4, the advanced large language model created by Microsoft-backed OpenAI, into Microsoft’s core products. Teevan’s team threw itself into efforts like Copilot, an AI-based tool that works across the Microsoft 365 suite of software that includes Word, Excel, and Outlook, to do tasks such as summarizing meetings, drafting emails, and analyzing data. The pandemic helped influence their thinking about “how AI can change communication and collaboration and help us work together better and understand information better,” she says.

For all the talk about potential uses of AI in the distant future, Teevan is focused on how it can make our lives easier now. “We’re all inventing something new in the context of AI,” she says. “And doing that well really requires business leaders to lead like scientists.”

Looking ahead—and Teevan also leads Microsoft’s “future of work” initiative—language models will help with gathering knowledge at a much faster pace, she says. “I think we’re going to see a fundamental shift in what knowledge is, how knowledge is captured, and how people produce knowledge—and start getting very intentional about how people produce knowledge,” Teevan says. “What makes a conversation useful? What helps your reflection on the conversation afterward? ​​In a Microsoft context, particularly within an organization, how you do that becomes really exciting.”

Tech titans have not minced words about their quest for AI supremacy: “A race starts today,” Microsoft CEO Satya Nadella said on Feb. 7, announcing a new version of its Bing search engine driven by AI. One of Nadella’s strongest assets in this competition is Kevin Scott, Microsoft’s CTO and executive vice president of AI. Scott spearheaded Microsoft’s $1 billion investment in OpenAI in 2019, which immediately placed one of the world’s most advanced AI labs in Microsoft’s corner. OpenAI’s CEO Sam Altman credited Scott with being “most of the reason why we’ve wanted to partner with Microsoft from the very beginning” in a podcast interview earlier this year. Microsoft injected an additional $10 billion into OpenAI this year.

Microsoft had a major AI hiccup in February, however, when a conversation between Bing’s new chatbot and New York Times columnist Kevin Roose went off the rails, with the AI demanding Roose leave his wife for it. Scott characterized the conversation as an “outlier” in an interview with the Verge but immediately ensured the AI’s code was tweaked to shut down its ability to wander into those types of conversations.

These days, one of Scott’s main AI priorities is the development of “copilots,” or AI assistants for virtually any task. The programming assistant GitHub Copilot, for example, is already helping more than 1 million developers code. Microsoft plans to add copilots to the Windows Terminal and Word—like a highly evolved version of Clippy—and hopes that in the future, these copilots will be an essential part of airline-ticket purchases, drug discovery, and much more.

If generative AI is going to revolutionize the future of work, Salesforce’s Clara Shih will almost certainly be among those leading the charge.

Shih, as CEO of the cloud computing giant’s AI division, drives its efforts to help companies use new AI technology, while minimizing the risks of doing so. Those risks aren’t trivial: not only do off-the-shelf tools like ChatGPT sometimes make things up, but unless users opt out, OpenAI can use the data that people put into ChatGPT to train its own models. “People understand that there’s tremendous upside, and that businesses will be transformed by AI,” Shih says, but adds, “The first thing that comes to everyone’s mind is how do I use generative AI safely?”

Marc Raibert does not have a whole lot of respect for robots—which is funny coming from a man who has chosen to spend much of his professional life with them. “Robots are as dumb as toasters, they’re as dumb as doorknobs,” he says. “They do what you tell them to do, but you usually need a pretty well-defined environment in which they can do it.”

Raibert means to change all that. He’s the founder and chairman of Boston Dynamics—best known for its doglike robots that can move about on four legs and perform work like inspecting factories for safety concerns, using their camera-equipped eyes and noise-detection ears to act as sentries at military bases, or investigating suspicious packages. Last year Raibert expanded his portfolio, establishing the Boston Dynamics AI Institute, aiming to give robots not just mobility and function, but also nimbleness and smarts, two things they deeply lack.

Ted Chiang is perhaps the world’s most celebrated living science-fiction author. His short, carefully hewn stories explore how our inner worlds and our societies would react to unexpected rifts in the fabric of science. How would it feel to receive a hormone injection that drastically improved your cognitive function? What if learning an alien language changed the way you perceived time? And if humanity were to create artificial life, what obligations would we owe it?

