This week, a group of tech industry leaders released an open letter warning of the imminent threat posed by artificial intelligence, likening it to the risks of a “pandemic and nuclear war.”
This open letter is just one of many recent attempts to draw attention to situations in which AI is untrustworthy and to question its potentially unfair or harmful effects.
A group of researchers from the University of Notre Dame say it’s important to ask a slightly different question. So what would it be like to develop trustworthy artificial intelligence?
They collaborated with technical experts within the U.S. military and researchers at Indiana University-Purdue University Indianapolis (IUPUI) and Indiana University to develop a comprehensive and systematic approach to creating trustworthy AI. doing.
Their project, called Trusted AI, identified six widely shared values, called the “Dimensions of Trusted AI.” The six dimensions are:
- explainability — Can you explain how AI arrives at inference?
- Safety and Robustness — Will AI perform as expected in a real live environment as well as in the lab?
- fairness — Can you guarantee that AI will not reproduce patterns of prejudice or discrimination?
- privacy — Are you confident that the data your AI uses will be kept safe and confidential?
- environmental health — Can AI be trained and developed with minimal environmental impact?
- Accountability and audit responsibility — Can you identify the person responsible? And can you confirm that the AI is working as expected?
According to researchers, a major challenge in developing trustworthy AI is ensuring that each aspect of the process is robust at every stage of the process, from initial data collection to the output the AI provides, or “inference.” To be able to transmit information. Only when there is an unbroken “chain of trust” can you be sure that the end result is trustworthy.
The principal investigator of the Trusted AI project is Christopher Sweet, Associate Director of Cyber Infrastructure Development at the Center for Research Computing.
Sweet, who is also an assistant professor in the Department of Computer Science and Engineering, emphasizes that the development process for Trusted AI is a cycle, not a one-time effort.
“It’s an iterative process,” Sweet explains. “These technologies are constantly evolving, as are the datasets they rely on and the social contexts in which they are used. It’s not about declaring victory. to demonstrate that it is an ongoing practice that requires the participation and engagement of
Charles Vardeman, a computational scientist at the Research Computing Center (CRC) and an assistant professor in the School of Computer Science and Engineering, leads the Trusted AI subproject. Bardeman said the team is working far beyond high-profile technologies and applications to prevent AI harm.
“People realize AI is powering things like Alexa and ChatGPT, but that’s really just the tip of the iceberg,” he says. “Most people interact with AI on a regular basis without even knowing it.
Adam Zaika, assistant professor in the School of Computer Science and Engineering, is leading the Trusted AI subproject, which focuses on how humans and machines work together to make trusted decisions. He and his colleagues developed a method to train AI to recognize fake images by training it to mimic human perception.
Another Trusted AI subproject, led by CRC Senior Associate Director and Professor of Practice Paul Brenner, is applying the Trusted AI recipe to create technology for the U.S. Navy.
Brenner, a faculty member at iNDustry Labs, ND Energy, and the Wireless Institute, explains: New machine learning tools such as natural language processing and knowledge graphs can help mine data and identify root causes of failures. ”
The stumbling block, Brenner said, is that most commercial machine learning tools are “black boxes.” They make inferences based on large datasets. What they don’t provide is an explanation of how or why they arrived at a particular reasoning.
Brenner’s team is developing new approaches beyond the “black box” for military applications. Working with the U.S. Naval Station near Crane, Indiana, Brenner and a group of 10 undergraduate researchers from the University of Notre Dame were pre-labeled for more accurate and explainable results. You’re building a machine learning tool trained on a specialized set of data.
Brenner emphasizes that in addition to the new tools and technologies his project develops, it also includes far-reaching implications that will continue to have an impact for decades to come.
“We are looking forward to sharing what we have learned with a wider group of students,” Brenner said. In addition to training students directly involved in research, the team educated young students on Trusted AI principles through presentations and by welcoming her 40 high school students to the University of Notre Dame campus for CRC’s Summer Scholars Program. educate the
“We are developing new approaches to AI that are urgently needed,” says Brenner. “At the same time, we are cultivating future military personnel, academics, and technology industry leaders who will make trustworthy AI a reality.”
Trusted AI is part of the Scalable Ametric Lifecycle Engagement (SCALE) workforce development program funded by the Undersecretary of Defense Research and Engineering Office’s Trusted & Assured Microelectronics Program.
was first issued and crc.nd.edu upon June 2.
