When it comes to artificial intelligence, MIT and IBM have been there from the beginning: laying the groundwork and creating some of the first programs—AI's predecessors—and theorizing how machine “intelligence” might arise.
Today, partnerships like the MIT-IBM Watson AI Lab that began eight years ago continue to provide expertise for tomorrow's AI technologies. This is critical for the industries and workforces that stand to benefit, especially in the short term: from $3-4 trillion in projected global economic benefits and an 80 percent increase in productivity for knowledge workers and creative tasks, to the significant incorporation of generative AI into business processes (80 percent) and applications (70 percent) over the next three years.
While the industry has seen a boom in notable models, mainly in the last year, academia continues to drive innovationcontributing to most of the highly cited research. Success in the MIT-IBM Watson AI Lab includes 54 patent disclosures, over 128,000 citations with an h-index of 162, and over 50 industry-driven use cases. Some of the lab's many achievements include improved stent placement using AI imaging techniques, reduced computational overhead, model shrinkage while maintaining performance, and interatomic potential modeling for silicate chemistry.
“The lab has a unique ability to identify the 'right' problems to solve, which sets us apart from peers,” says Aude Oliva, director of the MIT Lab and director of strategic industry engagement at the MIT Schwarzman College of Computing. “Furthermore, the experience our students gain working on enterprise AI challenges translates on their competitiveness on the labor market and the promotion of a competitive industry.
“The MIT-IBM Watson AI Lab has had a profound impact by bringing together a rich set of collaborations between IBM and MIT researchers and students,” says Provost Anantha Chandrakasan, co-chair of the MIT lab and Vannevar Bush Professor of Electrical Engineering and Computer Science. “By supporting cross-cutting research at the intersection of artificial intelligence and many other disciplines, the lab is advancing fundamental work and accelerating the development of transformative solutions for our nation and the world.”
Long-term work
As artificial intelligence continues to gain attention, many organizations are struggling to turn the technology into meaningful results. AND Gartner Study 2024 states that “at least 30% of generative AI projects will be abandoned after proof of concept by the end of 2025”, which demonstrates the ambition and widespread hunger for AI, but the lack of knowledge on how to develop and apply it to create immediate value.
This is where the lab shines, combining research and implementation. The majority of the lab's research portfolio for the current year is focused on the exploitation and development of new features, capabilities or products for IBM, the lab's corporate members or real-world applications. The latter include large language models, AI hardware, and foundational models including multimodal, biomedical, and geospatial. Inquiry-minded students and trainees are invaluable in this pursuit, offering enthusiasm and new perspectives while accumulating domain knowledge to help generate and engineer advances in the field, as well as opening new frontiers of exploration using artificial intelligence as a tool.
Arrangements from AAAI 2025 Presidential Panel on the Future of Artificial Intelligence Research support the need for input from academia-industry collaborations, such as labs in the AI arena: “Researchers have a role to play in providing independent advice and interpretation of these results (from industry) and their implications. The private sector is focused more on the short-term, and universities and society more on the long-term.”
Combining these strengths with an emphasis on open information and open science can spur innovation that neither could achieve on their own. History shows that adopting these principles, sharing code and sharing research brings long-term benefits to both the sector and society. Consistent with the missions of IBM and MIT, the laboratory through this collaboration shares technologies, findings, governance, and standards in the public sphere, thereby increasing transparency, accelerating reproducibility, and ensuring credible progress.
Created to combine MIT's deep research expertise with IBM's industrial R&D capabilities, the lab aims to drive breakthroughs in fundamental artificial intelligence methods and hardware as well as new applications in areas such as health care, chemistry, finance, cybersecurity, and robust business planning and decision-making.
Bigger isn't always better
Today, large entry-level models are giving way to smaller, more specialized models that provide better performance. Contributions from lab staff such as Song Han, associate professor in MIT's Department of Electrical Engineering and Computer Science (EECS), and IBM Research's Chuang Gan have made this possible through work such as once and for all AND AWQ. Such innovations increase performance through better architectures, algorithm shrinkage, and activation-aware weight quantization, allowing models such as language processing to run at the edge with faster speeds and lower latency.
