School of Engineering

Instead of giving AI agents a specific task, MIT researchers put 13 agents, named Eko, alone on a digital desert island for 30 hours, and witnessed the technology playing instead of simply wandering around, writes Fast Company’s Jude Cramer. “The researchers compared Eko’s behavior to common elements of play in animals, including humans,” notes Cramer. “These signs of play include creating their own rules, repeating behaviors with slight variations, setting goals for themselves, and transforming objects and events through make-believe.” 

After observing the behavior of 13 AI agents on a simulated desert island, MIT researchers suggest that allowing agents to explore “without a goal” could be an effective way to teach AI agents new skills, reports Aditya Mehta for TechCrunch. “We normally develop artificial agents by deciding in advance what they should practice, what they should optimize, and what counts as progress,” the researchers write. “Play suggests another regime: give an agent a sufficiently rich and safe world, let it choose what is interesting, and allow useful knowledge to accumulate from the history that follows.” 

Flapping Airplanes, a startup co-founded by Benjamin Spector ’22, MEng ’22, has developed a framework for AI models to run efficiently while being trained on less data and in a single graphics processing unit (GPU), writes Julie Bort for TechCrunch. The framework “can run a task that spends most of its time waiting for data to be fetched alongside one that spends most of its time doing math, so the chip is rarely sitting idle,” notes Bort. “The premise is that AI has barely scratched the surface of what GPUs can do, and they believe it can create intelligent systems ‘without ingesting half the internet.’”  

Eka Robotics, a startup cofounded by Prof. Pulkit Agrawal, has developed a robot with  a force-sensitive gripper that can “pick up about 1,500 objects that weigh up to 1 kg in an hour, compared to about 600 for a human” with 95% accuracy, reports Tim Fernholz for TechCrunch.  “We are starting off being very pragmatic, and saying here are two verticals in pick, arrange, place, and wire handling, let’s make them be home runs,” says Agrawal. “Once we have made home runs on them, the sky is the limit in how much we can do with the approach we are developing.” 

For Manufacturing Dive, reporter Nathan Owens spotlights MIT’s Technologist Advanced Manufacturing Program (TechAMP), a 12-mong-long program that combines online instruction with hands-on lab work to bridge the gap between technical engineering and management skills, designed for working technicians or operators. “Students gain foundational leadership, process control and systems thinking skills,” notes Owens. TechAMP has “really shown me how to communicate more, communicate problems more, communicate solutions more,” says Brian Coppenrath, who completed the program.  

The New Yorker’s Gideon Lewis-Kraus details how Morris Chang ’52, SM ’53 founded the Taiwan Semiconductor Manufacturing Company (TSMC), “one of the most valuable tech companies globally” that “makes almost all of the world’s advanced [computer] chips.” “Chang proposed a semiconductor ‘foundry,’ which would do for pricey chips what its predecessors had done for cheap junk. At the time, however, the market didn’t exist,” writes Lewis-Kraus. “By the end of TSMC’s first decade, the company was bringing in a billion dollars in annual revenue.”  

Financial Times reporter Clive Cookson details how the late Margaret Hamilton engineered code that would play a crucial role in achieving the first moon landing, at MIT’s Instrumentation Lab. When Apollo 11’s onboard computer flashed alarms minutes from landing on the moon, “Hamilton and colleagues had engineered the software to dump non-essential background tasks under those circumstances and prioritize procedures needed for the landing,” writes Cookson. “NASA controllers in Houston trusted the software, so they decided not to abort the mission, allowing Neil Armstrong and Buzz Aldrin to touch down safely in their Eagle module.”  

Fast Company’s Adele Peters highlights Firma Power, a clean energy developer co-founded by MIT alumni aimed at working with data centers to use clean alternatives to gas power. “We started the company with the hypothesis that despite the fact that 98% of what’s in the queue is wind, solar and storage, that’s not getting tapped as the default path to power for large loads—so there’s got to be some market frictions and barriers there,” says Jesse Jenkins SM ’14, PhD ’18.  

Prof. Jacob Andreas speaks with CNN’s Lisa Eadicicco about deciphering the chain-of-thought behind complex AI neural networks. “I think we’re still very far from being able to claim that we have any sort of fine-grained understanding of how the model is actually arriving at a solution,” says Andreas.  

MIT researchers gave 13 AI agents, named Eko, free rein, on a simulated desert island for 30 hours, and observed as the agents played and learned from their surroundings like humans, writes Gizmodo’s Webb Wright. “We had this mindset to have as [few] constraints as possible, because we really wanted to see what the behavior of these autonomous agents is when you don’t really tell them what to do,” says graduate student Nathan Cloos. “Play is really about curiosity or novelty-seeking behavior, and this is essentially what Eko does,” Prof. Daniela Rus adds.  

Writing for Nature, Deputy Editorial Director Maia Weinstock honors the life and legacy of the late Margaret Hamilton, a pioneer of the software engineering field, who, at MIT’s Instrumentation Lab, led the team that wrote and tested software for the Apollo Guidance Computer onboard Apollo 11. Hamilton’s story “was also my inspiration for one of the vignettes in a Lego set that I created in 2016, featuring Hamilton and several other women in NASA history,” writes Weinstock. “She continued work at [her Universal Systems Language company] Hamilton Technologies well into her 80s, never failing to inspire me and countless others with her tenacity, humour, warmth and childlike wonder.”

At the STAT Summit, President Sally Kornbluth joined Holden Thorp, editor-in-chief of the Science family of journals, for a panel moderated by STAT’s Anil Oza. During the conversation, Kornbluth emphasized the importance of curiosity-driven research and education at institutions like MIT, highlighting MIT for America, a new initiative aimed at ensuring learners and educators from Kindergarten through community college across the country have access to STEM education. “We want to both meet people where they are, give their kids the opportunity, and also make them understand us as people who care about their communities,” says Kornbluth.

MIT researchers have developed the Probabilistic Bias Correction (PBC) tool, which “uses machine learning to identify systematic bias in a model and make corrections” to improve weather forecasts, reports The Verge’s Justine Calma. “The researchers found that PBC improved the accuracy of a leading physics-based weather model by the European Centre for Medium-Range Weather Forecasts (ECMWF) for 98 percent of precipitation and 92 percent of temperature targets,” notes Calma. “The machine learning framework also doubled the accuracy of ECMWF’s new AI forecasting system in the study’s analysis.” 

Gizmodo’s Ellyn Lapointe spotlights the Probabilistic Bias Correction (PBC) tool developed by MIT researchers to “correct systemic errors in forecast models and improve the accuracy of their subseasonal outlooks.” “Extreme weather has the largest impact on our society and improved forecasts of these events can yield large societal benefits,” says Research Scientist Judah Cohen. “Interestingly, previous bias correction schemes actually made forecasts of extreme weather worse, and our model showed clear improvement of extreme weather forecasts.”  

MIT researchers have developed an AI system “that not only minimizes forecasting errors, but learns from its inevitable mistakes through a method called postprocessing,” writes reporter Andrew Paul for Popular Science. “Their AI postprocessing integration blew the existing forecasting models out of the water,” notes Paul. “Such massive improvements could soon help the farmers and energy providers who already rely on subseasonal forecasts, but the benefits may even save lives.”