Microsoft has acquired Semantic Machines, an artificial intelligence startup which could help Cortana hold more natural dialog with users. The California company has focused its efforts on so-called “conversational AI,” moving beyond the more basic back-and-forth currently supported by the Google Assistant, Apple’s Siri, and Amazon’s Alexa.
Right now, talking with one of those virtual helpers could leave you doubting the “intelligence” part of AI. A big part of that is their difficulty identifying meaning: effectively picking up on the thread of a conversation, and using that to understand different requests and commands.
Semantic Machines is taking a different approach. It has built a number of machine learning components which work together for a smarter AI that can hold that conversational thread. That includes a Conversation Engine which pulls semantic intent from voice or text and then creates a self-updating, learning framework to comprehend the nuances of that discussion. A new speech recognition engine has been built in-house, together with a proprietary speech synthesis engine so that the AI can talk back just as naturally as the user.
Reinforcement learning is a big component for the system. For a start there’s a proprietary data capture and annotation system, which the Microsoft Azure team is likely to be looking at with great interest, and which Semantic Machine plans to use to build a huge training system for AI using both speech and text. That also includes reinforcement learning, with real-time feedback from individual conversations going on to improve the AI’s efficacy as a whole.
While it’s not alone in looking to smarter AI and machine learning, Semantic Machine’s roster of experts do have some serious credentials among them. That includes Larry Gillick, formerly chief speech scientist at Apple; Dan Roth, who created Voice Signal Technologies, one of the foundation parts of voice interface systems on the iPhone and more; and professor of computer science, Dan Klein.
“With the acquisition of Semantic Machines, we will establish a conversational AI center of excellence in Berkeley to push forward the boundaries of what is possible in language interfaces,” David Ku, CVP and chief technology at Microsoft AI & Research, said of the acquisition. “Combining Semantic Machines’ technology with Microsoft’s own AI advances, we aim to deliver powerful, natural and more productive user experiences that will take conversational computing to a new level.”
Artificial Intelligences – and how they can power smart assistants – are in vogue right now, and not just in smartphones. The growth of home AI technology, led by Amazon’s Alexa and Google’s Assistant, has arguably been even more aggressive than in mobile devices, though Microsoft has struggled to make a dent there. Cortana, its assistant, has cropped up in a couple of devices, including Harman-Kardon’s Invoke smart speaker. Nonetheless, it’s failed to gain the same sort of developer traction.