Inside Amazon Alexa’s venture bets on the future of voice and AI

“Alexa, generate a new rock song.”

Commands like that will soon be an option for Alexa users, thanks to one of a handful of announcements Amazon made at CES around weaving generative AI into its voice platform.

Those included partnerships with Splash, an AI-powered music creation platform powered by voice commands; Character.ai, a cast of fictional personas and historical figures for users to chat with; and Volley, an AI-powered voice game maker.

Two of those startups—Splash and Volley—got their start thanks to backing from Amazon’s voice-focused venture arm, Alexa Fund. The division’s portfolio offers a glimpse into how the company is thinking about the future of AI in voice applications. Tech Brew sat down with Paul Bernard, director of the Alexa Fund, at CES to understand how he bets on big ideas in the space.

The fund started at the same time as Alexa as a way to build up the ecosystem around Amazon’s digital assistant. Bernard said it’s since evolved beyond just voice applications to encompass three broad categories: smart devices, AI, and entertainment.

Alexa Fund operates “much more like a traditional venture fund,” Bernard said. If Amazon identifies a “gap in the roadmap” and is looking for a new capability immediately, that’s more likely to be the purview of the acquisitions team or the partnerships team, he said.

“We’re really about placing bets on the five-year-plus horizon that might eventually become things that are relevant for Amazon,” he said. “But it’s really more about longer-term adjacencies and trends.”

Within the fund’s categories, one of the biggest investment themes at the moment is AI agents, a type of task-specific automations, Bernard said.

“The current class of AI technology is bringing to the fore the potential of automation and task execution that has long been part of the promise of personal assistants like Alexa, but is now much closer to becoming a reality,” Bernard said.

Along those lines, the Alexa Fund has invested in a company called Imbue, which aims to bring more reasoning power to large language models (LLMs). The company is building foundation models aimed at handling more complex tasks, like, say, the various nuances of scheduling or rescheduling a work meeting with a large group, Bernard said.

“So if you think about transformer-based large language models there, they understand patterns of the world. They predict the next word effectively. That’s very good at Q&A and summarization,” Bernard said. “It’s not really great, though, at agentic work and reasoning.”

In the entertainment realm, Bernard said the Alexa Fund is “really interested” in the intersection of generative AI capabilities and existing intellectual property in the media world, later mentioning the Alexa product’s team’s work with Disney on a custom bot as one example of how this could look. He also sees AI-generated music like Splash’s as a big growth area, noting that it “will eventually be trained on existing catalogs and music, once the rights for all that get figured out.”

Entertainment can ultimately serve as an important component for making voice platforms more engaging and intuitive for everyday use, Bernard said.

“You’re already seeing a lot of evolution on the richness of voice experiences, like the…tonality,” Bernard said. “The ability to generate personalized and custom voice sounds is getting a lot easier.”

“You can imagine…a world where Alexa is this assistant that does everything for you. There’s an alternate vision [for Alexa] in which you could hire different AIs for different things in your day, all of which could have different personalities and different voices.”