During training, the players first face simple one-player games, such as finding a purple cube or placing a yellow ball on a red floor. They advance to more complex multiplayer games like hide and seek or capture the flag, where teams compete to be the first to find and grab their opponent’s flag. The playground manager has no specific goal but aims to improve the general capability of the players over time.
Why is this cool? AIs like DeepMind’s AlphaZero have beaten the world’s best human players at chess and Go. But they can only learn one game at a time. As DeepMind cofounder Shane Legg put it when I spoke to him last year, it’s like having to swap out your chess brain for your Go brain each time you want to switch games.
Researchers are now trying to build AIs that can learn multiple tasks at once, which means teaching them general skills that make it easier to adapt.
One exciting trend in this direction is open-ended learning, where AIs are trained on many different tasks without a specific goal. In many ways, this is how humans and other animals seem to learn, via aimless play. But this requires a vast amount of data. XLand generates that data automatically, in the form of an endless stream of challenges. It is similar to POET, an AI training dojo where two-legged bots learn to navigate obstacles in a 2D landscape. XLand’s world is much more complex and detailed, however.
XLand is also an example of AI learning to make itself, or what Jeff Clune, who helped develop POET and leads a team working on this topic at OpenAI, calls AI-generating algorithms (AI-GAs). “This work pushes the frontiers of AI-GAs,” says Clune. “It is very exciting to see.”
The Download: Introducing our TR35 list, and the death of the smart city
Spoiler alert: our annual Innovators Under 35 list isn’t actually about what a small group of smart young people have been up to (although that’s certainly part of it.) It’s really about where the world of technology is headed next.
As you read about the problems this year’s winners have set out to solve, you’ll also glimpse the near future of AI, biotech, materials, computing, and the fight against climate change.
To connect the dots, we asked five experts—all judges or former winners—to write short essays about where they see the most promise, and the biggest potential roadblocks, in their respective fields. We hope the list inspires you and gives you a sense of what to expect in the years ahead.
Read the full list here.
The Urbanism issue
The modern city is a surveillance device. It can track your movements via your license plate, your cell phone, and your face. But go to any city or suburb in the United States and there’s a different type of monitoring happening, one powered by networks of privately owned doorbell cameras, wildlife cameras, and even garden-variety security cameras.
The latest print issue of MIT Technology Review examines why, independently of local governments, we have built our neighborhoods into panopticons: everyone watching everything, all the time. Here is a selection of some of the new stories in the edition, guaranteed to make you wonder whether smart cities really are so smart after all:
– How groups of online neighborhood watchmen are taking the law into their own hands.
– Why Toronto wants you to forget everything you know about smart cities.
– Bike theft is a huge problem. Specialized parking pods could be the answer.
– Public transport wants to kill off cash—but it won’t be as disruptive as you think.
Toronto wants to kill the smart city forever
Most Quayside watchers have a hard time believing that covid was the real reason for ending the project. Sidewalk Labs never really painted a compelling picture of the place it hoped to build.
The new Waterfront Toronto project has clearly learned from the past. Renderings of the new plans for Quayside—call it Quayside 2.0—released earlier this year show trees and greenery sprouting from every possible balcony and outcropping, with nary an autonomous vehicle or drone in site. The project’s highly accomplished design team—led by Alison Brooks, a Canadian architect based in London; the renowned Ghanaian-British architect David Adjaye; Matthew Hickey, a Mohawk architect from the Six Nations First Nation; and the Danish firm Henning Larsen—all speak of this new corner of Canada’s largest city not as a techno-utopia but as a bucolic retreat.
In every way, Quayside 2.0 promotes the notion that an urban neighborhood can be a hybrid of the natural and the manmade. The project boldly suggests that we now want our cities to be green, both metaphorically and literally—the renderings are so loaded with trees that they suggest foliage is a new form of architectural ornament. In the promotional video for the project, Adjaye, known for his design of the Smithsonian Museum of African American History, cites the “importance of human life, plant life, and the natural world.” The pendulum has swung back toward Howard’s garden city: Quayside 2022 is a conspicuous disavowal not only of the 2017 proposal but of the smart city concept itself.
To some extent, this retreat to nature reflects the changing times, as society has gone from a place of techno-optimism (think: Steve Jobs introducing the iPhone) to a place of skepticism, scarred by data collection scandals, misinformation, online harassment, and outright techno-fraud. Sure, the tech industry has made life more productive over the past two decades, but has it made it better? Sidewalk never had an answer to this.
“To me it’s a wonderful ending because we didn’t end up with a big mistake,” says Jennifer Keesmaat, former chief planner for Toronto, who advised the Ministry of Infrastructure on how to set this next iteration up for success. She’s enthusiastic about the rethought plan for the area: “If you look at what we’re doing now on that site, it’s classic city building with a 21st-century twist, which means it’s a carbon-neutral community. It’s a totally electrified community. It’s a community that prioritizes affordable housing, because we have an affordable-housing crisis in our city. It’s a community that has a strong emphasis on green space and urban agriculture and urban farming. Are those things that are derived from Sidewalk’s proposal? Not really.”
Rewriting what we thought was possible in biotech
What ML and AI in biotech broadly need to engage with are the holes that are unique to the study of health. Success stories like neural nets that learned to identify dogs in images were built with the help of high-quality image labeling that people were in a good position to provide. Even attempts to generate or translate human language are easily verified and audited by experts who speak a particular language.
Instead, much of biology, health, and medicine is very much in the stage of fundamental discovery. How do neurodegenerative diseases work? What environmental factors really matter? What role does nutrition play in overall human health? We don’t know yet. In health and biotech, machine learning is taking on a different, more challenging, task—one that will require less engineering and more science.
Marzyeh Ghassemi is an assistant professor at MIT and a faculty member at the Vector Institute (and a 35 Innovators honoree in 2018).