Artificial Intelligence, and the concerns it raises, are increasingly at the center of our public debates about the technology.
The challenge is that we still know relatively little about the factors within AI that create the problem. At times, AI acts in ways that make it appear to be intentionally trying to deceive people. Another challenge is assessing the threats that AI poses. While “loss of control,” where AI goes rogue, is drawing attention at the moment, more practical safety concerns are emerging.
These issues were the subject of two articles released on Tuesday by Northeastern University.
In “Can scientists build a lie detector for AI chatbots,” the author covers Northeastern researcher David Bau who is trying to understand why chatbots seem at times to actively deceive users. The real challenge in understanding this problem is that “researchers still don’t really understand how to get to the bottom of a chatbot’s miscalculation or potential deception, because they don’t have a strong grasp on how AI models produce their answers, the researchers say.”
To get beyond that problem, Bau is putting chatbots through “theoretical MRI machines to understand the synaptic connections” that define AI neural networks. In the same way that MRI’s can track neural breakdowns in human brains, Bau is hoping to look inside the AI “black box” and identify synaptic patterns that signal the machine is intentionally being deceptive.
If he is successful, then Bau hopes to develop a “white box” that can serve as a “lie detector that can be used to help uncover when an AI is convincingly being deceptive.”
In “After weeks of AI doomsday warnings, experts want to shift the debate to practical safety,” the article covers the concerns over AI doomsday scenarios before turning to more practical worries. “I think the bigger threat is bad actors leveraging AI for cybersecurity attacks, privacy attacks, hacking society, hacking humans,” Usama Fayyad of Northeastern University said. “The threat is not the AI itself. The threat is what people can do with it.”
