It’s crunch time on climate change, and companies, governments, philanthropists, and NGOs around the world are starting to take action, be it through donating huge sums of money to the cause, building a database for precise tracking of carbon emissions, creating a plan for a clean hydrogen economy, or advocating for solar geoengineering—among many other initiatives. But according to Nvidia, to really know where and how to take action on climate change, we need more data, better modeling, and faster computers. That’s why the company is building what it calls “the world’s most powerful AI supercomputer dedicated to predicting climate change.” The system will be called Earth-2 and will be built using Nvidia’s Omniverse, a multi-GPU development platform for 3D simulation based on Pixar’s Universal Scene Description. In a blog post announcing Earth-2 late last week, Nvidia’s founder and CEO Jensen Huang described his vision for the system as a “digital twin” of Earth. Digital twins aren’t a new concept; they’ve become popular in manufacturing as a way to simulate a product’s performance and tweak the product based on feedback from the simulation. But advances in computing power and AI mean these simulations have become much more granular and powerful, with the ability to drive meaningful change—and that’s just what Huang is hoping for with Earth-2. “We need to confront climate change now. Yet, we won’t feel the impact of our efforts for decades,” he wrote. “It’s hard to mobilize action for something so far in the future. But we must know our future today—see it and feel it—so we can act with urgency.” Plenty of climate models already exist. They quantify factors like air pressure, wind magnitude, and temperature and plug them into equations to get a view of climate patterns in a given region, representing those regions as 3D grids. The smaller the region, the more accurate a model can be before becoming unwieldy (in other words, models must solve more equations to achieve higher resolution, but trying to take on too many equations will make a model so slow that it stops being useful). This means most existing climate models lack both granularity and accuracy. The solution? A bigger, better, faster computer. “Greater resolution is needed to model changes in the global water cycle,” Huang wrote. “Meter-scale resolution is needed to simulate clouds that reflect sunlight back to space. Scientists estimate that these resolutions will demand millions to billions of times more computing power than what’s currently available.” Earth-2 will employ three technologies to achieve ultra-high-resolution climate modeling: GPU-accelerated computing; deep learning and breakthroughs in physics-informed neural networks; and AI supercomputers—and a ton of data. The ultimate aim of this digital twin of our planet is to spur action that will drive meaningful change, both in terms of mitigating the negative impacts of climate change on populations and mitigating climate change itself. Extreme weather events like hurricanes, wildfires, heat waves, and flash floods are increasingly taking lives, damaging property, and forcing people to flee from their homes; you’ve doubtless seen the dire headlines and heartbreaking images on the news. If we could accurately predict these events much further in advance, those headlines would change. Huang hopes Nvidia’s model will be able to predict extreme weather changes in designated regions decades ahead of time. People would then know to either not move to certain areas at all, or to build the infrastructure in those areas in a way that’s compatible with the impending climate events. The model will also aim to help find solutions, running simulations of various courses of actions to figure out which would have the greatest impact at the lowest cost. Nvidia has not shared a timeline for Earth-2’s development nor when the supercomputer will be ready to launch. But if its Cambridge-1 supercomputer for healthcare resea...

