The University of Michigan has joined forces with artificial intelligence development company OpenAI to establish new research funding opportunities and AI resources. This collaboration aims to advance the University’s AI research initiatives. The partnership is part of NextGenAI, OpenAI’s $50 million effort, designed to support AI projects at various research institutions.
According to Engineering professor Michael Wellman, the partnership, one of 15 similar collaborations OpenAI has formed, has been in the works for several months. Wellman provided insights into the process, stating that the University would soon issue a call for proposals from faculty for potential projects seeking funding. “We are right now preparing a call for proposals, which we will be publishing to the University community and Ann Arbor campus,” Wellman said. “Unfortunately, we are only going to be able to support a small number of these in the first round, but we are looking to showcase really the breadth of really interesting ideas that people all around campus have related to AI.”
Selected projects will provide researchers with access to crucial resources via an Application Programming Interface. This includes funding for computational resources to facilitate intensive research endeavors. “They’ll get some resources, of course, which will include some funding that can be spent on the research,” Wellman explained. “And especially important, is access to OpenAI’s models and tools. They’ll also be able to get certain amount of credits to use for API access to the latest open AI models, and this is something that they can also have access to through paid services, but this may be able to let them really engage with the models at higher scale and intensity than they could in their current research.”
Richard Lewis, a professor of psychology, linguistics, and cognitive science, weighed in on AI’s capabilities. In an email to The Daily, Lewis wrote that “It is astonishing how quickly many of the benchmarks used to assess AI capabilities have been rendered obsolete. But one thing has remained true throughout all of these advances: We do not yet have systems that robustly achieve human-level cognitive abilities from training on human-scale data. Some think that this is revealing of a fundamental problem with current approaches, but I think it is way too early to know.”
Nandini Valluru, an information student, noted that AI, despite its advancements, struggles with replicating human qualities, like emotion and personal voice. “I think AI struggles with replicating human emotion and spirit,” Valluru wrote. “An AI-generated piece of writing always lacks the emotion and voice of a human-written piece. You know when you read something that is so human or watch something that feels very human? Art is human, and I don’t think AI could do it as well as humans.”
Wellman also acknowledged the risks and benefits associated with AI advancements, emphasizing the University’s responsibility to monitor the technology’s development conscientiously. “Artificial intelligence is really leading to very transformative effects in our society and the kinds of problems we can solve and how we work, how we learn, how we really do everything,” Wellman said. “New kinds of biases could appear or could be magnified by AI. There’s also lots of potential ways that AI can help democratize access to knowledge and lead to tremendous benefits.”
Wellman highlighted the crucial role of academic institutions in maintaining ethical and safe AI development, particularly at a public university like the University of Michigan. “As academics, we represent the public sector, especially at a public university like the University of Michigan, and so we have a mandate to help society navigate this AI transformation in ways that are beneficial to human beings and to society at large,” Wellman said.
Valluru shared her thoughts on ethical concerns, emphasizing the significance of diverse perspectives and inclusion of minority groups in AI development. “We need legislation and regulations surrounding AI and where it is used,” Valluru wrote. “Evaluate training data and models routinely. Involve diverse perspectives. If models are built with only one perspective, then those perspectives are championed in the technology, leaving behind other groups of people.”
Wellman expressed optimism about the collaboration’s potential for a long-term partnership between the University and OpenAI. “These are going to be modest size projects, but they’ll be relatively short term, at least for just a year,” Wellman said. “We do hope that this collaboration will be successful, and we may be able to have an ongoing relationship and follow on projects, but it really will help these investigators pursue their cutting edge ideas.”
