Erik Avakian, technical counselor at Information-Tech Analysis Group, stated that he applauds NIST’s efforts to succeed in out for group suggestions, however he additionally cautioned that it would backfire. For instance, what if AI brokers flood the feedback with self-serving options?
Such an AI remark flood might do numerous unhealthy issues, he stated, together with making the ultimate suggestions “unhealthy man pleasant” or just “AI poisoning the precise suggestions.” If that assault occurred, Avakian stated, the perfect response from NIST can be to conduct in-person interviews. “That may be the one method. Possibly human interviews or regional workshops the place they bring about individuals in,” he stated.
Though Avakian stated that the preliminary NIST report is “actually a welcome begin, there are potential dangers that the overlays might not go far sufficient as they relate to rising assault vectors distinctive to AI. The [report] addresses fundamentals reminiscent of mannequin integrity and entry management, however these alone won’t dig deep sufficient into addressing cutting-edge assault vectors distinctive to AI,” Avakian stated.
