It’s the most polite throwdown you’ll ever hear.
Put Amazon Echo and Google Home in a room together and what do they do? Apologize.
Modeled after the traditionally subservient and ever-aiming-to-please secretary — a fiction at best, an expectation and demand at worst — Echo and Home excel at such intellectually untaxing tasks as setting timers, reading weather forecasts and telling you when your next meeting is. Unsurprisingly, perhaps, they also excel at apologizing. Put the two of them in a room together as part of an experiment to see if they can carry on a conversation sans human intervention, and it’s a sea of “I’m sorrys.”
“Sorry,” Home says, “I don’t know how to help with that.”
“Sorry,” Echo says, “I didn’t understand the question I heard.”
Better still is when they apologize to each other—two robotic servants unable to understand each other’s automated responses. They’re polite, that’s for sure. Apologize to the Echo—powered by Amazon’s virtual assistant, Alexa — and it’ll tell you it’s all right, and Google Home will chime in, too.
But the best is when the two get stuck in an endless loop together, one that’s stopped only by barking at one or both of them to “stop” or “shut up.”
It’s not a real conversation, of course — at least not a conversation you’d likely want to be a part of. But it’s a start, and it’s not hard to imagine a future in which these devices will have more realistic dialogues, ones in which they build on each other’s knowledge (or try to outdo the competition by one-upping it with information).
Rather than offering to show me Bing search results on my phone’s Alexa app, for example, it would be nice to see Echo outsource the question to a nearby friend — the Google Home. That would take cooperation among the tech giants, though, and at this point, that seems even less likely than a virtual assistant that can understand you perfectly.
And that’s the next hurdle these devices will need to clear: understanding us better. When I ask Echo a question it can’t answer, it should ask me to rephrase the question, or ask what I mean by something.
And they need to get better at understanding follow-up questions. Though Google Home can, for example, respond to “How are you?” by saying “Awesome! Thanks. How can I help?” it’s only within these narrow confines that a back-and-forth can occur. Otherwise, it’s like talking to an inanimate object (which it is, literally): After Google Home answers my weather question and tells me it’s going to rain today, I ask “When?” but get nothing; neither Google nor Alexa understands what I’m talking about. (Even when I say “Ok Google, when is it supposed to rain?” I get nothing — that’s another problem.)
Voice assistants should be able to remember the question I just asked and, using that context, answer follow-up questions. Or they should, at the very least, ask me if I’d like to know more. They should be like a good butcher, handing you one item while asking, always, “What else?”
But right now, most virtual assistants can’t handle nested inquiries and ideas, and they don’t have a firm enough grasp on context to keep a conversation going. What’s more, sometimes they’re simply not listening. Unless you’re prompted to give a response, or unless you say “Ok, Google” or “Ok, Alexa” first, the devices don’t know you’re talking to them — which is great for privacy, not so great for mimicking the way humans communicate.
The way around this isn’t to give up and let these devices listen to everything we say; it’s to imbue them with the ability to learn when to ask you for a response. To do this, developers will have to sharpen virtual assistants’ natural language processing and machine learning, so the more we talk to these assistants, the better they’ll be at understanding us.
Developers also need to give these devices self-awareness, so that the virtual assistants know when they don’t know something. When Watson was on Jeopardy, for example, it calculated not only a response, but also a level of certainty. Ideally, the Home and Echo of the future will be able to do something similar, and when they’re uncertain, ask you for clarification.
It’s one thing to ask our virtual assistants to know a ton; it’s another to ask them to know all the tons they don’t yet know—to have a certain kind of intellectual self-awareness.