PolyAI CTO Shawn Wen says models that listen and speak simultaneously have been developed, but finding answers fast enough for natural voice conversations remains a challenge.
The discussion also considers customer-service systems that must solve problems and earn trust, alongside Otter’s work on digital representations of people that may be used in meetings.
Shawn Wen, PolyAI's chief technology officer, told the HumanX conference last month that systems have been created that can speak and listen at the same time. The next step, he added, is to find answers faster so that conversation sounds natural.
In phone support, these systems shouldn't sound robotic. They should also make the caller believe they can solve their problem. Wen estimated that customers might prefer not to talk to a human if an AI system can help them.
Alex Gay, marketing director at Otter, a meeting recording service, said automation can advance if systems can tell who is speaking, understand what they are asking for, and connect that information to what an organization already knows.
Otter is also working on digital versions of people that could be used in meetings. Gay said their voices need to be able to express emotion so they can engage in conversations beyond simple questions and answers.
Getting it right remains a challenge. Wen said systems often miss important words, resulting in the full meaning of a conversation not being captured. Gay added that a mistake in the initial capture can lead to incorrect actions and reduce trust in the service. Otter is continuing to improve its capture, and it also needs to improve language support.
Otter also wants to look at ways to let everyone involved in a conversation know that it's being recorded, even when its tool isn't participating. Wen said that in corporate phone calls, it's important for people to know they're talking to an AI.
Top comments (0)