Windows Phone is still a distant third to Apple andAndroid in the smartphone market, but Microsoftis hoping to change that with the introduction of Windows Phone 8.1- and more importantly itspersonal digital assistant Cortana. Microsoft claims that Cortana isn't like your average virtual assistant. She's supposed to be a little wittier, more personable, and capable of learning more about you than Siri or Google Now. Business Insider talks to Microsoft's Marcus Ash, Partner Group program manager. Excerpts:
Microsoft interviewed real-life personal assistants when creating Cortana.
If you want to make a real humanistic connection with that technology, the best thing you can do is find a set of humans that do the job we think this phone should be able to do. We [asked] them, 'What do you do to really make the person that you work for happy? What types of [tasks] do they ask you to do?'
The other area we were focusing on was how much personality we should attribute to this assistant. These machine-learning systems need a lot of data. So if you don't ask the right questions, then you're not going to get the right data, and then the system can't train itself.
You need a pleasant sounding voice. You need to make sure that voice sounds as human as you can possibly make it. When you ask a question that you would ask a normal person, the system should respond the same way a person would respond. So we really thought of all those problems.
BI: In what other ways did you study real personal assistants?
MA: We interviewed these people that had these high-stress jobs, meaning they were assisting people who were celebrities where it really matters that you're getting things right. We interviewed between five and seven assistants over the course of one week.
And we had them keep a journal, and we looked through those journals and we looked back and did exit interviews. We asked them tell us about the relationship with the person [he or she] worked for. We said, tell me the types of things you do for them. Tell me how much they have to ask you to do things for you versus how proactive you are. That's where we got a lot of insight.
It's all about trust. This person tells me very private information. And this person expects me to keep this private information between us. They didn't go into details, but you can imagine the kinds of things that an assistant that follows that type of person around might see or hear.
BI: How do trust and personality translate to a phone?
MA: The notebook in our case belongs to Cortana. It's actually her view of you. It's based on what you say, and it's based on you giving access and having this trust relationship build. She'll never put anything in your notebook that you're not aware of, or that you don't trust. But she'll make inferences about you based on the information you tell her.
This is a very personal thing for people. Especially as we look forward to the future at all the types of things that our devices are going to know about us. Knowing that you can trust this device to do the right things with that data is really one of the key points we honed in on early by talking to these assistants. Personality, I thought, was a real breakthrough for us. If you don't have a personality, it's really hard for people to trust you.
BI: Can Cortana make the virtual assistant more valuable?
MA: People have been working on speech systems for years. They're very complex. They're very difficult to get right. The thing that gives me a lot of hope on these systems is that we've reached a point where we're collecting so much data about speech that the speech systems are improving at such a rapid rate. It's much better than it was five years ago, and it's growing at such an exponential rate because there's so much data being poured into smartphones. So that makes me feel great.
BI: Could Cortana influence shoppers to turn to Windows Phone rather than Apple or Android?
MA: We think it's [the virtual assistant] going to be one of the next big things that distinguishes these platforms.
How good is the assistant and the contextual learning technology on this phone? So I think that's a longerterm vision. Even in the shorter term, the idea that Windows Phone has done something that's interesting and unique and that we've got a distinct point of view about, it feels like we're getting a lot of pickup.
BI: What makes Cortana different or better than Google Now?
MA: We focus a little bit more on contextual triggers that we think people will actually understand. We always talk about 3 triggers-we talk about time, we talk about location and we talk about people. Those are the triggers people get. Let's just focus on a set of things that are going to be of high utility and limit the number of triggers so that people can understand what the system is capable of.
When we look at Google, they've made some pretty clear decisions.
It's about getting you quickly and efficiently to Google's services. It's not about personality. There's just something really delightful that makes people smile about having an anthropomorphic personality inside this assistant. We studied this a lot and looked at people's reaction in labs; it just makes people smile. It also opens up this type of trust relationship we talk a lot about. It's ultimately a question of who can get the most information in a privacy- controlled way that's still personal.
Microsoft interviewed real-life personal assistants when creating Cortana.
If you want to make a real humanistic connection with that technology, the best thing you can do is find a set of humans that do the job we think this phone should be able to do. We [asked] them, 'What do you do to really make the person that you work for happy? What types of [tasks] do they ask you to do?'
The other area we were focusing on was how much personality we should attribute to this assistant. These machine-learning systems need a lot of data. So if you don't ask the right questions, then you're not going to get the right data, and then the system can't train itself.
You need a pleasant sounding voice. You need to make sure that voice sounds as human as you can possibly make it. When you ask a question that you would ask a normal person, the system should respond the same way a person would respond. So we really thought of all those problems.
BI: In what other ways did you study real personal assistants?
MA: We interviewed these people that had these high-stress jobs, meaning they were assisting people who were celebrities where it really matters that you're getting things right. We interviewed between five and seven assistants over the course of one week.
And we had them keep a journal, and we looked through those journals and we looked back and did exit interviews. We asked them tell us about the relationship with the person [he or she] worked for. We said, tell me the types of things you do for them. Tell me how much they have to ask you to do things for you versus how proactive you are. That's where we got a lot of insight.
It's all about trust. This person tells me very private information. And this person expects me to keep this private information between us. They didn't go into details, but you can imagine the kinds of things that an assistant that follows that type of person around might see or hear.
BI: How do trust and personality translate to a phone?
MA: The notebook in our case belongs to Cortana. It's actually her view of you. It's based on what you say, and it's based on you giving access and having this trust relationship build. She'll never put anything in your notebook that you're not aware of, or that you don't trust. But she'll make inferences about you based on the information you tell her.
This is a very personal thing for people. Especially as we look forward to the future at all the types of things that our devices are going to know about us. Knowing that you can trust this device to do the right things with that data is really one of the key points we honed in on early by talking to these assistants. Personality, I thought, was a real breakthrough for us. If you don't have a personality, it's really hard for people to trust you.
BI: Can Cortana make the virtual assistant more valuable?
MA: People have been working on speech systems for years. They're very complex. They're very difficult to get right. The thing that gives me a lot of hope on these systems is that we've reached a point where we're collecting so much data about speech that the speech systems are improving at such a rapid rate. It's much better than it was five years ago, and it's growing at such an exponential rate because there's so much data being poured into smartphones. So that makes me feel great.
BI: Could Cortana influence shoppers to turn to Windows Phone rather than Apple or Android?
MA: We think it's [the virtual assistant] going to be one of the next big things that distinguishes these platforms.
How good is the assistant and the contextual learning technology on this phone? So I think that's a longerterm vision. Even in the shorter term, the idea that Windows Phone has done something that's interesting and unique and that we've got a distinct point of view about, it feels like we're getting a lot of pickup.
BI: What makes Cortana different or better than Google Now?
MA: We focus a little bit more on contextual triggers that we think people will actually understand. We always talk about 3 triggers-we talk about time, we talk about location and we talk about people. Those are the triggers people get. Let's just focus on a set of things that are going to be of high utility and limit the number of triggers so that people can understand what the system is capable of.
When we look at Google, they've made some pretty clear decisions.
It's about getting you quickly and efficiently to Google's services. It's not about personality. There's just something really delightful that makes people smile about having an anthropomorphic personality inside this assistant. We studied this a lot and looked at people's reaction in labs; it just makes people smile. It also opens up this type of trust relationship we talk a lot about. It's ultimately a question of who can get the most information in a privacy- controlled way that's still personal.
No comments:
Post a Comment