Thanks to deep learning, the central mysteries of structural biology are falling like dominos. Just last year, DeepMind shocked the biomedical field with AlphaFold, an algorithm that predicts protein structures with jaw-dropping accuracy. The University of Washington (UW) soon unveiled RoseTTAFold, an AI that rivaled AlphaFold in predictive ability. A few weeks later, DeepMind released a near complete catalog of all protein structures in the human body. Together, the teams essentially solved a 50-year-old grand challenge in biology, and because proteins are at the heart of most medications, they may also have seeded a new era of drug development. For the first time, we have unprecedented insight into the protein engines of our cells, many of which had remained impervious to traditional lab techniques. Yet one glaring detail was missing. Proteins don’t operate alone. They often associate into complexes—small groups that interact to carry out critical tasks in our cells and bodies. This month, the UW team upped their game. Tapping into both AlphaFold and RoseTTAFold, they tweaked the programs to predict which proteins are likely to tag-team and sketched up the resulting complexes into a 3D models. Using AI, the team predicted hundreds of complexes—many of which are entirely new—that regulate DNA repair, govern the cell’s digestive system, and perform other critical biological functions. These under-the-hood insights could impact the next generation of DNA editors and spur new treatments for neurodegenerative disorders or anti-aging therapies. “It’s a really cool result,” said Dr. Michael Snyder at Stanford University, who was not involved in the study, to Science. Like a compass, the results can guide experimental scientists as they test the predictions and search for new insights into how our cells grow, age, die, malfunction, and reproduce. Several predictions further highlighted how our cells absorb external molecules—a powerful piece of information that could help us coerce normally reluctant cells to gulp up medications. “It.gives you a lot of potential new drug targets,” said study author Dr. Qian Cong at the University of Texas Southwestern Medical Center. The Cell’s Lego Blocks Our bodies are governed by proteins, each of which intricately folds into 3D shapes. Like unique Lego bricks, these shapes allow the proteins to combine into larger structures, which in turn conduct the biological processes that propel life. Too abstract? An example: when cells live out their usual lifespan, they go through a process called apoptosis—in Greek, the falling of the leaves—in which the cell gently falls apart without disturbing its neighbors by leaking toxic chemicals. The entire process is a cascade of protein-protein interactions. One protein grabs onto another protein to activate it. The now-activated protein is subsequently released to stir up the next protein in the chain, and so on, eventually causing the aging or diseased cell to sacrifice itself. Another example: in neurons during learning, synapses (the hubs that connect brain cells) call upon a myriad of proteins that form a complex together. This complex, in turn, spurs the neuron’s DNA to make proteins that etch the new memory into the brain. “Everything in biology works in complexes. So, knowing who works with who is critical,” said Snyder. For decades, scientists have relied on painfully slow processes to parse out those interactions. One approach is computational: map out a protein’s structure down to the atomic level and predict “hot spots” that might interact with another protein. Another is experimental: using both biological lab prowess and physics ingenuity, scientists can isolate protein complexes from cells—like sugar precipitating from lemonade when there’s too much of it—and use specialized equipment to analyze the proteins. It’s tiresome, expensive, and often plagued with errors. Here Comes the Sun Deep learning is now shining light on the whole enterprise....

With the pace of emissions reductions looking unlikely to prevent damaging climate change, controversial geoengineering approaches are gaining traction. But aversion to even studying such a drastic option makes it hard to have a sensible conversation, say researchers. Geoengineering refers to large-scale interventions designed to alter the Earth’s climate system in response to global warming. Some have suggested it may end up being a crucial part of the toolbox for tackling global warming, given that efforts to head off warming by reducing emissions seem well behind schedule. One major plank of geoengineering is the idea of removing excess CO2 from the atmosphere, either through reforestation or carbon capture technology that will scrub emissions from industrial exhausts or directly from the air. There are limits to nature-based CO2 removal, though, and so-called “negative emissions technology” is a long way from maturity. The other option is solar geoengineering, which involves deflecting sunlight away from the Earth by boosting the reflectivity of the atmosphere or the planet’s surface. Leading proposals involve injecting tiny particles into the stratosphere, making clouds whiter by spraying sea water into the atmosphere, or thinning out high cirrus clouds that trap heat. In theory, this could reduce global warming fairly cheaply and quickly, but interfering with the Earth’s climate system carries unpredictable and potentially enormous risks. This has led to widespread opposition to even basic research into the idea. Earlier this year, a test of the approach by Sweden’s space agency was cancelled following concerted opposition. But this lack of research means policymakers are flying blind when weighing the pros and cons of the approach, researchers write in a series of articles in the latest issue of Science. They outline why research into the approach is necessary and how social science in particular can help us better understand the potential trade-offs. In an