​At the 2018 annual conference of the American Society for Health Care Risk Management (October 7-10 in Nashville), the keynote panel "Artificial Intelligence: Challenges and Opportunities for Risk Management" was presented by Anu Banerjee, PhD, MS, MHM, of Arnot Health in New York state; Erin Grace, MHA, and Rear Admiral Jeffrey Brady, MD, MPH, both of the Agency for Healthcare Research and Quality's (AHRQ) Center for Quality Improvement and Patient Safety; and Elliot Fishman, MD, of Johns Hopkins Hospital in Baltimore, Maryland. The panel explored the risks, opportunities, challenges, and benefits of artificial (or augmented) intelligence (AI) for healthcare organizations. Nothing can replace the human elements of healthcare, the panelists emphasized. Rather, AI can serve as an enhancement to the care clinicians provide. AI can handle significant amounts of data; however, Grace noted a caveat: the data used to train the system must be real and significant. Agreed Banerjee: "Data is only valuable when it's converted to information. We are using AI to make the data smarter." Likewise, Banerjee said, predictive models, such as heat maps, can be used in preparation for AI use, as the goal of translation from data into actionable information is the same. "Make sure that it's used for process improvement, not just data trending," Banerjee advised. AI can be used to reduce errors and adverse events, to support diagnostic accuracy, to support personalized disease treatment, to optimize electronic health records, and to support quality improvement initiatives, Banerjee suggested. Fishman, a radiologist, described reading a computed tomography scan: "AI may be the best second reader we'll ever have. It can identify abnormal findings for us to review." He described the process of teaching a computer program to identify pancreatic tumors: "The program's success rate is about 90%." An additional benefit of AI analysis, said Fishman, "is that it all happens in the background—it's not taking hours or days." Fishman also emphasized the benefits for the patient: "It's helping us treat better" because the diagnosis becomes more specific. However, Fishman posed the question of responsibility for the diagnosis. For example, while reading an image, if the AI flags something that the clinician dismisses but that is actually a positive finding, who is responsible for the error? "Those will be interesting challenges," agreed the panelists. Brady also reminded attendees that the added value of AI should be assessed as it evolves, and cautioned that, as with any new supportive tool, the benefits and limits of AI need to be clearly identified, understood, and managed. However, Fishman predicted that how clinicians practice will completely change in a few decades, and he speculated that success may well depend on how clinicians adapt to incorporating of AI into care delivery processes. Grace emphasized that with deliberate, successful implementation, AI has the potential to support clinicians in their work, providing data that allow the clinician to "provide the best care for each and every patient we see."