When it comes to artificial intelligence, MIT and IBM have been there from the beginning, building the foundational work, creating some of the first programs (precursors to AI), and theorizing how machine “intelligence” might be achieved.
Today, collaborations like the MIT and IBM Watson AI Lab, launched eight years ago, continue to provide the expertise that will enable future AI technologies. This is critical, especially for the industries and workforce that will benefit in the short term. From a projected global economic impact of $3-4 trillion over the next three years and an 80% productivity increase for knowledge workers and creative tasks, to the significant incorporation of generative AI into business processes (80%) and software applications (70%).
While industry has seen a notable model boom primarily over the past year, academia continues to drive innovation and contributes most of the highly cited research. The MIT-IBM Watson AI Lab has achieved success in the form of 54 patent disclosures, over 128,000 citations with an h-index of 162, and over 50 industry-driven use cases. The lab’s many accomplishments include improving stent placement with AI imaging techniques, significantly reducing computational overhead, scaling down models while maintaining performance, and modeling interatomic potentials in silicate chemistry.
“This lab is uniquely positioned to identify the ‘right’ problems to solve, which sets us apart from other institutions,” said Ord Oliva, MIT lab director and director of strategic industry engagement in the MIT Schwarzman College of Computing. “Furthermore, the experience our students gain from tackling these enterprise AI challenges will foster a competitive edge in the job market and a competitive industry.”
“The MIT-IBM Watson AI Lab has made a tremendous impact by bringing together the rich collaboration between IBM and MIT researchers and students,” said Professor Anantha Chandrakasan, MIT co-chair of the lab and the Vannevar Bush Professor of Electrical Engineering and Computer Science. “By supporting cross-cutting research at the intersection of AI and many other fields, the Institute advances fundamental research and accelerates the development of innovative solutions for our country and the world.”
long term job
As interest in AI continues to grow, many organizations struggle to translate the technology into meaningful outcomes. A 2024 Gartner study found that “at least 30% of generative AI projects will be abandoned after proof of concept by the end of 2025,” demonstrating ambition and widespread aspirations for AI, but a lack of knowledge about how to develop and apply it to create immediate value.
This is where laboratories play an important role in bridging research and development. A large portion of the Lab’s research portfolio this year is geared toward the use and development of new capabilities, capabilities, or products for IBM, the Lab’s corporate members, or real-world applications. The final model consists of foundational models that include large-scale language models, AI hardware, and multimodal, biomedical, and geospatial models. Inquiry-minded students and interns are invaluable in this pursuit, providing enthusiasm and new perspectives while accumulating expertise that helps unlock and design advances in the field, and using AI as a tool to open new frontiers for exploration.
The findings of the AAAI 2025 President’s Panel on the Future of AI Research support the need for collaborative contributions from academia and industry, such as laboratories in the field of AI. “Academics have a role to play in providing independent advice and interpretation of the results[from industry]and their results. The private sector is more focused on the short term, and universities and societies are more focused on the longer term.”
Combining these strengths, along with promoting open source and open science, can spark innovation that cannot be achieved by either alone. History shows that adopting these principles, sharing code and making research accessible has long-term benefits for both industry and society. In line with the mission of IBM and MIT, through this collaboration, the Institute brings technology, discovery, governance, and standards to the public sphere, thereby increasing transparency, accelerating reproducibility, and ensuring trusted progress.
The lab was created to combine MIT’s deep research expertise with IBM’s industrial R&D capabilities, aiming to develop core AI methods and hardware breakthroughs as well as new applications in areas such as healthcare, chemistry, finance, cybersecurity, and robust planning and decision-making for business.
Bigger is not necessarily better
Large foundational models are now being replaced by smaller, more task-specific models that deliver better performance. Contributions from lab members such as Song Han, associate professor in MIT’s School of Electrical Engineering and Computer Science (EECS), and Chuang Gan of IBM Research make this possible through research such as Once for All and AWQ. These innovations improve efficiency through better architectures, algorithmic reduction, and activation-aware weight quantization, allowing models such as language processing to run faster and with lower latency on edge devices.