editorial, Edward A. Parson from the University of California, Los Angeles, notes that critics often point to the fact that solar geoengineering is a short-term solution to a long-term problem that is likely to be imperfect and whose effects could be uneven and unjust. More importantly, if solar geoengineering becomes acceptable to use, we may end up over-relying on it and putting less effort into emissions reductions or carbon removal. This point is often used to argue that solar geoengineering can never be acceptable, and therefore research into it isn’t warranted. But Parson argues that both the potential harms and benefits of solar geoengineering are currently hypothetical due to a lack of research. Rejecting an activity due to unknown harms might be justified in extreme circumstances and when the alternative is acceptable, he writes. But the alternative to solar geoengineering is potentially catastrophic climate change—unless we drastically ramp up emissions reductions and removals, which is far from a sure thing. Part of the rationale for preventing solar geoengineering research is that it will drive socio-political lock-in that makes its deployment more likely. But Parson points out that rather than preventing its deployment, blocking research into solar geoengineering may actually lead to less-informed, more dangerous deployments by desperate policymakers further down the line. One way to overcome some of the resistance to research in this area might be to make the debate around it more constructive, writes David W Keith from Harvard University in a policy paper. And the best way to do that is to disentangle the technical, political, and ethical aspects of the debate. Appraising the pros and cons of solar geoengineering involves many different fields, from engineering to climate science to economics. But often, experts in one of these areas will give an overall judgment on the technology despite not being in a position to assess critical aspects of it. The...

Just a few months ago, most of us had never heard of an NFT. Even once we figured out what they were, it seemed like maybe they’d be a short-lived fad, a distraction for the tech-savvy to funnel money into as the pandemic dragged on. But it seems NFTs are here to stay. The internet has exploded with digital artwork that’s being bought and sold like crazy; in the third quarter of this year, trading volume of NFTs hit $10.67 billion, a more than 700-percent increase from the second quarter. Last month, both Coinbase and Sotheby’s announced plans to launch NFT marketplaces. As a quick refresher in case you, like me, still don’t totally get it: NFT stands for non-fungible token, and it’s a digital certificate that represents ownership of a digital asset. The certificates are one-of-a-kind (that’s the non-fungible part), are verified by and stored on a blockchain, and allow digital assets to be transferred or sold. Depending who you ask, NFTs were a thing as early as 2013 or 2014—but they didn’t really hit headlines until earlier this year, when artists like Grimes and Beeple sold their digital creations for millions of dollars. Soon everyone from Jack Dorsey to George Church to the NBA started jumping on the NFT bandwagon. And you’ve probably heard about the bizarre phenomenon that is the Bored Ape Yacht Club. This is just the beginning of an ever-growing list of artists, celebrities, crypto-enthusiasts, and others who are betting NFTs are the future of collectible art. Re-enter Beeple, the American artist (whose given name is Mike Winkelmann) whose collage of 5,000 pieces of digital art, titled Everydays: The First 5000 Days, sold for $69 million in a Christie’s auction in March. Another piece of his sold this week, and though it went for less than half what Everydays did, it’s bringing a whole new twist to the NFT art world. The new work, titled Human One, is a life-sized 3D video sculpture, and Winkelmann called it “the first portrait of a human born in the metaverse.” It shows a person in silver clothing and boots, wearing a backpack and a helmet (which is something of a cross between that of an astronaut and a motorcyclist) trekking purposefully across a changing landscape. It was purchased for $29 million at Christie’s 21st Century Evening Sale on Tuesday by Ryan Zurrer, a Swiss venture capitalist. introducing HUMAN ONE beeple (@beeple) October 28, 2021 The piece is a box whose four walls are video screens, with a computer at its base. It’s over seven feet tall and can be viewed from any angle. But its key feature is the fact that it will be continuously updated, supposedly for the rest of Winkelmann’s life. “I want to make something that people can continue to come back to and find new meaning in. And the meaning will continue to evolve,” he said. “That to me is super-exciting. It feels like I now have this whole other canvas.” The artist plans to change the imagery that appears in the box regularly. It will be sort of like having one of those digital photo frames, except instead of family and friends on a small, flat screen, who-knows-what will appear in 3D and larger than life. If some of the images in Everydays are any indication, Zurrer may end up seeing some pretty striking political commentary in his living room, or office, or wherever he chooses to keep Human One. “You could come downstairs in the morning and the piece looks one way,” Winkelmann said. “Then you come home from work, and it looks another way.” However, he won’t be changing the piece according to any sort of schedule, but rather as the fancy strikes him—and, he noted, in response to current events. If Zurrer chooses to keep the piece in his home or another private location, that would establish a sort of artistic intimacy between him and Winkelmann, with Zurrer being privy to the artist’s ideas and creativity in real time. Though Human One was doubtless an expensive and highly complex project, it’s likely just the beginning of a whole new type of “l...

