By David Cox, Director, MIT-IBM Watson AI Lab
Industry and academia have collaborated in artificial intelligence research for decades, but in recent years the power balance in this relationship has shifted in ways that are detrimental to AI progress and the sustainability of the field.
Most existing arrangements between industry and academia are either “work for hire,” which often is too narrowly defined to attract the brightest minds in academia to participate, or “buy the lab,” which effectively end collaborations by hiring researchers away from academia and prevent the next generation of AI talent from receiving the education and research opportunities that will lead to AI progress in the future, cannibalizing the future pipeline to serve the needs of the present.
A new working model between industry and academia is needed, one in which stable, long-term industry-academic partnerships enable continued AI advancement while preserving our society’s capacity to conduct fundamental research and train future generations of AI experts.
In a long-term partnership, academic and industry researchers must work collaboratively as equals, rather than industry merely sponsoring research or pulling faculty or students out of academia.
Instead of traditional top-down or single-organization decision-making, successful partnerships should be guided by more inclusive decision-making approaches – for example, through joint committees, with equal representation of academic and industry members, each of whom feels a strong responsibility to the collaboration and to the advancement of AI.
We believe our MIT-IBM Watson AI Lab collaboration offers a new model for engaging between academia and industry. Below are five key advantages to such a model, and an explanation of why it’s the surest path to transformational progress in AI research.
AI is exploding with new and expanding subfields, and conducting rapid and meaningful AI research demands cross-disciplinary knowledge, along with intense focus. Strong long-term partnerships between academia and industry are positioned to integrate a broad range of academic disciplines — from computer science, mathematics and logic to biology, linguistics, economics and even the arts — with industry’s real-world perspective, domain knowledge, and access to data. Furthermore, advances in AI demand new ideas and a creative, ambitious workforce, along with substantial computational and financial resources. With academia being a fertile source for the former and industry uniquely positioned to provide the latter, unifying the two takes full advantage of their complementary strengths.
Because expansion of AI has broad implications for all people and communities, its creation and development should reflect a diversity of backgrounds and viewpoints. Part of the value of a peer research approach is in the variety of perspectives, expertise, and experience levels it offers. Students bring fresh ideas and eagerness to immerse and learn quickly, to experiment and take risks, to deeply focus on a novel problem or solution, and to earn a scientific reputation (and a degree) for themselves. Experienced academic and industry researchers share deep expertise in their chosen areas that comes from years, potentially decades, of focus, failures, and breakthroughs; scientific rigor and principled approaches; and an understanding of the broader context in which technology can be brought into service.
One of the greatest accelerators in AI progress is the openness with which academic and industry players have shared the fruits of their research. Yet it is not uncommon that when talented AI researchers leave academia and join industry, their research becomes more closed and less accessible to the field, slowing the overall development of AI as a field. We recognize that there is substantial value in an open ecosystem in which industry and academia work in close collaboration with one another, sharing their results and technologies with the wider AI community. By publishing in top scientific conferences and journals, and open-sourcing data and code, we can feed the research ecosystem and accelerate rather than stifle the development of AI.
Read the source article in Information Week.