Microsoft reveals AI tech for virtual agents worth talking to

Chatting with Cortana and other virtual agents looks set to get a lot more useful, with Microsoft announcing a new breed of intelligent agent at Build 2019. The software behemoth showed off its latest conversational interface at the annual developer event, debuting a completely fresh way of training – and interacting with – assistant technologies.

Traditionally, virtual agents would be trained on a set of skills or abilities. As a human user talks with them, and makes requests, the virtual agent compares those queries with the talents it has been educated to enact.

It's enough for the basics, but it's still hardly witty conversation, nor especially useful. The agents are limited by their programing, and how that programing can be combined. At its most basic, that cam mean an agent that can't track one interaction to the next, forgetting the prior discussion at each stage.

More damaging, though, is the fact that such traditional agents typically can't combine multiple skills together for a result that's greater than the sum of its parts. It's there, among other things, where Microsoft believes it can carve out a niche.

Its conversational AIs are a lot more flexible than the voice interfaces we're familiar with now. For a start, there's memory: the agent can remember what was jus being talked about, and so infer meaning from one stage of the request to the next. If you were working with a document, for instance, you could potentially ask "Hey Cortana, show me all Word files from March 2019," and then follow up with "only those over a megabyte."

It's not just memory of conversations, though. The new virtual agent framework also supports multi-domain and multi-agent experiences, combining different agent platforms and ecosystems into one. Different skills and back-end services can be woven together in the background, delivering more capable and more contextually-aware agents to users. It needn't even be limited just to Microsoft, either: external services can also be tapped into.

Microsoft hasn't done all this on its own, mind. The company acquired Semantic Machines last year, a conversational AI specialist working on speech and text training for agents. The startup caught Microsoft's attention in part by its so-called Conversation Engine. That promised to tease semantic intent from voice and text, then automatically build a self-updating and learning framework by which the nuances of the conversation could be analyzed and understood.

The future, Microsoft suggests, is one where every organization has its own roster of agents to call upon. Not only that, like the most capable teams those agents would be able to seamlessly collaborate and interoperate. With products like Cortana-powered "Surface Buds" rumored to take on AirPods, complete with a smarter AI whispering in your ear, that could well give Microsoft a significant edge.

The proof of that pudding is in the tasting, of course, and for that we'll have to wait. Microsoft stays that the new conversational engine is going to be integrated into Cortana, as you'd expect, but also made available to external developers via the Microsoft Bot Framework and the Azure Bot Service.