A review explores the role of AI in engineering, assessing the benefits and challenges of the synergy between the two fields. A 2004 DARPA contest pitted AI vehicles against one another in a race on 150 miles of dirt roads. The best-performing vehicle made it less than eight miles of the way. The next year, five vehicles finished a 132-mile course, and today driverless cabs are active in several major cities. Enthusiasts have suggested AI could improve transportation and manufacturing, medicine, consumer goods, and military technology. Rama Chellappa, Guru Madhavan, Ed Schlesinger, and John Anderson assess those claims by exploring several case studies, including autonomous cars and planes, AI-assisted surgery, AI-guided closed loop anesthesiology, AI and robotics, and AI-assisted discovery of new materials. Among the questions the authors raise: should AI vehicles be fully autonomous if more accidents occur when humans are in control? Who will get left behind as AI remakes the world economy? The authors discuss weaknesses in data-driven models, such as domain shift, lack of robustness to adversarial attacks, bias in decision-making, and lack of explainability—along with issues of safety, privacy, bias, and other ethical and governance issues. As we enter an era of machines that can sense, learn, and make decisions without human supervision, the authors call for research into appropriate regulatory frameworks to address questions of safety, liability, and governance of engineered AI systems.
PNAS Nexus
Engineering and AI: Advancing the synergy
11-Mar-2025
The authors declare no competing interests. The views presented in this article are those of the authors and not necessarily of the National Academies of Sciences, Engineering, and Medicine. This paper results from discussions at a forum on engineered AI systems organized by the National Academy of Engineering and the Johns Hopkins Whiting School of Engineering in June 2024.