Graduate student Matthew Groh discusses Detect Fakes, a research project he co-created aimed at teaching people how to detect deepfakes, with Wall Street Journal reporter Ann-Marie Alcántara. Groh recommends people pay attention to the context of an image or video, noting that people can “pay attention to incentives and what someone is saying and why someone might be saying this.”
Kevin Hu SB ’13, SM ’15, PhD ’19 co-founded Metaplane, a startup aimed at providing users with data analytics-focused tools, reports Kyle Wiggers for TechCrunch. “Metaplane monitors data using anomaly detection models trained primarily on historical metadata. The monitors try to account for seasonality, trends and feedback from customers, Hu says, to minimize alert fatigue, “writes Wiggers.
A review led Prof. Marzyeh Ghassemi has found that a major issue in health-related machine learning models “is the relative scarcity of publicly available data sets in medicine,” reports Emily Sohn for Nature.
Researchers from the MIT-IBM Watson AI Lab and the Harvard Natural Language Processing Group developed the Giant Language model Test Room (GLTR), an algorithm that attempts to detect if text was written by a bot, reports Megan Morrone for Fast Company. “Using the ‘it takes one to know one’ method, if the GLTR algorithm can predict the next word in a sentence, then it will assume that sentence has been written by a bot,” explains Morrone.
Research scientist Ryan Hamerly and his team are working to harness “the low power consumption of hybrid optical devices for smart speakers, lightweight drones and even self-driving cars,” reports The Economist.
MIT spinout Gaia A is developing a forest management building tool aimed at providing foresters with the resources to make data-driven decisions, reports Haje Jan Kamps and Brian Heater for TechCrunch. “The company is currently using lidar and computer vision tech to gather data but is ultimately building a data platform to tackle some of the big questions in forestry,” writes Kamps and Heater.
Research affiliate Ramin Hasani speaks with Kimberly Adams of Marketplace about how he and his CSAIL colleagues solved a differential equation dating back to the early 1900s, enabling researchers to create an AI algorithm that can learn on the spot and adapt to evolving patterns. The new algorithm “will enable larger-scale brain simulations,” Hasani explains.
Fortune reporter Gabby Shacknai spotlights Joy Buolamwini PhD ’22 and her research in racial bias in AI. “After finishing grad school, Biolamwini decided to continue her research on A.I.’s racial bias and quickly realized that much of this was a result of the non-diverse datasets and imagery used by a disproportionately white, male tech workforce to train A.I. and inform its algorithms,” writes Shacknai.
Graduate student Kevin Frans co-founded OpenAI, a for-profit research lab that aims to provide free public access to artificial intelligence systems, reports Hiawatha Bray for The Boston Globe. “Our mission is to put AI is the hands of everyone,” says Frans.
Popular Science reporter Charlotte Hu writes that MIT researchers have developed a new machine learning model that can depict how the sound around a listener changes as they move through a certain space. “We’re mostly modeling the spatial acoustics, so the [focus is on] reverberations,” explains graduate student Yilun Du. “Maybe if you’re in a concert hall, there are a lot of reverberations, maybe if you’re in a cathedral, there are many echoes versus if you’re in a small room, there isn’t really any echo.”
MIT researchers have developed a new technique aimed at protecting images from AI generators, reports Kyle Barr for Gizmodo. The program uses "data poisoning techniques to essentially disturb pixels within an image to create invisible noise, effectively making AI art generators incapable of generating realistic deepfakes based on the photos they’re fed,” reports Kyle Barr for Gizmodo."
Researchers from MIT and other institutions have found that an AI system called KataGo can be consistently tricked into losing at the strategy game of Go, reports Jeremy Kahn for Fortune. This research highlights potential similar vulnerabilities in “other AI systems trained in a similar manner,” writes Kahn.
Scientists at MIT have developed “a machine learning model that can capture how sounds in a room will propagate through space,” report Kyle Wiggers and Devin Coldewey for TechCrunch. “By modeling the acoustics, the system can learn a room’s geometry from sound recordings, which can then be used to build a visual rendering of a room,” write Wiggers and Coldewey.
Writing for The Wall Street Journal, MIT Prof. Katherine Kellogg and Stanford Prof. Melissa Valentine explore the challenges of introducing AI technologies in the workplace, focusing on the fashion industry. "Getting workers to actually use the technologies will turn out to be just as important as making sure the systems work in the first place," they write.
MIT researchers have developed a new machine learning model that can identify and track blobs of plasma created in controlled nuclear-fusion research, reports Ed Browne for Newsweek. “Fusion research is a complex, multidisciplinary project that requires technologies from many fields,” explains graduate student Woonghee “Harry” Han.