House prices have soared during the last year and a half, and the implications aren’t great (unless you’re a homeowner looking just to sell and not buy). Homelessness and housing insecurity have risen dramatically over the course of the pandemic, with millions of people unable to afford to live where they want to, and many unable to afford to live anywhere at all. A supply shortage is just one among many factors contributing to these problems, but it’s a big one; houses take a long time to build, require all sorts of permissions and inspections and approvals, and are, of course, expensive. A Seattle-based company wants to be part of changing this, and they’ve just joined forces with a partner to make home building more sustainable and efficient while driving down its costs. Last week, construction tech company NODE, which got its start at Y Combinator, announced a merger with Green Canopy, a vertically-integrated developer, designer, general contractor, and fund manager. The new company’s goal is to offer accessible, green housing options at scale. “The construction industry is ripe for disruption and evolution,” said NODE co-founder Bec Chapin. “It’s a giant industry that has been losing productivity over decades and is not meeting our most crucial demands for housing.” NODE’s approach is similar to that of Las Vegas-based Boxabl, which ships pre-fabricated “foldable” houses to its customers in a 20-foot-wide load that can be up and running in as little as a day. In a 2018 GeekWire interview, Chapin said the company was “developing a component technology in which the walls, floors and ceilings are in separate pieces, as well as all of the things needed to make it a complete house: kitchens, baths, heating systems, etc. These houses can be packed more efficiently, then easily assembled on site.” NODE homes come in flat-pack kits that fit in standard shipping containers, and they don’t require specialists to assemble; they’re essentially the IKEA furniture of houses (though IKEA furniture can, admittedly, be much harder to put together than the company would have you think, as you know if you’ve ever purchased one of their bed frames or shelving units.). Their assembly is guided by software and can be done by generalist construction workers, or even by homeowners themselves. A 2019 McKinsey report noted that modular construction is seeing a comeback, largely thanks to the impact of new digital tools. Consumer perception of prefab housing is becoming more positive as the design of these homes gets more modern and visually appealing. Most importantly, modular construction assisted by digital technologies can make home building up to 50 percent faster, at a cost that’s comparable to or lower than traditional building costs. Green Canopy NODE’s homes are priced between $90,000 (for a 260-square-foot-home that somehow fits in a kitchen and a bathroom) to $150,000 for a 500-square-foot model. These figures are well below the cost of using traditional building methods for homes of comparable size in the company’s native Seattle area. So they seem to be on the right track. But where they’re really looking to set themselves apart from competitors is in their focus on sustainability. “We started Node because buildings account for 47 percent of carbon emissions, yet all of the technology exists for buildings to be carbon negative,” Chapin said. The company’s homes are designed to be carbon neutral or carbon negative; they’re ultra energy-efficient and they use non-toxic materials. Their insulation, for example, is made of recycled denim, glass, and sand instead of fiberglass. The homes can also be outfitted with solar panels or mini wind turbines, and thus could end up generating more energy than they consume, enabling homeowners to sell power back to the grid. The newly-merged company recently raised $10 million in new funding, and expects to double in size over the next year (it currently has 31 employees). Initially focused on the “...

