Credit: Thinkstock

We need virtual assistants that talk to each other

IDG NEWS SERVICE: 'Siri, Google Assistant, Cortana and Alexa — figure out a solution and get back to me.'

Publisert Sist oppdatert

Everybody talks about (and often to) the Big Four virtual assistants — Siri, Alexa, Cortana and Google Assistant. But many other companies are working on virtual assistants, too.

Huawei is working on a virtual assistant for the Chinese market.

Samsung offers Bixby on its Galaxy S8 or S8+ smartphones.

Voice recognition giant Nuance offers an enterprise ready virtual assistant called Nina, which specializes in knowing the limits of A.I. and kicking queries over to a team of human assistants when necessary. Nuance this month announced a Nina "skill" on Amazon's Alexa platform.

European telecom Orange offers a virtual assistant called Djingo. A french startup is building an x.ai-like meeting scheduling virtual assistant called Julie Desk. And noHold makes a highly customizable virtual assistant called Albert.

There are dozens of others.

The problem with choice in voice-based virtual assistants is that you have to choose one and stick with it. And you shouldn't have to.

The memorize-the-magic-words option

"You should be able to tell Alexa, 'Ask Siri'," Amazon senior vice president of devices, David Limp, said at a conference this week.

Limp's sketchy vision for how virtual assistants should work together is wrong in three ways.

First, he imagines the user telling one all-purpose virtual assistant to "ask" another specific all-purpose virtual assistant. In this imagining, Apple would create a Siri "skill" and become an adjunct to the Alexa platform, something that's inconceivable. Apple maintains Siri to provide an interface feature to Apple hardware like iPhones, iPads, Apple TV and Macs -- not to provide benefits to Amazon Echo users. Siri will never be an Alexa "skill."

Second, this is how "skills" work on Alexa, and it's fatally flawed. In order to use "skills," the user has to specify the service, then say the magic words that enable that "skill" to produce the desired result. Alexa has thousands of skills. But Amazon is relying on users to somehow find skills and memorize their commands. This is why most "skills" are barely used at all.

Third, the only successful models for the future of virtual assistants that work together is either branding awareness (an assistant functions as part of a wider range of features or another product) or direct subscription payments. Current all-purpose assistants are monetized through hardware sales and advertising revenue.

The idea that virtual assistants of the future will work like Alexa's "skills" is very unlikely.

Limp's scenario is one of several suggesting that virtual assistants need to work together.

The open-source option

Playground Global founder and CEO Andy Rubin founded both Danger and Android, and ran Android for Google for years. Now he's got a billion-dollar startup called Essential that recently announced a line of smartphones and other devices.

One of those devices is a virtual assistant appliance called Essential Home, which Rubin says should run Siri, Alexa, Google's Assistant or any virtual assistant customers want to use. The virtual assistant appliance is for consumers, but the concept could be applied to businesses as well. The model is: We build the virtual assistant appliance hardware, but you (the customer) figure out which virtual assistant to put on it.

Based on what Rubin has said, he has no solution to the problem of getting virtual assistants to work with each other, only to get existing virtual assistants to work with his own Home device. He's offering choice, but the actual decision is left to the user.

The bring-your-own-assistant option

Dennis Mortensen is the CEO and founder of x.ai, which makes a virtual assistant that schedules meetings via email.

Mortensen envisions the coming age of BYOA — Bring Your Own Assistant. In his model, the value of each employee or executive in the future will be not only based on their skills, experience and knowledge, but also on the quality of the virtual assistants they use and their skill in using them.

In other words, virtual assistants used in business will not be developed or mandated by the company, but chosen by the employee and carried from one job to the next. This vision is almost certainly accurate, although I would also imagine that companies will deploy and even build their own assistants to add to those chosen and used by employees.

Mortensen also doesn't address the core problem: How do you choose which assistant to use, or decide which module, plug-in or "skill" is best?

The choose-it-for-me option

It's clear that the major tech companies -- Apple, Google, Microsoft, Amazon, Samsung and others -- are working on general purpose virtual assistant services and that their strategy is to open their platforms to third-party add-ons. It's equally clear that if each add-on has to be discovered and conjured up individually, this entire approach is dead in the water.

Even smartphone apps, which hold the advantage of reminding the user with an icon and providing visual interface reminders, are far too difficult to discover and remember to use. The virtual assistant equivalent of apps -- plug-ins or add-ons that are invisible and require memorization (command-line interfaces) -- simply won't be used. And as a result, the incentive to even create them in the first place won't exist.

The obvious solution is for the general-purpose virtual assistant to select the app for the user.

Right now, virtual assistants make decisions about which source of information to use. For example, when you ask for the weather, it won't check your calendar. It will recognize the category of query and reach out to whatever weather service the assistant is hard-wired to use.

Virtual assistants should work initially like apps, then later not at all like apps. At first individual users should find, and in some cases pay for or subscribe to, specific virtual assistant services such as a meeting scheduler or a flight-booking add-ons.

Once installed, however, the user of those services should be the virtual assistant, not the human.

It should be possible, in fact, to install or subscribe to several flight-booking add-ons. Later, when you want to fly somewhere, you should tell your assistant, "book me a flight to Chicago next Wednesday," and the assistant itself should query all the add-ons to arrive at the best price or best combination of cost, travel time, flight schedule, airline carrier and other factors.

Although this general approach is obvious, it's not obvious that current virtual assistant artificial intelligence is up to the job yet. We may have to wait a few years.

But when it does arrive, we'll find ourselves in Mortensen's BYOA world, where employees will choose the vertical assistants or assistant plug-ins. Employees will choose them, but the assistants will use them.