Friendly avatars: How AI is going mainstream

A decade ago we had Clippy, the offensive paperclip, who popped up whilst we were trying to do things. Many people hated him with intense passion. Now we have Siri, the friendly helper – she may be amusingly limited but many people do love her. These two (can I say individuals?) may have their differences but are both, loosely speaking, the same thing… avatars we emotionally engage with.

And their population is multiplying steadily. From Dom [YouTube advert], Domino’s Pizzas friendly US personality, to Frank, MyWave’s new personal assistant, which looks likely to sweep across the globe sometime very soon.

So what does this mean for life and business?

“Clippy treated everyone the same and didn’t adapt over time,” explains Stephen Hoover, CEO of Xerox PARC during his keynote at the Intelligent Assistants Conference (IAC) earlier this week. This is why he annoyed people so much. Now this is exactly what’s changing in the avatar and personal assistant space.

Lana Novikova Co-Founder of Heartbeat Technologies, an AI and emotion analytics startup summarised the situation rather neatly in a panel discussion. “My 10 year old daughter talks to Siri like she’s her friend. So, when I asked my daughter what I should say to a group of experts on the subject she said: ‘Have Siri remember what I asked yesterday!’”.

Personal assistants have come a long way since Clippy. But hardly surprising they are nowhere near there yet. They don’t really know our preferences. But they will. “[Think] if 15 years ago you’d pulled out a Discman and tried to talk to it you’d have looked crazy,” Andy Mauro, who leads the Cognitive Innovation Group at Nuance tells me at the event.

“Some tech takes ages to mature,” he adds. “Nobody says geez the evolution of the internet is slow. This is the beginning of consumer interaction with AI.”

My attendance at this event was courtesy of Nuance and yet the story presented by everyone there is remarkably consistent. And while everyday acceptance of this technology may be driven by the likes of Apple and Google, this in turn only serves to set higher expectations in the business space.

The most obvious example of this is in a call centre, where a negative experience can seriously annoy people and can have a shocking impact on brand reputation. The solution that pretty much everyone has hit upon is to use virtual agents to answer simple queries, to escalate the problem to real people where necessary, and to use real call centre conversations to train the machines to do better.

“It is easy to say I am going to have my virtual agents learn from humans,” warns Mauro “but [in the end] it is all going to come down to delivery.”

There are numerous players in the space – both big and small. Yet as the technology is so cutting edge, and C-suite buy-in is needed to make it work across numerous company siloes, it does seem likely larger organisations will have the edge.

Naturally enough Nuance is very keen to place itself at the centre of all this development. And Mauro who has been with the company for 15 years – an incredible tenure when you consider the changes that have taken place – is very passionate about both the possibilities and the company.

“In the enterprise space people don’t care who they talking to,” he says. All most customers care about is the quickest easiest fix to their problems. This means the actual avatar can be irrelevant. But more is expected in its place.

Martin Kedback, head of business development and support at large Swedish bank, Swedbank – has successfully deployed Nuance’s Nina virtual agent in a Google-style chat box. He provides some insight into how this works in practice. The overwhelming majority of his customers ask the same three questions: “Where is the nearest branch?” “How do I get foreign currency?” and “Where can I get cash?” Taking these three out the mix alone, frees up employees to deal with more complex and ruminative challenges.

Case studies are always positive though, so what has proved a let-down? “Nina could be more intuitive about questions,” says Kedback adding that he and his colleagues had too high expectations about her ability to learn.

Yet this is not really the fault of Nuance or any other companies in the field. It is more a reflection of the state of the industry. After all, we are still at the start of a fairly long trajectory but Mauro believes in around a decade the fully personalised experience complete with – non-annoying – anticipatory and predictive behaviour will be standard.

So, what is holding it back? To manually program all the information needed to make today’s virtual agents really smart would “cost too much and take too long” says Mauro. The breakthrough will only come when computers truly learn from people. Yet this is a tricky thing to achieve because there are two sorts of information. There is hard statistical data and there is the softer more difficult to pin-point work-based know how that exists in all our heads. “The grey net” as Mauro coins it.

“The industry is somewhat marred by a binary approach,” he adds ruefully.

And so, in a way this takes us full circle back to avatars. If we are presented with a virtual agent it is the overt ‘personality’ of that machine that will endear or repel us. This means when you side-line the avatar – because after all, you only need an answer – it is only by giving it the broader depth of human ‘personality’ that it is likely to work. 

Animating a paper clip is by far the easiest option…


Further reading:

2014 Report: Are ‘Mind Reading’ Apps Good or Bad?


« Acer is bringing free computers to schoolchildren in Chile


"Virtual Assistants": Where is customer service heading? »


Do you think your smartphone is making you a workaholic?