Most people know that the land masses on which we all live represent just 30 percent of Earth’s surface, and the rest is covered by oceans. The emergence of the continents was a pivotal moment in the history of life on Earth, not least because they are the humble abode of most humans. But it’s still not clear exactly when these continental landmasses first appeared on Earth, and what tectonic processes built them. Our research, published in Proceedings of the National Academy of Sciences, estimates the age of rocks from the most ancient continental fragments (called cratons) in India, Australia, and South Africa. The sand that created these rocks would once have formed some of the world’s first beaches. We conclude that the first large continents were making their way above sea level around three billion years ago, much earlier than the 2.5 billion years estimated by previous research. A Three-Billion-Year-Old Beach When continents rise above the oceans, they start to erode. Wind and rain break rocks down into grains of sand, which are transported downstream by rivers and accumulate along coastlines to form beaches. These processes, which we can observe in action during a trip to the beach today, have been operating for billions of years. By scouring the rock record for signs of ancient beach deposits, geologists can study episodes of continent formation that happened in the distant past. The Singhbhum craton, an ancient piece of continental crust that makes up the eastern parts of the Indian subcontinent, contains several formations of ancient sandstone. These layers were originally formed from sand deposited in beaches, estuaries and rivers, which was then buried and compressed into rock. We determined the age of these deposits by studying microscopic grains of a mineral called zircon, which is preserved within these sandstones. This mineral contains tiny amounts of uranium, which very slowly turns into lead via radioactive decay. This allows us to estimate the age of these zircon grains, using a technique called uranium-lead dating, which is well suited to dating very old rocks. The zircon grains reveal that the Singhbhum sandstones were deposited around three billion years ago, making them some of the oldest beach deposits in the world. This also suggests a continental landmass had emerged in what is now India by at least three billion years ago. Interestingly, sedimentary rocks of roughly this age are also present in the oldest cratons of Australia (the Pilbara and Yilgarn cratons) and South Africa (the Kaapvaal Craton), suggesting multiple continental landmasses may have emerged around the globe at this time. Rise Above It How did rocky continents manage to rise above the oceans? A unique feature of continents is their thick, buoyant crust, which allows them to float on top of Earth’s mantle, just like a cork in water. Like icebergs, the top of continents with thick crust (typically more than 45km thick) sticks out above the water, whereas continental blocks with crusts thinner than about 40km remain submerged. So if the secret of the continents’ rise is due to their thickness, we need to understand how and why they began to grow thicker in the first place. Most ancient continents, including the Singhbhum Craton, are made of granites, which formed through the melting of pre-existing rocks at the base of the crust. In our research, we found the granites in the Singhbhum Craton formed at increasingly greater depths between about 3.5 billion and 3 billion years ago, implying the crust was becoming thicker during this time window. Because granites are one of the least dense types of rock, the ancient crust of the Singhbhum Craton would have become progressively more buoyant as it grew thicker. We calculate that by around three billion years ago, the continental crust of the Singhbhum Craton had grown to be about 50km thick, making it buoyant enough to begin rising above sea level. The rise of continents had a profound inf...

