Researchers at MIT have released a new working paper that aims to quantify the severity and speed with which AI systems could replace human workers, reports Hiawatha Bray for The Boston Globe. The paper concluded that “it’s not enough for AI systems to be good at tasks not performed by people,” explains Bray. “The system must be good enough to justify the cost of installing it and redesigning the way a job is done.”
A new working paper by MIT researchers predicts “only 23% of wages linked to vision-related tasks could be feasibly cost-effectively replaced by AI,” reports Gil Press for Forbes. The researchers “argue that their findings apply also to generative AI or the automation of language-related tasks,” writes Press.
Wall Street Journal reporter Lindsey Choo spotlights Principal Research Scientist Matthias Winkenbach and his work developing an AI model to help delivery drivers find the best routes. The model would “take into consideration complex real-world constraints,” such as allowing drivers to, “choose a route that may not be the shortest but allows them to park more conveniently or unload packages in safer spaces,” writes Choo.
Researchers at MIT have discovered how a new computational imaging algorithm can capture user interactions through ambient light sensors commonly found in smartphones, reports Davey Winder for Forbes. “By combining the smartphone display screen, an active component, with the ambient light sense, which is passive, the researchers realized that capturing images in front of that screen was possible without using the device camera,” explains Winder.
Writing for Wired, Institute Prof. Daron Acemoglu predicts that expectations for generative AI will need to recalibrated during the year ahead. Acemoglu notes that he believes in 2024, “generative AI will have been adopted by many companies, but it will prove to be just ‘so-so automation’ of the type that displaces workers but fails to deliver huge productivity improvements.”
Writing for Wired, research scientist Kate Darling highlights the importance of addressing the fundamentally human behaviors that have been incorporated into AI chatbots. “Research in human-computer and human-robot interaction shows that we love to anthropomorphize—attribute humanlike qualities, behaviors, and emotions to—the nonhuman agents we interact with, especially if they mimic cues we recognize,” writes Darling. “And, thanks to recent advances in conversational AI, our machines are suddenly very skilled at one of those cues: language.”
Philip Adama Abel '15 co-founded Cleva, “a banking platform for African individuals and businesses to receive international payments by opening USD accounts,” reports Tage Kene-Okafor TechCrunch. “Long term, we are open to Cleva evolving from just being a product-only service to being a platform issuing APIs to do a bunch of other things that help us distribute services across other African countries or around the world,” says Abel.
MIT researchers have used machine learning to uncover the different kinds of sentences that most likely to activate the brain’s key language processing centers, reports Kyle Wiggers and Devin Coldewey for TechCrunch. The model, “was able to predict for novel sentences whether they would be taxing on human cognition or not,” they explain.
Prof. David Autor speaks with Bloomberg about the future of generative AI and the technology’s potential impact on productivity and the labor market. “When we interact with AI, we need to learn how to treat it not as authoritative, but as a guide to support decision making, and that’s really critical,” says Autor.
MIT researchers have found that job applicants who used AI to improve spelling and grammar on their resumes were eight percent more likely to be hired, reports Bryan Robinson for Forbes. “If you take two identical workers with the same skills and background, the one with the better-written resume is more likely to get hired,” explains graduate student Emma van Inwegen. “The takeaway is that employers actually care about the writing in the resume; it’s not just a correlation.”
Researchers at MIT and elsewhere developed an artificial intelligence predictive model that can be used to detect which strains of Covid-19 could become dominant and lead to a new wave of illness, reports Ruth Reader, Carmen Paun, Daniel Payne and Eric Schumaker for Politico. The model, “found three strong predictors of a dominant variant: the number of infections a strain causes in its first week relative to the number of times it appears in sequencing, the number of mutations in the spike protein, and the number of weeks since the current dominant variant began circulating,” they note.
Former Postdoc Rana el Kaliouby speaks with Boston Globe reporter Janelle Nanos about her work focused on infusing emotional intelligence into AI. “I’m an optimist, but I’m a realist as well, and I can see where this can go wrong,” says Kaliouby. “I think one of the biggest concerns around AI is bias and how we’re building bias into these models and then deploying them at scale everywhere around the world.”
Prof. Manolis Kellis speaks with Brian Bergstein of The Boston Globe’s “Say More” podcast about the potential influence of artificial intelligence on biological advances and medicine. “I would say there is no way to answer biological questions without computational science nowadays,” says Kellis. “The genome and all of the technologies that it enabled transformed biology into an information-based discipline.”
Dynamic Labs, co-founded by Itai Turbahn '11 and Yoni Goldberg '09, MEng '10, offers “tech for crypto and non-crypto companies alike to create seamless login experiences backed by digital wallets,” reports Marco Quiroz-Gutierrez for Fortune. “The firm’s wallet-based business is twofold,” explains Quiroz-Gutierrez. “It offers a customizable all-in-one service that can either push the crypto wallet technology to the background for less savvy users or put it front and center for Web3 natives.”
Prof. Manolis Kellis speaks with USA Today reporter Josh Peter about the potential impact of AI in developing undetectable performance-enhancing drugs (PEDs). "The most feasible approach would be using generative AI to alter existing PEDs that trigger drug tests in a way that makes those drugs undetectable by current testing technology,” Kellis notes.