Some see evidence that the swift progress of AI is pushing oversight Have you ever had to rethink, redesign or override an entire further up the agenda for many organizations. “AI is becoming a AI-based system due to questionable or unsatisfactory results? catalyst for stronger governance practices in many companies,” says Kimberly Nevala, Director of Business Strategies for SAS. “If AI is directly influencing your customers or automating critical operational decisions, you can’t take it for granted that the data will be just right, the models will just work and resultant outcomes will meet expectations. As organizations advance down the AI path and see these very real impacts – both positive and negative – they are 19% 24% moving quickly to put that next level of oversight in place.” UNSURE YES Consider the survey responses to the question of “who conducts reviews or evaluations of AI outputs?” C-level executives such as the chief digital officer, chief data officer or chief analytics officer top the list of answers. At first glance, this may seem encouraging – after all, executive-level oversight would seem to be a positive reflection of the seriousness with which AI oversight is viewed at these organizations. But should these high-level executives be the ones to review AI systems outputs? “These executives have big responsibilities,” says Accenture’s Chowdhury. “We need to make sure they are providing oversight of AI at the systems level, and that their oversight is matched by equally rigorous oversight from managers and others at more tactical levels – multiple layers of oversight.” For Intel’s Greer, the response to a separate question sheds light on the need for a more sophisticated approach to AI oversight. “The survey asked if respondents had to rethink, redesign or override an entire AI-based system due to questionable or unsatisfactory results,” says Greer. “Fifty-seven percent of respondents indicated 57% that they hadn’t, which would imply that AI systems have been more successfully developed and deployed than any other systems, NO technology or otherwise, that have been deployed. Like any other systems, AI capabilities aren’t perfect, and particularly at this stage in our evolution they will require much trial and error. I think that if oversight processes were working at a high level, we wouldn’t see so many of these executives saying everything is working perfectly.” 15