MIT’s CSAIL researchers have designed an innovative approach using AI models to explain the behavior of other systems, such as large neural networks. Their method involves “automated interpretability agents” (AIA) that generate intuitive explanations and the “function interpretation and description” (FIND) benchmark for evaluating interpretability procedures. This advancement aims to make AI systems more understandable…
MIT neuroscientists used an artificial language network to identify which sentences activate the brain’s language processing centers. They found that more complex or unusual sentences elicit stronger responses, while straightforward or nonsensical sentences barely engage these regions. The study suggests that linguistic properties such as surprisal and complexity influence brain activation. The research was funded…
Artificial intelligence has strong potential to impact diverse fields. The MIT panel explored the implications of generative AI for art and design. The discussion focused on AI’s role in fostering ambiguity, creating tangible experiences, and managing expectations. The panelists emphasized the need to consider AI’s impact on creativity, biases, and human understanding of technology.
Irene Terpstra ’23 and Rujul Gandhi ’22, two MIT engineering students, are leveraging natural language for AI systems. Terpstra’s team is using language models to assist in chip design, while Gandhi is developing a system to convert natural language instructions for robots. Gandhi is also working on speech models for low-resource languages, seeing potential in…
MIT had a remarkable year in 2023, from President Sally Kornbluth’s inauguration to breakthroughs in various fields. Highlights include AI developments, Nobel Prize wins, climate innovations, and advancements in health and art. MIT remained at the forefront of cutting-edge research, positioning itself as a leader in science and technology.
Using deep learning, MIT researchers have discovered compounds with high potential to kill drug-resistant bacteria like MRSA. These compounds demonstrate low toxicity against human cells, making them strong drug candidates. MIT’s Antibiotics-AI Project aims to find new antibiotics using deep learning models, and the research has been published in Nature. The project received funding from…
MIT researchers have introduced a new technique that gives artists greater control over animations in movies and video games. Using mathematical functions called barycentric coordinates, the method allows artists to define how 2D and 3D shapes move and bend in space, providing flexibility and a more natural look. The approach has potential applications in various…
MIT researchers have discovered that image recognition difficulty for humans has been overlooked, despite its importance in fields like healthcare and transportation. They developed a new metric called “minimum viewing time” (MVT) to measure image recognition difficulty, showing that existing datasets favor easy images. Their work could lead to more robust and human-like performance in…
MIT researchers have developed a fast machine-learning-based method to calculate transition states in chemical reactions. The new approach can predict transition states accurately and quickly, in contrast to the time-consuming quantum chemistry techniques. The model can aid in designing catalysts and understanding natural reactions, potentially impacting fields like pharmaceutical synthesis and astrochemistry.
The MIT Energy and Climate Hack brought together students from various fields to find rapid solutions for the global energy and climate crisis. Companies presented challenges, and teams had two days to develop solutions, with AI emerging as a valuable tool. The event highlighted the need for cooperation and diverse expertise in addressing climate change.…
The MIT-Pillar AI Collective has selected three fellows for fall 2023. They are pursuing research in AI, machine learning, and data science, with the goal of commercializing their innovations. The Fellows include Alexander Andonian, Daniel Magley, and Madhumitha Ravichandra, each working on innovative projects in their respective fields as part of the program’s mission to…
MIT researchers have found that modern computational models derived from machine learning are approaching the goal of mimicking the human auditory system. The study, led by Josh McDermott, emphasizes the importance of training these models with auditory input, including background noise, to closely match the activation patterns of the human auditory cortex. The research aims…
Researchers from MIT and the Chinese University of Hong Kong have developed a technique called neural lithography, using real-world data to build a photolithography simulator that can more accurately model the manufacturing process of optical devices. This approach could lead to the creation of more efficient optical devices for various applications.
Greek mathematician Euclid, known as the father of geometry, revolutionized the understanding of shapes over 2,000 years ago. Today, MIT professor Justin Solomon applies modern geometric techniques to diverse problems, from machine-learning model testing to medical imaging and generative AI. He fosters diversity in geometric research and aims to improve unsupervised machine learning models.
MIT Generative AI Week featured a flagship full-day symposium and four subject-specific symposia, aiming to foster dialogue about generative artificial intelligence technologies. The events included panels, roundtable discussions, and keynote speeches, covering topics such as AI and education, health, creativity, and commerce. The week concluded with a screening of the documentary “Another Body,” followed by…
MIT leaders and scholars release policy briefs outlining a framework for U.S. artificial intelligence (AI) governance, aiming to enhance U.S. leadership and limit potential harm. The approach involves extending current regulatory and liability approaches and emphasizes identifying the purpose and intent of AI tools. The project aims to address various regulatory challenges in the AI…
MIT researchers developed an automated onboarding system that improves human-AI collaboration accuracy by training users when to trust AI assistance. Their method uses natural language to teach rules based on the user’s past interactions with AI, leading to a 5% improvement in image prediction tasks.
MIT researchers have developed an Automatic Surface Reconstruction framework using machine learning to design new compounds or alloys for catalysts without reliance on chemist intuition. The method provides dynamic, thorough characterization of material surfaces, revealing previously unidentified atomic configurations. It operates more cost-effectively, efficiently, and is available for global use.
At the “Generative AI: Shaping the Future” symposium, keynote speaker Rodney Brooks highlighted the risk of overhyping AI’s capabilities, emphasizing the need for responsible development. The event at MIT included discussions on generative AI’s potential for positive impact, collaborative research, and the importance of ethical integration into society.
Researchers from MIT, Harvard University, and the University of Washington have developed a new approach to reinforcement learning that leverages feedback from nonexpert users to teach AI agents specific tasks. Unlike other methods, this approach enables the agent to learn more quickly despite the noisy and potentially inaccurate feedback. The method has the potential to…