Siri's getting an upgrade. Here's some advice from someone who's been there

By Richard Nieva , written on January 22, 2013

From The News Desk

Late last week, Apple put up a job listing. Not for a developer, per se, or a high level executive, or new members of its retail army. It’s looking for more of a creative writer. The company is aiming to teach its old Siri new tricks. Creative, witty tricks. According to 9to5Mac, the listing called for someone with a “love for language, wordplay and conversation.”

We’re looking for a uniquely creative individual to help us evolve and enrich Siri, our virtual personal assistant. Siri’s known for ‘her’ wit, cultural knowledge, and zeal to explain things in engaging, funny, and practical ways. The ideal candidate is someone who combines a love for language, wordplay, and conversation with demonstrated experience in bringing creative content to life within an intense technical environment.

So who should Apple hire? One person who might have some good advice is Robert Hoffer, who headed what was then called ActiveBuddy, and developed what some might call Siri’s grandfather: SmarterChild.

SmarterChild was an instant messaging bot that you could chat with and it respond with information or sassy quips. If you had an AIM account in the mid-to-late 90s, you might remember it. You’d ask it to define words for you, or find movie listings, or – if you were a juvenile teen at the time -- obnoxiously curse at him while it demanded an apology. Sorry SmarterChild. Microsoft bought the company in 2002 and ultimately shut it down in 2009.

Hoffer is now the cofounder and president of AB2, a company focused on voice recognition for social networking apps. In a first-ever interview with the press about SmarterChild and its origins, I chatted with Hoffer today to see what makes good creative material for an artificially intelligent bot, the dangers of striking the wrong tone and what helps make a bot feel lifelike.

PandoDaily: What should Apple keep in mind when looking for Siri writers?

Hoffer: I think you have a couple of problems there. Once we dealt to Microsoft, we translated our system to nine or 10 languages. And so you need to have multiple writers because things that are interesting in English aren’t necessarily interesting in Mandarin. So one problem is the internationalization of the personality.

The second problem is finding writers that can strike the correct balance between being helpful and being funny or amusing. And if we were all funny and amusing, we’d all be writers.

The third problem is, you hamstring your creative talent because you need to define a personality where, if you slide the bar to too much sarcasm, you run the risk of being too offensive. And today, there’s no ability to trim that down on Siri. You can’t tell it how sarcastic you want it to be.

So what kind of tone does Apple need to strike?

You have the all problems of creating a character for the mass market. And the problem with creating a character for the mass market is, if you drive in the center of the road, you get hit by a car going in one direction or another. So, popular characters who are famous declare one side or other of the personality divide. So you can be very popular if you’re really, really sarcastic, for example. But only with about half the people. You can be popular if you’re really serious, but only with about half the people. So to create this namby-pamby generic character is very difficult.

You also can’t make it too artificially intelligent, or you introduce what’s called the uncanny chasm. That is, there’s a point at which a robot becomes uncomfortably creepy. It knows you too well. We had this application we developed called Knock Knock, and nobody ever let us launch it. One of the things that Siri doesn’t do is ever initiate the conversation. But that’s not how your friends behave. They message you all the time. So we had a robot that tells you knock knock jokes. We tried it on AOL – freaked people out.

What was your audience like?

We made SmarterChild a little sardonic and sarcastic, which is why the market we ended up capturing was the youth market. It skewed heavily young, like 70 or 80 percent teens. We launched on AOL AIM, Yahoo Messenger and MSN Messenger. And we would see traffic spike when it turned 3 p.m. on the coasts and all the teenagers were getting out of school.

We suspect Siri is appealing to the same sort of people. For example, my 13-year-old daughter loves Siri. I don’t particularly like it, because I don’t find it particularly helpful. Creating a personality for Siri is an ongoing struggle for Apple, on top of the struggle they have with natural language recognition.

What are some of the struggles dealing with younger users?

I had friends that were involved deeply in story at places like Pixar, and I relied on them to consult. And there were times when we still went over the line. One of the things people liked to do with SmarterChild, and delighted in, was trying to have it consent to having sexual relations with them. And that was an ongoing fight, because you could trick the bot. So one of the things you can’t do is have it repeat back what the consumer says. If you do, it might repeat back a phrase you might not have caught in your cache filter that’s very inappropriate.

But Apple has users all across the board.

Yes, that’s why I believe there can’t only ever be one bot. My reason is that, on the one hand you have characters like Bart Simpson or Bugs Bunny. Those characters have immensely popular personalities. But they don’t have 100 percent appeal. So you have to begin to create a range of personalities that are popular in certain niches.