As a result, the core, vision, multimodal, and multi-language models have delivered benefits, enabling lab research groups consisting of Oliva, MIT EECS Associate Professor Yoon Kim, and IBM Research Fellows Rameswar Panda, Yang Zhang, and Rogerio Feris to build on their work. This includes techniques to infuse models with external knowledge and the development of linear attention transformer methods to achieve higher throughput compared to other state-of-the-art systems.
Understanding and reasoning in vision and multimodal systems also provided benefits. It works like “Task 2Sim“And”Fuse” demonstrated the improved performance of a vision model when pre-training is performed on synthetic data, and how video action recognition can be improved by combining feeds from past and current feature maps.
As part of the improved AI effort, the lab teams of Gregory Wornell, professor of engineering at MIT EECS Sumitomo Electric Industries, Chuang Gan of IBM Research, and David Cox, vice president of core artificial intelligence at IBM Research and IBM lab director, have demonstrated that model adaptability and data performance can go hand in hand. Two approaches, EvoSkala AND Reasoning based on a chain of actions and thoughts (COAT) enable language models to make the most of limited data and computation by improving on previous generation attempts through structured iteration, narrowing down a better answer. COAT uses a meta-action structure and reinforcement learning to tackle reasoning-intensive tasks through self-correction, while EvoScale brings a similar philosophy to code generation, developing high-quality potential solutions. These techniques help enable resource-aware, targeted real-world implementation.
“The impact of MIT-IBM research on our efforts to develop large language models cannot be overstated,” says Cox. “We see smaller, more specialized models and tools having a huge impact, especially when they are combined. Innovations from the MIT-IBM Watson AI Lab are helping to shape these technical directions and influence the strategy we execute in the marketplace through platforms like watsonx.”
For example, numerous laboratory projects have contributed features, capabilities and applications to IBM solutions Granite visionwhich, despite its small size, delivers impressive computer vision designed for document understanding. This is happening at a time when there is a growing need to extract, interpret and reliably summarize information and data contained in long formats for enterprise needs.
Other developments that go beyond direct AI research and cross-disciplinary research are not only beneficial but also necessary to advance technology and elevate society, the AAAI 2025 panel concluded.
The work of Caroline Uhler and Devavrat Shah in the lab – Andrew (1956) and Erna Viterbi Professors at EECS and the Institute for Data, Systems and Society (IDSS) – along with Kristjan Greenewald at IBM Research transcends specializations. They develop causal discovery methods to discover how interventions influence outcomes and determine which interventions produce the desired outcomes. The research involves developing a framework that can either explain how “treatments” for different subpopulations might work, e.g. on an e-commerce platform, or mobility constraints in relation to morbidity outcomes. Findings from this body of work may have implications for the fields of marketing and medicine, education and risk management.
“Advances in artificial intelligence and other areas of computer science are influencing the way people formulate and solve challenges in almost every discipline. At the MIT-IBM Watson AI Lab, researchers recognize the cross-sectional nature of their work and its impact, examining problems from multiple perspectives and bringing real-world problems to industry to develop innovative solutions,” says Dan Huttenlocher, co-chair of the MIT laboratory, dean of the MIT Schwarzman College of Computing, and Henry Ellis Warren (1894) professor of electrical engineering and computer science.
An important factor in ensuring this research ecosystem thrives is the steady flow of student talent and contributions through MIT's Undergraduate Research Opportunities Program (UROP), MIT EECS 6A programand the new MIT-IBM Watson AI Lab internship program. In total, over 70 young researchers not only accelerated the development of their technical skills, but with the guidance and support of mentors, the lab gained knowledge in artificial intelligence fields to become emerging practitioners themselves. That's why the lab continually strives to identify promising students at all stages of their discovery of the potential of artificial intelligence.
“To unlock the full economic and social potential of AI, we must support 'useful and efficient intelligence,'” says Sriram Raghavan, IBM vice president of AI research and chair of the IBM Lab. “To translate the promise of AI into progress, it is critical that we continue to focus on innovation to develop efficient, optimized and tailored models that can be easily adapted to specific domains and use cases. Academia-industry collaborations like the MIT-IBM Watson AI Lab are helping drive the transformative breakthroughs that make this possible.”