The results show benefits for foundational, visionary, multimodal, and large-scale language models, allowing the research group of Oliva, MIT EECS Associate Professor Yoon Kim, and IBM research members Rameswar Panda, Yang Zhang, and Rogerio Feris to build on that work. This includes techniques for injecting external knowledge into the model and the development of linear attention transformation methods that increase throughput compared to other state-of-the-art systems.
Vision and multimodal systems understanding and reasoning have also benefited. Works such as “Task2Sim” and “AdaFuse” demonstrate improved performance of vision models when pre-trained on synthetic data and how video action recognition can be improved by fusing channels of past and present feature maps.
As part of the Lean AI effort, a team in the lab of Gregory Wornell, professor of engineering at MIT EECS Sumitomo Electric Industries, Chuang Gan of IBM Research, and David Cox, vice president of foundational AI at IBM Research and IBM director at the institute, showed that model adaptability and data efficiency can go hand in hand. Two approaches, EvoScale and Chain-of-Action-Thoughting (COAT), allow language models to make the most of limited data and computation by improving previous generations of trials and narrowing down better responses through structured iteration. COAT uses a meta-action framework and reinforcement learning to tackle reasoning-intensive tasks through self-correction. EvoScale, on the other hand, brings a similar philosophy to code generation and evolves high-quality candidate solutions. These techniques help enable resource-aware and targeted real-world deployment.
“It is difficult to overstate the impact that MIT and IBM research has had on our large-scale language model development efforts,” Cox said. “We’re seeing smaller, more specialized models and tools having a huge impact, especially when combined. Innovations from the MIT-IBM Watson AI Lab will help shape these technological directions and influence the strategies we’re taking to market through platforms like WatsonX.”
For example, numerous lab projects have contributed to the capabilities, capabilities, and uses of IBM’s Granite Vision. Despite its compact size, Granite Vision offers impressive computer vision designed to help you understand your documents. This comes at a time when there is an increasing need to extract, interpret, and authoritatively summarize information and data contained in long format for enterprise purposes.
Beyond direct research on AI, other cross-disciplinary outcomes are not only beneficial, but necessary to advance technology and improve society, the 2025 AAAI Panel concludes.
The research of Caroline Uhler and Devabrat Shah in the lab (both professors Andrew (’56) and Erna Viterbi of EECS and the Institute for Data, Systems and Society (IDSS)) and Christian Greenwald of IBM Research crosses disciplines. They uncover how interventions affect outcomes and develop causal discovery methods to identify which interventions achieve desired outcomes. This research includes the development of a framework that can elucidate how “cures” are deployed for different subpopulations, such as e-commerce platforms and movement restrictions related to morbidity. The findings from this body of research may have implications for the fields of marketing, medicine, education, and risk management.
“Advances in AI and other fields of computing are impacting the way people formulate and approach challenges in nearly every field. At the MIT-IBM Watson AI Lab, researchers recognize the cross-cutting nature of their work and its impact, examining problems from multiple perspectives and bringing in real-world problems from industry to develop new solutions,” said Dan Huttenlocher, MIT Lab Co-Chair and MIT Dean. Schwarzman College of Computing, Henry Ellis Warren (1894) Professor of Electrical Engineering and Computer Science.
A key factor in making this research ecosystem thrive is the steady influx of talented students and their contributions through MIT’s Undergraduate Research Opportunities Program (UROP), the MIT EECS 6A program, and the new MIT-IBM Watson AI Lab internship program. In total, more than 70 young researchers not only accelerated the development of their technical skills, but also gained knowledge in the AI domain and became new practitioners themselves through the guidance and support of their lab mentors. This is why the lab continually strives to identify promising students at every stage of exploring the potential of AI.
“To realize the full economic and social potential of AI, we need to foster ‘useful and efficient intelligence,’” said Sriram Raghavan, vice president of AI research at IBM and director of IBM Research. “To turn the promise of AI into progress, it is important to continue to focus on innovation to develop efficient, optimized, and fit-for-purpose models that can be easily adapted to specific domains and use cases. Collaborations between academia and industry, like the Watson AI Lab at MIT and IBM, will help drive the breakthroughs that make this possible.”