Recently, Chiang, 56, has stepped into a new role. In nonfiction pieces for the New Yorker, he has emerged as one of the sharpest critics of AI and the corporations behind it. In one viral piece, he compared ChatGPT to “a blurry jpeg of the web,” arguing that the very technology that makes the app so fluent is the reason for its inability to separate truth from fiction. In another, he took aim at the structures of power that new AI advances both arose from and reinforce. Without structural economic changes, he argued, the rise of AI threatens to worsen wealth inequality, weaken worker power, and fortify a tech oligarchy. “What does progress even mean, if it doesn’t include better lives for people who work?” he wrote. “What is the point of greater efficiency, if the money being saved isn’t going anywhere except into shareholders’ bank accounts?”

Black Mirror is perhaps the defining piece of pop culture about 21st century technology. And Charlie Brooker is the mastermind behind it. For more than a decade, Brooker has used his Channel 4–Netflix anthology television show to explore the potentially dystopian results of when our quests for technology advancement are taken to their logical, sometimes violent ends. In Brooker’s stories, radical technology appears and usually changes his characters’ lives at first for the better—and then for the much, much worse.

Artificial intelligence, of course, is one of the utmost subjects of Brooker’s imagination. The satirist has brought to life robotic autonomous killer bees, AI home assistants that are enslaved copies of one’s consciousness, and personalized television shows that are instantly generated based on viewers’ lives. His stories in this realm have earned the praise of actual AI experts for their imagination and accuracy about machine-learning processes.

The singer-songwriter Holly Herndon cannot sing in Spanish. She cannot nail melismatic Arabic vocal runs across multiple octaves. She certainly cannot sing any song you want, on demand, to wherever you are in the world.

But her digital twin, Holly+, can do all of that. Working with technologists, Herndon created a vocal deepfake of herself in 2021 by extensively training a neural network on her voice. Now, any amateur musician can use Holly+ to transform their pedestrian voice into hers, perfectly tuned and ethereal.

The idea of handing over your voice for public manipulation might sound dystopian, a gesture of human surrender to our new machine overlords. But Herndon’s intent is the opposite. She created Holly+ to spur her fellow artists to reclaim agency over their careers and creative autonomy in the midst of a technological revolution she feels will dramatically shift how we make and process art. The world is barreling toward an era of “infinite media,” Herndon says, where anyone can rap as Drake or paint as van Gogh. This makes it all the more crucial to give artists the power to determine what happens with their likenesses and voices.

Pelonomi Moiloa didn’t originally intend to work in AI. As a student at the University of Witwatersrand in South Africa in the early 2010s, she double-majored in biomedical engineering and electrical engineering because she was fascinated by artificial organs and limbs. But as she kept studying, she found that AI was playing an increasingly important role in biomedical engineering. She went on to earn a master’s degree in 2016 at Tohoku University in Japan, where she focused on the intersection of bioengineering and artificial intelligence.

Returning to South Africa, where AI jobs were hard to come by, Moiloa ended up working as a data scientist for a bank. But she never forgot her passion. At the Deep Learning Indaba in 2017, a conference that aimed to strengthen machine learning and artificial intelligence on the African continent, she met the others who would co-found LeLapa AI. The startup—whose name means home in the South African languages Sotho and Tswana—aims to improve the quality of life of Africans via AI. Moiloa, 30, serves as CEO.

The company’s first major project, Vulavula, uses AI to assist with natural language processing of African languages that are often neglected by other researchers—Zulu, Sotho, and Afrikaans are spoken by some 25 million people. These tools will assist companies that reach customers by call centers and text-based customer-service apps. Moiloa says the team would like to venture beyond just language-based AI and is considering developing products in robotics as well.

But for right now, the project Moiloa is most passionate about is training AI models to properly pronounce South African names. “African names get butchered quite badly, to the point of nonrecognition,” says Moiloa. “So that’s something I’m excited about because there’s such a big meaning behind names.”

At this foundational stage, Moiloa calls it essential that a diverse group of people are involved.

“Technology really has the capability to make life a little bit nicer for everyone. But if we don’t create it with that kind of intention, then the absolute opposite is true,” says Moiloa. “So it’s important for us as Africans to own that narrative and own what we want to write into our futures from a code perspective.”

It’s all too clear that AI recognizes certain types of people better than others. Few people have explored this shortcoming with as much depth and clarity as Stephanie Dinkins. The 39-year-old multimedia artist has been programming AIs to realistically depict Black women for years. But she’s found that they come up woefully short: even AIs programmed to think like her seem unable to talk about race or discrimination, or to conjure culturally specific references.