When it comes to brain computing, timing is everything. It’s how neurons wire up into circuits. It’s how these circuits process highly complex data, leading to actions that can mean life or death. It’s how our brains can make split-second decisions, even when faced with entirely new circumstances. And we do so without frying the brain from extensive energy consumption. To rephrase, the brain makes an excellent example of an extremely powerful computer to mimic—and computer scientists and engineers have taken the first steps towards doing so. The field of neuromorphic computing looks to recreate the brain’s architecture and data processing abilities with novel hardware chips and software algorithms. It may be a pathway towards true artificial intelligence. But one crucial element is lacking. Most algorithms that power neuromorphic chips only care about the contribution of each artificial neuron—that is, how strongly they connect to one another, dubbed “synaptic weight.” What’s missing—yet tantamount to our brain’s inner working—is timing. This month, a team affiliated with the Human Brain Project, the European Union’s flagship big data neuroscience endeavor, added the element of time to a neuromorphic algorithm. The results were then implemented on physical hardware—the BrainScaleS-2 neuromorphic platform—and pitted against state-of-the-art GPUs and conventional neuromorphic solutions. “Compared to the abstract neural networks used in deep learning, the more biological archetypes.still lag behind in terms of performance and scalability” due to their inherent complexity, the authors said. In several tests, the algorithm compared “favorably, in terms of accuracy, latency, and energy efficiency” on a standard benchmark test, said Dr. Charlotte Frenkel at the University of Zurich and ETH Zurich in Switzerland, who was not involved in the study. By adding a temporal component into neuromorphic computing, we could usher in a new era of highly efficient AI that moves from static data tasks—say, image recognition—to one that better encapsulates time. Think videos, biosignals, or brain-to-computer speech. To lead author Dr. Mihai Petrovici, the potential goes both ways. “Our work is not only interesting for neuromorphic computing and biologically inspired hardware. It also acknowledges the demand . to transfer so-called deep learning approaches to neuroscience and thereby further unveil the secrets of the human brain,” he said. Let’s Talk Spikes At the root of the new algorithm is a fundamental principle in brain computing: spikes. Let’s take a look at a highly abstracted neuron. It’s like a tootsie roll, with a bulbous middle section flanked by two outward-reaching wrappers. One side is the input—an intricate tree that receives signals from a previous neuron. The other is the output, blasting signals to other neurons using bubble-like ships filled with chemicals, which in turn triggers an electrical response on the receiving end. Here’s the crux: for this entire sequence to occur, the neuron has to “spike.” If, and only if, the neuron receives a high enough level of input—a nicely built-in noise reduction mechanism—the bulbous part will generate a spike that travels down the output channels to alert the next neuron. But neurons don’t just use one spike to convey information. Rather, they spike in a time sequence. Think of it like Morse Code: ­the timing of when an electrical burst occurs carries a wealth of data. It’s the basis for neurons wiring up into circuits and hierarchies, allowing highly energy-efficient processing. So why not adopt the same strategy for neuromorphic computers? A Spartan Brain-Like Chip Instead of mapping out a single artificial neuron’s spikes—a Herculean task—the team honed in on a single metric: how long it takes for a neuron to fire. The idea behind “time-to-first-spike” code is simple: the longer it takes a neuron to spike, the lower its activity levels. Compared to counting spikes, it’s an extremely sp...

While getting humans to Mars is likely to be one of the grandest challenges humanity has ever undertaken, getting them back could be even tougher. Researchers think sending genetically engineered microbes to the Red Planet could be the solution. Both NASA and SpaceX are mulling human missions to Mars in the coming decades. But carrying enough fuel to make sure it’s a round trip adds a lot of extra weight, which dramatically increases costs and also makes landing on the planet much riskier. As a result, NASA has been investigating a variety of strategies that would make it possible to produce some or all of the required fuel on Mars using locally-sourced ingredients. While the planet may be pretty barren, its atmosphere is 95 percent carbon dioxide and there is abundant water ice in certain areas. That could provide all the ingredients needed to create hydrocarbon rocket fuels and the liquid oxygen needed to support combustion. The most ambitious of NASA’s plans would be to use electrolysis to generate hydrogen and oxygen from water and then use the Sabatier reaction to combine the hydrogen with Martian CO2 to create methane for use as a fuel. The technology to do that at scale is still immature, though, so the more likely option would see methane shipped from Earth and oxygen generated in place using solid oxide carbon dioxide electrolysis (SOCE). That would still require 7.5 tons of fuel and 1 ton of SOCE equipment to be transported to Mars, though. Researchers from the Georgia Institute of Technology have outlined a new strategy in a paper in Nature Communications, which would use genetically engineered microbes to produce all the fuel and oxygen required for a return trip on Mars. “Carbon dioxide is one of the only resources available on Mars,” first author Nick Kruyer said in a press release. “Knowing that biology is especially good at converting CO2 into useful products makes it a good fit for creating rocket fuel.” The researchers’ proposal involves building