Add KITT.AI to the growing cluster of startups and established technology companies pushing the boundaries of machine intelligence in Seattle.
A spinout of the Allen Institute for Artificial Intelligence (AI2), KITT.AI received an investment from Founders’ Co-op and the $100 million Amazon Alexa Fund, which points to the technology KITT.AI is working on: natural language understanding. The amount of the financing was not disclosed.
Xuchen Yao and his co-founders, Guoguo Chen and Kenji Sagae, have spent their careers focused on this “moonshot” problem.
“We all thought that making computers understand human language was fascinating,” Yao says via e-mail. “It seems trivial and easy because children do it without obvious effort. But the problem is actually way harder than sending people to the moon (and taking them back).”
It’s worth taking a moment to distinguish natural language understanding from automatic speech recognition, a technology that many industry heavyweights—Nuance, SRI, Microsoft, Google, Apple, IBM, Amazon—have refined over the last decade. I dictated interview questions for Yao into my iPhone while sitting on the deck of an indoor swimming pool filled with kids taking lessons; it was remarkably accurate, even in such a noisy, echoing environment.
Speech recognition “is so common that it has become an infrastructure for (the language understanding part of) AI,” Yao says.
Natural language understanding (NLU) is the much more difficult task of deriving meaning from speech or text—a capability that would move computers closer to
what we think of as intelligence.
“Quite simply, if we can solve the full NLU problem, we’ve solved AI,” Oren Etzioni, CEO of the Paul Allen-founded AI2, writes in an e-mail. “Current systems like Siri and Echo just use templates to provide ‘canned’ responses, so we have a long ways to go.”
The big companies that are making progress on natural language understanding today are doing so in support of their own platforms: Microsoft Cortana, Siri from Apple, Google Now, and Alexa, the voice and intelligence inside Amazon Echo. (Amazon has been notably open with its voice technology, giving developers free access to its Alexa Voice Service and Alexa Skills Kit, which allow them to build new, custom capabilities using voice controls and connected devices. The company is backing the best of them through its investment fund, announced last summer.)
“There’s still quite some distance to NLU becoming an infrastructure that’s commonly available to every company,” Yao says.
Therein lies KITT.AI’s business: “We are a [software as a service] company that’ll provide Natural/Spoken Language Understanding as a service,” Yao says. “We will build an open platform and community for developers.”
Individual developers will have free access to the technology. The company will charge businesses, he says.
Founders’ Co-op managing director Chris DeVore fleshed out KITT.AI’s plans a bit more in a blog post announcing the funding last week: “KITT will begin releasing its first developer-facing products this Spring, including a customizable hotword detection engine”—co-founder Chen helped create the “OK Google” hotword technology—“and a conversational understanding service, with a longer-term vision of becoming the ‘Twilio/Slack for NLU’.”
KITT.AI says its hotword detection technology is lightweight, with a runtime of less than 3 megabytes, and does not require an Internet connection. Other applications of the company’s technology include answering natural language questions from Web text or knowledge bases and parsing language into executable code.
The company’s Semantic Lighting API (application programming interface) is an example of the latter. It allows nuanced control of LED smart lights with natural language commands. For example, in response to the command, “Give me something romantic,” the lights turn red.
The API works