In Dinkins’ recent project, Not the Only One, she trained an AI on three generations of Black women, in an attempt to give an AI cultural roots, deep history, and a perspective that is too often absent from an overwhelmingly homogeneous field. Dinkins says the AI “struggles to have fluid conversation. But every now and again, it outputs a gem I have to wrestle with in terms of how, where, and why it got that idea.” The project helped Dinkins land the $100,000 LG Guggenheim Award, given to an artist pushing the boundaries of technology-based art.

Dinkins also runs an art and tech incubator called AI.Assembly, and has increasingly spent her time trying to spur more participation from people in undertapped communities who don’t believe AI is for them. “We simply cannot afford to ignore or be repulsed by AI,” she says. “It is changing our world exponentially. At the very least, we have to acknowledge it and see what that means to our individual lives.”

Correction, April 29, 2025
The original version of this story mischaracterized Dinkins’ $100,000 award. It was the LG Guggenheim Award, given jointly by LG and the Guggenheim, not solely by the Guggenheim.

Few have sparked the public’s imagination of the potential horrors and wonders of AI like Lilly Wachowski, who wrote and directed The Matrix in 1999 with her sister Lana. The Matrix and its sequels warned of an AI-controlled dystopia in which humans are enslaved by machines and harvested as an energy source. Their vision has had an outsize influence upon many influential AI philosophers and researchers: the AI thinker Eliezer Yudkowsky, for example, named it one of his favorite movies.

This year, Wachowski has been less focused on AI’s existential threat than how its near-term rollout might exacerbate inequalities, including in the film industry. The Screen Actors Guild is currently locked in a labor dispute with Hollywood studios, and one of the major points of contention is AI’s future role in filmmaking. In June, Wachowski took to Twitter to criticize a future in which AI could be used by movie studios to replace actors. “I do vehemently object to the use of AI as a tool to generate wealth,” she wrote. “Technology should be used to benefit humanity (has Star Trek not taught us anything!?) … not for the ultra rich to continue to f-ck over working folks and eliminate jobs for ever larger salaries and dividend payouts.”

The first thing that hit Manu Chopra when he walked into the room was the dirt. It was 2017 and Chopra, then 21, was on a field visit to a data company in Mumbai as part of his new job in AI. Inside the hot, dusty room he saw around 30 men hunched over laptops under a barely moving ceiling fan. When Chopra spoke to them, they told him they were earning $0.40 per hour. He didn’t have the heart to tell them the data they were generating was worth at least 10 times that amount, perhaps much more. “I thought, This cannot be the only way this work can happen,” he says.

The data that makes today’s cutting-edge AI systems possible often originates from factories in the Global South, where workers toil for low wages to teach autonomous vehicles how to drive or, increasingly, rate the reliability of chatbots. Seeing this firsthand led Chopra, now 27, to found Karya: a nonprofit that would do things differently. Karya not only pays its workers at least $5.00 per hour (around 20 times the Indian minimum wage) for their work, it pays them again every time a company licenses it to build a new AI. Much of the work Karya does right now is collecting datasets of Indian languages that have so far been sidelined from the AI boom. That data will go toward building AI systems in those languages that work not just accurately, but also equitably.

Read More: The Workers Behind AI Rarely See Its Rewards. This Indian Startup Wants to Fix That

“I genuinely feel this is the quickest way to move millions of people out of poverty if done right,” Chopra, who was born into poverty and won a scholarship to Stanford that changed the course of his life, told TIME. “Wealth is power. And we want to redistribute wealth to the communities who have been left behind.”

Poverty in sub-Saharan Africa is already deadly—and only growing deadlier, thanks to a dangerous cocktail of climate change, poor infrastructure, and the legacy of colonialism.

Kate Kallot, 32, has focused on a subtle but nonetheless critical issue: a lack of data. Amini, a Nairobi-based startup she founded last year, uses satellite imaging and AI to collect and crunch environmental data to understand what’s happening at ground level, down to the square meter. The result, she says, includes new tools to help people at the grassroots—think smallholder farmers looking to improve their productivity—as well as big companies that will invest in Africa once they have a confident way of tracking these conditions. The result could be transformative.

