Talking to Alexa is too often a one-sided affair. Certainly, the primary purpose of a digital assistant is to listen, but its responses tend to favor cold efficiency -- get in and get out. If a human spoke to you the same way, you'd think they were being passive aggressive in order to get you to shut up and leave them alone.

While this dynamic makes sense from a utilitarian perspective, it runs counter to the stated goals of all digital assistants (Siri, Google Assistant, and the rest) -- that being conversational language.

To really take their usefulness to the next level, digital assistants need to understand more than just words. They have to take into account context and even the mood or personality of the speaker.

SEE ALSO:You can now use Alexa as your home's PA system

Most digital assistants use AI to understand context to some degree, so they can respond to follow-up queries. Google voice search has been able to respond to follow-up questions since at least 2013. You can ask, "Who is the prime minister of the UK?" and, when Google tells you it's Theresa May, you can follow up with, "When was she born?" to see that May was born on Oct. 1, 1956.

However, that kind of contextual awareness doesn't usually last beyond a couple of additional queries. Also, real, human conversations involve both parties often listening and speaking at the same time. If you're responding to someone IRL, you're constantly reading their expressions and adjusting. You may even interject from time to time, so it's essential that the other party is also observing, ready to respond to such moments.

The next frontier

Digital assistants are really bad at all that subtle stuff that happens in normal conversations today. But Microsoft snuck out a recent announcement that gives us hope for the future. The company claims to have taken a big leap in conversational AI by introducing "full duplexing" -- that is, the ability to have a verbal conversation with an AI-driven assistant where it can speak and listen simultaneously.

“It’s the biggest deal," Shane Mac, CEO of Assist, an automated assistant platform, told Mashable. "This is the solution for the ‘I do not understand’ problem. This is how the act of listening becomes understanding. When [digital assistants] start doing this stuff, people will say, 'Holy shit, that’s what i want.’”

Not needing to say "Alexa" or "Cortana" wake words over and over would certainly be a convenience, though it would also mean we'd need to re-orient our expectations for when an assistant is actually listening to the content of what's being said and not just buffering audio to scan for a wake word. That changes the privacy equation, and users would need to clearly opt-in to being listened to -- and thus recorded in some form -- more often.

Mashable Light SpeedWant more out-of-this world tech, space and science stories?Sign up for Mashable's weekly Light Speed newsletter.By signing up you agree to our Terms of Use and Privacy Policy.Thanks for signing up!

"You have to give the customer the controls to be in control of their privacy," says Mac. "At the end of the day, you have to balance privacy and convenience. Everyone’s going to want this."

Credit: Microsoft

But full duplexing tech has the potential to make our interactions with digital assistants much more natural. You'll be able to speak further instructions when the AI is in the middle of responding to something else -- say, telling it to dim the lights when it's in the middle of reading back an email.

That doesn't necessarily mean the conversations are more rude, just more human. How many times have you remembered something just when the person you're talking to just began a long diatribe? "Hold on a second" usually does the trick without offending, and would theoretically work with an assistant capable of full duplexing.

"We know that this is a direction that the whole [digital assistant] space is headed towards," says Kiesha Clayton, senior communications manager for Microsoft AI. "Instead of having a one-trigger turn conversation, it's this full-duplex, multi-turn conversation where you can keep chatting, you don't need the wake word."

China leading the way

Full duplexing isn't available in any of the major U.S. assistants, not even Microsoft's voice in the space, Cortana. (Google, Apple, and Amazon declined to comment for this article, though it's clear none of their assistants offer full duplexing right now.) Microsoft has only applied the tech to a "companionship" chatbot called XiaoIce (pronounced shiow-ice), a chatbot Microsoft introduced in 2014 and is very popular … in Asia.

"Xiaoice delivers the weather on news stations, she reads poetry, she sings songs," explains Clayton. "In China, they're really leading the way with what we could see as the future with these bots."

Thanks to Microsoft's partnership with Xiaomi, XiaoIce is now on several smart home devices, and the chatbot is said to have more than 200 million users. At least one of them ended up talking to XiaoIce for longer than 4 hours, Microsoft claims, with more than 1,600 individual verbal exchanges.

That's one Chatty Cathy -- a rap session like that is almost inconceivable when stacked against the rote weather updates and timer-setting that occupy the vast majority of Alexa's and Siri's verbal bandwidth.

"Right now chatbots are better at being answering machines than they are conversation enablers," says Michael Facemire, a principal analyst at Forrester Research. "Most chatbots let you ask one question, and get one answer back. It's very, very important that we have [full duplexing] longterm."

Still, it's hard to gauge just how personal or useful XiaoIce's responses are without a transcript. And let's be real -- Microsoft's record with chatbots is mixed, to say the least. It inadvertently allowed the internet to train its Twitter bot, Tay, into repeating racist garbage, and its newer effort, Zo, isn't exactly lighting the world on fire with its thoughtful responses. Just check out this excerpt of my conversation with it:

In other words, full duplexing is a great tool to create conversations, but the AI that drives them clearly needs work. Still, if we're ever really going to get to the semi-omnipotent Star Trek computer we all dream about, the ability to listen while talking is something all digital assistants will need to learn. And, for once, Microsoft gets to lead the conversation.

---

Featured Video For You

---

This cute robot is every hacker's ultimate nemesis

---

(责任编辑：关于我们)