Zoom ensures that user content is not exploited to train its AI — Siècle Digital

After embarrassing itself with disgruntled users, Zoom decided to edit a blog post about its terms of service. Subject of contention: A misunderstanding of the data used to train the platform’s generative artificial intelligence (AI) systems.

The vagueness around the consent »

Zoom recently updated its terms of service, specifying that it will not use its customers’ audio, video or chat content to train its AI models without their consent. It was precisely this last point that made many shiver, as the study did not specify the methods of obtaining consent. Some feared, for example, that the appearance of a simple pop-up window during a call could be considered consent to the use of their data.

Many interpreted the move as a way to give the company broad control and copyrights over customer data, potentially including anything they showed or discussed on a call. Zoom leverages AI models to power features like meeting summaries.

On Aug. 11, the company changed its blog post to reassure users. ” Zoom does not use audio, video, chat, screen sharing, attachments, or other communications such as customer content (such as survey results, whiteboards, and reactions) to train Zoom or third-party AI models “, he wanted to clarify.

Nurse training raises questions

With the boom in generative AI kicked off by the release of ChatGPT last November, more and more companies are leveraging the technology to enhance their products, such as Zoom.

However, these systems are built on large amounts of data from the Internet and the provenance of this data is increasingly questionable. In addition to fears about privacy violations, many also fear possible copyright infringement.

For these various reasons, the European Union wants to force large companies to specify where the information used to train their models comes from. This is one of the measures contained in the AI ​​Act, legislation that will regulate the use of AI in the Old Continent.