At This Point, Zoom Could Use Another Pandemic

For a while, Zoom was the most important company in America. Take a look at some recent headlines: • Over the weekend, full-stack developer Alex Ivanovs noted on his Stack Diary platform that Zoom had updated its terms of service, with language assuming automatic user consent for the company’s utilization of “Customer Content” and “Service Generated Data” for “machine learning, artificial intelligence, training, testing.” After the discovery was shared on Hacker News, Zoom’s chief product officer wrote responses both on the forum and on the company blog, declaring on Monday that “we will not use audio, video, or chat customer content to train our artificial intelligence models without your consent.” Ivanovs found, however, that the app still doesn’t allow any opt-out option for Service Generated Data collection; Gizmodo further noticed that there’s no easy means to refuse sharing your customer content with the company’s optional, A.I.-powered, chatbot-heavy Zoom IQ. Naturally, as the New York Times reported Monday, Zoom employees were not happy with the new mandate, “express frustration about the time and money they’d waste while commuting.” One anonymous employee message reported by the San Francisco Standard: “The irony is you need to go to office but every meeting is via Zoom.” • British researchers put out a new paper on Monday noting that Zoom’s audio-suppression tools aren’t strong enough to prevent deep-learning models from being able to detect the text that you type on your Mac’s keyboard, even if that text isn’t transmitted through Zoom—because the models are able to detect which keys you’re pressing just based on their relative sounds. Although the notorious tech has been part of Zoom’s arsenal for years—the company introduced A.I.-powered transcription programs back in 2017—its work in this field has only recently come under more scrutiny and skepticism. Last year, Zoom declared that it would consider incorporating “emotion A.I.,” software with a dubious ability to gather and use data from users’ facial expressions, tone of voice, and conversational language.