Less than one month after enduring public backlash over confusion about how it trains its artificial intelligence tools, Zoom says it’s prepared to plow forward with a ChatGPT-style chatbot to embed in its videoconferences and accompanying chats. Zoom believes its beefed-up AI Companion AI tool could boost productivity for its users but it also risks reigniting concerns over the type of data that AI is trained on.
Zoom’s AI Companion, according to a company blog post, is actually a rebrand and expansion of ZoomIQ, which previously let users summarize chat threads and generated automated responses to written chat questions. Next year, Zoom says users will be able to chat with the AI Companion and ask it for help on a variety of office tasks, from quickly getting prepped for a meeting and receiving consolidated summaries of past sessions, to searching for specific documents. Zoom says users will have the ability to interact with the AI Companion in the middle of a meeting to file support tickets or research answers to questions raised in real-time. In essence, Zoom wants its companion to serve the role of a true, private virtual assistant.
These conversational features won’t ship until Spring, 2024, according to a company blog post.
“With AI Companion empowering you in Zoom, you can save time, improve the quality of your work, stay more connected with teammates no matter where or when they work, and receive insightful coaching that will help you level up skills like delivering a great presentation,” Zoom Chief Product Officer Smita Hashim said.
Zoom says its approach to AI relies on both its own proprietary large language model in combination with Meta’s Llama 2, OpenAI, and Anthropic. This, “federated approach to AI,” Zoom says, should let the AI Companion quickly incorporate new advancements from multiple different models. Zoom did not immediately respond to Gizmodo’s request for comment for more details.
The conversational chatbot tools are just one of several AI features Zoom is rolling out to try and boost user productivity on the platform in the coming months. The company says its AI Companion, which is already available to paid Zoom subscribers, will let users catch up on highlights from a meeting and ask specific questions about what’s been discussed in a meeting. Once the meeting wraps up, Zoom claims the Companion will automatically transcribe the session and segment it into “smart chapters,” and highlight important information. The companion will also create summaries of meetings that highlight important topics and potential next steps.
By the end of September, Zoom says it plans to release a feature where its AI Companion can help generate email responses with the AI judging the proper tone and length. Around the same time, Zoom says the Companion will be able to take a variety of missed chats and summarize them “so you can see the big picture more easily.” Later this Fall, Zoom claims AI Companion will be able to automatically detect meeting intent through chat messages. Looking forward, Zoom believes its users will be able to interact with the AI Companion to ask for feedback on how to improve presentation skills and track how long they were talking versus listening during a meeting.
Zoom is trying to jump back on the AI saddle this week after wading into a clunky controversy over unclear language regarding the types of data it uses to train AI models. Last month, the company was forced to backtrack and update a blog post clarifying that it would not use Customer Content, such as call audio, video, chat, attachments, or screen-sharing, to train its AI despite previously seeming to suggest it could. Zoom’s most recent blog post similarly claims it does not use any audio, video, chat, screen sharing, or other “communications like customer content” to train Zoom’s AI or other third-party model. Naturally, one might wonder then exactly what data is actually used to train the system. Zoom did not respond to our request for comment asking for clarity here.