Strategic Areas: MIT-Google Program for Computing Innovation
Computing for Science and The Planet
The health of the planet is already a strategic area for MIT and an active area of research within the College. Examples of relevant projects in this area include how to leverage artificial intelligence (AI) for environmental monitoring, forecasting, and climate change mitigation, and how the intersection of computing with other disciplines in science and engineering can spawn new technologies with positive environmental impact, such as the creation of denovo biodegradable materials. In addition, computation and AI catalyze and support research in other areas of science, and synergistic combinations of computation with scientific research is solicited across many fields, including biology, chemistry, and physics.
With a dedicated pillar for ethical and responsible AI and computing, the College infuses good practices of computing into all MIT sciences, humanities, business, and engineering disciplines. A deeper and broader understanding of the impact of ML deployment in domains such as fairness, bias, transparency, equity, policy, responsibility, and accountability are of utmost importance for democratizing and developing safer AI technologies. Ethical and responsible computing goes beyond AI, and includes questions of privacy, security and dependability as well as questions of accessibility.
We are moving into a world where computing is no longer just a tool for automating mundane tasks, but a partner in a variety of creative endeavors, from scientific discovery to art and engineering. But a number of opportunities remain before we can fully realize that potential. For example, in the context of scientific discovery, how can we move from the current crop of systems, which can make remarkably accurate predictions after learning on massive amounts of data to systems that can formulate new hypotheses, suggest experiments, and even pose new questions worthy of exploration? How do we foster productive partnerships between machines and human creators? How can machines perceive images and sounds the way people do, and can we leverage this to produce images and sounds that will elicit specific human reactions? At what point does a machine generated artifact become art? What new technical capabilities are required before we can trust machines as full creative partners?
A core topic of computing is efficiency. Computational efficiency is key both in enabling new applications of computing as well as in allowing existing applications to execute with fewer resources or in more constrained devices. Supporting more efficient computing requires advances at every level of the computing stack, from novel computer architectures, to more efficient systems software, to improved programming technology, to novel algorithms for specific applications. Advances that span multiple of these levels can have especially big performance impacts, as illustrated by the success of domain specific compilers capable of exploring algorithmic choices, or the success of specialized hardware for deep learning. Traditionally, there has been a tradeoff between performance on one hand and developer effort and expertise required on the other. Techniques that achieve performance without the need of extensive human intervention and without the risk of introducing bugs can be especially desirable.