Google, Privacy Reps Debate AI Transparency Standards

Consumer protection for artificial intelligence systems is a lot harder for the FTC without clear visibility into system decision-making, said Electronic Frontier Foundation Tech Projects Director Jeremy Gillula Tuesday during the agency’s seventh policy hearing. Some companies have made an effort, but it’s an ongoing problem, he said. Consumers and researchers might not necessarily need every detail about machine learning and artificial intelligence decisions, said Google Brain Team Senior Staff Research Scientist Martin Wattenberg. Google isn't giving “the full matrix of every weight in the neural network, but we’re giving them information that’s useful at the level that they want in terms of a concept that they’re actually interested in.” Wattenberg emphasized progress made in coming up with ways to understand these systems: “They no longer need to be considered black boxes.” Google recommended practices for fair artificial intelligence use, which covers interpretability, privacy and security. Computer & Communications Industry Association Competition and Regulatory Policy Director Marianela Lopez-Galdos questioned whether laws that focus on consumer welfare are sufficient to address machine learning issues.