four football fields’ worth of photobioreactors—essentially liquid-filled transparent tubes—which will be used to grow photosynthetic cyanobacteria. While it is possible to get these microbes to produce fuels themselves, they are fairly inefficient at it. So instead, they will be fed into another reactor where enzymes will break them down into simple sugars, which are then fed to genetically modified E. coli bacteria that produce a chemical called 2,3-butanediol. On Earth this chemical is primarily used to make rubber, and burns too inefficiently to be used as a fuel. But thanks to Mars’ low gravity, it is more than capable of powering a rocket engine there, and also uses less oxygen than methane. “You need a lot less energy for lift-off on Mars, which gave us the flexibility to consider different chemicals that aren’t designed for rocket launch on Earth,” said Pamela Peralta-Yahya, who led the research. The process also generates 44 tons of excess oxygen that could be used for life support. The one catch is that if the system was built with today’s state-of-the-art technology, it would require 2.8 times as much material to be delivered to Mars compared to the most likely NASA strategy. However, once there it would use 32 percent less power, and resupply missions would only need to carry 3.7 tons of nutrients and chemicals rather than 6.5 tons of methane every time. And modeling studies suggest that by optimizing the biological processes involved and designing lighter-weight materials, a future system could actually weigh 13 percent less than the NASA solution and use 59 percent less power. The biggest barrier at the minute might be the fact that current NASA regulations prohibit sending microbes to Mars due to fears of contaminating the pristine environment. The researchers acknowledge that they will have to develop foolproof biological containment strategies before the proposal could be seriously considered. But if we want to make round trips to Mars a regular f...

AI research wunderkind, DeepMind, has long been all fun and games. The London-based organization, owned by Google parent company Alphabet, has used deep learning to train algorithms that can take down world champions at the ancient game of Go and top players of the popular strategy video game Starcraft. Then last year, things got serious when DeepMind trounced the competition at a protein folding contest. Predicting the structure of proteins, the complex molecules underpinning all biology, is notoriously difficult. But DeepMind’s AlphaFold2 made a quantum leap in capability, producing results that matched experimental data down to a resolution of a few atoms. In July, the company published a paper describing AlphaFold2, open-sourced the code, and dropped a library of 350,000 protein structures with a promise to add 100 million more. This week, Alphabet announced it will build on DeepMind’s AlphaFold2 breakthrough by creating a new company, Isomorphic Labs, in an effort to apply AI to drug discovery. “We are at an exciting moment in history now where these techniques and methods are becoming powerful and sophisticated enough to be applied to real-world problems including scientific discovery itself,” wrote Demis Hassabis, DeepMind founder and CEO, in a post announcing the company. “Now the time is right to push this forward at pace, and with the dedicated focus and resources that Isomorphic Labs will bring.” Hassabis is Isomorphic’s founder and will serve as its CEO while the fledgling company gets its feet, setting the agenda and culture, building a team, and connecting the effort to DeepMind. The two companies will collaborate, but be largely independent. “You can think of [Isomorphic] as a sort of sister company to DeepMind,” Hassabis told Stat. “The idea is to really forge ahead with the potential for computational AI methods to reimagine the whole drug discovery process.” While AlphaFold2’s success sparked the effort, protein folding is only one step—arguably simpler than others—in the arduous drug discovery process. Hassabis is thinking bigger. Though details are scarce, it appears the new company will build a line of AI models to ease key choke points in the process. Instead of identifying and developing drugs themselves, they’ll sell a platform of models as a service to pharmaceutical companies. Hassabis told Stat these might tackle how proteins interact, the design of small molecules, how well molecules bind, and the prediction of toxicity. That the work will be separated from DeepMind itself is interesting. The company’s not insignificant costs have largely been dedicated to pure research. DeepMind turned its first profit in 2020, but its customers are mostly Alphabet companies. Some have wondered if it’d face more pressure to focus on commercial products. The decision to create a separate enterprise based on DeepMind research seems to indicate that’s not yet the case. If it can keep pushing the field ahead as a whole, perhaps it makes sense to fund a new organization—or organizations, seeded by future breakthroughs—as opposed to diverting resources from DeepMind’s more foundational research. Isomorphic Labs has plenty of company in its drug discovery efforts. In 2020, AI in cancer, molecular, and drug discovery received the most private investment in the field, attracting over $13.8 billion, more than quadruple 2019’s total. There have been three AI drug discovery IPOs in the last year, and mature startups—including Exscientia, Insilico Medicine, Insitro, Atomwise, and Valo Health—have earned hundreds of millions in funding. Companies like Genentech, Pfizer, and Merck are likewise working to embed AI in their processes. To a degree, Isomorphic will be building its business from the ground up. AlphaFold2 is without a doubt a big deal, but protein modeling is the tip of the drug discovery iceberg. Also, while AlphaFold2 had the benefit of access to hundreds of thousands of freely available, already modeled protein s...