“Data is the start of any economic revolution,” says Kallot. “Our thesis is that one of the reasons why the continent hasn’t been able to develop itself as fast as the Global North is because of the lack of data.”

Kallot was born in France and spent her childhood visiting family in several African countries. She ran AI chipmaker Nvidia’s emerging-markets business before leaving to found Amini. She is not the first to realize that environmental data can aid development, but her company, which has raised $2 million in early-stage funding, has positioned itself as a leader focused on Africa.

One relevant data point: the continent is home to 65% of the world’s uncultivated arable land. Amini data can not only offer farmers insights on best practices, it might also unlock development that could feed the world as climate change wreaks havoc on today’s breadbaskets. The data also gives insurance companies confidence to offer policies to these small farmers, protecting them from financial ruin in case of the worst climate-related events. This is especially urgent as regulators in the U.S. and Europe begin to implement rules that require big companies—including firms that might invest in Africa—to disclose the climate risks in their supply chains.

And there are other, novel applications. Amini hopes its data can facilitate the preservation of African forests and other natural environments that store carbon in exchange for funding from countries in the Global North. Amini monitors to measure what’s actually being protected—and to ensure that it stays that way. “If we reach the level we are aiming for, you’ll see a much different Africa in a couple of years,” says Kallot.

One of the first things Dr. Ziad Obermeyer realized when he was training in emergency medicine was how hard it was to make decisions that have incredibly high stakes. “It’s really agonizing,” he says over Zoom. “I would go home after my shifts and be stressed about all this stuff that happened, thinking, ‘I should have kept that woman in the hospital. I shouldn’t have sent her home.’”

AI could play a role in tempering some of that anxiety, says Obermeyer, who is now an associate professor at the University of California, Berkeley, School of Public Health, where he focuses on the intersection of machine learning and health. (He also continues to practice emergency medicine in underserved parts of the U.S.) AI, he believes, can help doctors make better decisions—including ones about who should be tested for a heart attack, since missed cardiac events can be catastrophic—while propelling new discoveries, such as ways to reduce unexplained pain among underserved populations.

Ask Alison Darcy to describe how to design an AI companion to help people feel happier, and she’ll tell you the first ingredient is Spock, the logic-driven Star Trek character who struggles with human emotions. Toss in some Kermit the Frog, who’s insightful and never lectures, and then add her late friend Eric Bayer, who had immense compassion—and a hypnotizing way of drawing people in until they revealed truths they hadn’t even realized they were hiding.

All those traits fuse together to create Woebot: a kind, often humorous chatbot that uses AI to function like an automated therapist. “It’s an emotional assistant that’s there for you in tricky moments and always has your best interest at heart,” says Darcy, a clinical research psychologist. Importantly, she says, the chatbot has “a fun dynamic, as much as a therapeutic one.”

As devastating wildfires sweep across California year after year, the state’s home-insurance market is struggling. Beset by enormous claims, Allstate stopped offering new policies in the state last year. State Farm pulled out in May. California homeowners face dwindling options, and many have had to go without insurance entirely.

Nathaniel Manning may have a solution. He worked at the White House Office of Science and Technology Policy under the Obama Administration, helping to make data from the U.S. Agency for International Development and the Federal Emergency Management Agency more accessible to the private sector. One of the prime industries interested in that information, he noticed, was the insurance sector. “I got really obsessed with the space,” he says. “This is the industry that is actually protecting people from climate change.”

In 2019, Manning, 38, co-founded Kettle, a company that’s trying to use AI to create a more nimble insurance market. (He now serves as head of operations.) Most insurance companies use models based on historical data to price policies. They can look at records and see, for instance, that fires occur about once every 50 years in a given area, and then estimate the cost of an insurance policy accordingly. But Manning says this method is getting less effective: “When you have the climate changing, looking in the rearview doesn’t work nearly so well anymore.”

Kettle has a different approach. In addition to historical data, the company is using satellite imagery, weather data, and machine-learning techniques to form what it says is a more accurate picture of the wildfire risks facing California homeowners. This allows the company to offer affordable insurance to homeowners in the state deemed risky by other insurers’ methods, but relatively safe to Kettle. The hope is that the method will create an insurance market that actually prices in climate risk, incentivizing people to move to safer areas. “Some parts of Malibu [are] in the top 5% [of wildfire risk],” says Manning. “That’s really dangerous … You should be paying a lot to live there if you get insurance.” The company has raised about $30 million in venture capital, and it’s recently been growing. Kettle expanded beyond California to the rest of the lower 48 states earlier this summer, and launched a new product insuring against damage from hurricane-force winds. It’s also planning to soon launch an insurance product for businesses. “People don’t realize that it is this data-driven [competition],” says Manning. “Whoever has the best models wins.”

