Microsoft: Better transcription AI will mean fewer meetings

Collaboration applications will eventually have more accurate tools for transcribing meetings. And as the quality of transcriptions rises, the need to attend all meetings in person will fall, a Microsoft executive said.

Ilya Bukshteyn, general manager of Microsoft Teams devices, said AI-powered transcribing of conversations within collaboration software will eventually let people skip meetings and catch up on discussions later. Transcriptions will include more accurate search technology to locate specific information.

"AI will enable us to have fewer meetings and get rid of the FOMO [fear of missing out]," Bukshteyn said. "I can actually have a better meeting after the fact by quickly watching the recording and having the key pieces pointed out to me."

This week, Bukshteyn joined Cisco and Five9 executives on a panel discussion of AI at the Enterprise Connect virtual conference. How AI will improve collaboration applications is important because companies will depend on the software to hold post-pandemic meetings between employees working inside and outside the office.

Third-party transcription software for collaboration applications has been around for a few years from vendors like Fireflies.ai and Otter.ai. Recently, Cisco and Microsoft started offering live transcription services within Webex Teams and Teams, respectively.

Unfortunately, the service quality limits their use as a meeting replacement.

"AI systems have inaccuracies," said panelist Jonathan Rosenberg, CTO at cloud contact center vendor Five9. "They're all predicting things and guessing things, and accuracy isn't 100% for any applications."

Collaboration vendors use AI for more than transcribing meetings. It is also used for outside noise cancellation during sessions and to create virtual backgrounds behind participants.

But the reliability of those features is not as critical as with transcripts. The latter loses value dramatically if the transcribing technology misidentifies people or produces a transcript with many errors.

Jon Arnold, an analyst at J Arnold & Associates, said that although vendors may claim that their transcription services are roughly 95% accurate, a meeting video recording "will debunk that pretty quickly."

"It's not nearly as good … when you start using a lot of acronyms and complex language," Arnold said.

Nevertheless, there's hope, he said. The AI technology learns from mistakes corrected by the user and improves in time.

"It never makes the same mistake twice," Arnold said. "What starts at 80% accurate will inch itself up to 100%."