Four and a half years ago, a robot named Flippy made its burger-cooking debut at a fast food restaurant called CaliBurger. The bot consisted of a cart on wheels with an extending arm, complete with a pneumatic pump that let the machine swap between tools: tongs, scrapers, and spatulas. Flippy’s main jobs were pulling raw patties from a stack and placing them on the grill, tracking each burger’s cook time and temperature, and transferring cooked burgers to a plate. This initial iteration of the fast-food robot—or robotic kitchen assistant, as its creators called it—was so successful that a commercial version launched last year. Its maker Miso Robotics put Flippy on the market for $30,000, and the bot was no longer limited to just flipping burgers; the new and improved Flippy could cook 19 different foods, including chicken wings, onion rings, french fries, and the Impossible Burger. It got sleeker, too: rather than sitting on a wheeled cart, the new Flippy was a “robot on a rail,” with the rail located along the hood of restaurant stoves. This week, Miso Robotics announced an even newer, more improved Flippy robot called Flippy 2 (hey, they’re consistent). Most of the updates and improvements on the new bot are based on feedback the company received from restaurant chain White Castle, the first big restaurant chain to go all-in on the original Flippy. So how is Flippy 2 different? The new robot can do the work of an entire fry station without any human assistance, and can do more than double the number of food preparation tasks its older sibling could do, including filling, emptying, and returning fry baskets. These capabilities have made the robot more independent, eliminating the need for a human employee to step in at the beginning or end of the cooking process. When foods are placed in fry bins, the robot’s AI vision identifies the food, picks it up, and cooks it in a fry basket designated for that food specifically (i.e., onion rings won’t be cooked in the same basket as fish sticks). When cooking is complete, Flippy 2 moves the ready-to-go items to a hot-holding area. Miso Robotics says the new robot’s throughput is 30 percent higher than that of its predecessor, which adds up to around 60 baskets of fried food per hour. So much fried food. Luckily, Americans can’t get enough fried food, in general and especially as the pandemic drags on. Even more importantly, the current labor shortages we’re seeing mean restaurant chains can’t hire enough people to cook fried food, making automated tools like Flippy not only helpful, but necessary. “Since Flippy’s inception, our goal has always been to provide a customizable solution that can function harmoniously with any kitchen and without disruption,” said Mike Bell, CEO of Miso Robotics. “Flippy 2 has more than 120 configurations built into its technology and is the only robotic fry station currently being produced at scale.” At the beginning of the pandemic, many foresaw that Covid-19 would push us into quicker adoption of many technologies that were already on the horizon, with automation of repetitive tasks being high on the list. They were right, and we’ve been lucky to have tools like Zoom to keep us collaborating and Flippy to keep us eating fast food (to whatever extent you consider eating fast food an essential activity; I mean, you can’t cook every day). Now if only there was a tech fix for inflation and housing shortages. Seeing as how there’ve been three different versions of Flippy rolled out in the last four and a half years, there are doubtless more iterations coming, each with new skills and improved technology. But the burger robot is just one of many new developments in automation of food preparation and delivery. Take this pizzeria in Paris: there are no humans involved in the cooking, ordering, or pick-up process at all. And just this week, IBM and McDonald’s announced a collaboration to create drive-through lanes run by AI. So it may not be long before you c...