In 2022, medical researchers took tumor samples from 143 patients with advanced blood cancers, and tested them against 139 cancer drugs. An AI system judged which drugs had been most effective against each patient’s tumor sample, and patients were matched with the therapy predicted to be most effective.

Of the 56 patients given personalized drug recommendations, 54% had their cancer kept under control nearly a third longer compared with their prior therapy. Matching cancer patients to the right drug is a difficult task—in recent decades physicians have analyzed the genomes of cancers to prescribe a handful of purpose-built drugs; however, the new approach allows a broader range of drugs to be used in a more targeted way. This, says Andrew Hopkins, founder and CEO of U.K.-based biotech company Exscientia, which developed the drug-selection technology, is one example of how AI is already improving outcomes for patients.

While the AI used in this trial was relatively simple, Exscientia is also building more complicated systems to design new drugs. Exscientia became the first company to enter an AI-designed drug into clinical trials in 2020—it took 12 months to identify the drug candidate (one that aims to treat obsessive-compulsive disorder), a process that can take many years with traditional methods. Since then, it has brought five more AI-designed drugs to clinical trials, including ones that target lung cancer and autoimmune diseases, says Hopkins.

Hopkins, 52, founded Exscientia in 2012, spinning the company out of his work at the University of Dundee, where he held the chairs of medicinal informatics and translational biology. Before that, he spent a decade at pharmaceutical giant Pfizer. Exscientia’s mission, he says, is to automate drug discovery in its entirety—identifying protein targets for drugs, designing drugs that are a good match for those targets, and, as in the 2022 study, selecting which drug from a range is best for a given patient.

In 1996, Hopkins was a Ph.D. student working in the lab to find drugs to treat HIV. One confounding issue with the virus was its adaptability, meaning it would often develop resistance to new drugs within weeks of their use. At 2 a.m. one night, Hopkins had an epiphany of sorts: if drug discovery could be automated using AI, the rate at which scientists found new treatments could accelerate rapidly. Hopkins was especially fond of late-night walks. “We used to have lab meetings at midnight,” Hopkins says. “We often used to go to the pub and then go back to the lab and carry on.”

Hopkins says he and his current team have the same spirit. “What’s incredible about the staff of Exscientia,” says Hopkins, “is how everyone is really sort of engaged in a mission: How do we improve and solve the problem of drug discovery?”

When the Senegalese artist Linda Dounia Rebeiz types “buildings in Dakar” into OpenAI’s text-to-image model DALL-E, it returns squat, decrepit low-rises covered in dirt and peeling paint. They are unrecognizable to Rebeiz compared with the vibrant architecture she sees every day that fills the Senegalese capital.

This is just one example of why Rebeiz, 29, rarely ever uses large-scale models like DALL-E or Midjourney. She finds them inflexible, rudimentary, and riddled with biases that reinforce stereotypes or misconceptions, especially when it comes to images of the Global South. “With DALL-E, it seems impossible to get around the bias and the issues,” she says. “You just had to endlessly reprompt, and it wasn’t working.”

Instead, Rebeiz mostly creates her art with generative adversarial networks (GANs), neural-net architectures that allow her to train AI carefully on her own datasets. Rebeiz has taken pictures of hundreds of images of Senegalese flowers and historical buildings and scanned many more from national archives that had yet to appear anywhere online, and made them into public datasets for others to build upon. Using these novel yet historical datasets, she has created several striking projects, including Once Upon a Flower, which simulates how humans might perceive floral images once global warming has killed off actual flowers.

Rebeiz has also taken a leadership role in encouraging other Black artists to use GANs and participate in a new frontier that currently gets so much wrong about their culture. This summer, she curated a group show on the digital art gallery Feral File featuring 10 Black artists working in AI. “I have to put my drop in the ocean,” she says. “Even if it’s an infinitesimal difference, there’s still a sliver of hope that we can figure out ways of making the data something we can interrogate and change.”