© 2021 WYSO
Our Community. Our Nation. Our World.
Play Live Radio
Next Up:
0:00
0:00
Available On Air Stations

The Prime Effect: Alexa, AI And What Makes Jeff Bezos Tick

Jeff Bezos speaks at the the Amazon re:MARS convention in Las Vegas. (John Locher, File/AP Photo)
Jeff Bezos speaks at the the Amazon re:MARS convention in Las Vegas. (John Locher, File/AP Photo)

Who is Jeff Bezos? In the second part of our series Amazon: The Prime Effect, we hear what the creation of Alexa tells us about the man who built Amazon.

Guests

Brad Stone, senior executive editor for Bloomberg Technology. Author of “Amazon Unbound” and “The Everything Store.” (@BradStone)

Also Featured

Janet Slifka, director of research science in the Alexa AI division at Amazon. She’s worked on Alexa technology since 2012.

First Person: ‘Origins Of Alexa’ Transcript

This hour is our second in our series Amazon: The Prime Effect — where we take an in-depth look at the myriad ways Amazon has changed the way we shop, work and how we live. In this episode, we focus on Amazon founder Jeff Bezos — his ambitions, and how he built the company and its culture.

[Tape] BEZOS: Our vision was in the long-term it would become the Star Trek computer. You could ask it anything and ask it to do things for you, to find things for you and it would be easy to converse with in a natural way.

CHAKRABARTI: That’s Jeff Bezos in a 2016 interview, describing his initial inspiration for Alexa, which began development in 2011. Consider the ambition behind that statement. He wanted to develop a computer, like the one in Star Trek. A seemingly all-knowing, linguistically sophisticated computer that you could chat with — frustrated futuristic Scottish brogue or not — but at that point, a computer that only existed in television fiction.

JANET SLIFKA: What was exciting about it [was] Amazon was stepping into that area without having a long history in that. Like at that time in 2012, if you listed the technology companies that worked in language understanding, I don’t think Amazon would have been in the list.

SLIFKA: My name is Janet Slifka. I’m a director of research science in our Amazon Alexa AI organization.

CHAKRABARTI: Slifka had the kind of expertise Bezos needed — a career in speech and acoustics that had already taken her to MIT and speech recognition software companies. She joined Amazon in 2012. The Alexa project was still secret.

SLIFKA: It was early, we didn’t have anything that worked yet. But lots of enthusiasm to put something together.

CHAKRABARTI: Enthusiasm and urgency, because remember, Jeff Bezos wanted to make something like the Star Trek computer. So, to make that happen, the product had to be smart, it had to be able to learn, which meant, the Alexa team needed a ton of data.

[Tape] BEZOS: Very challenging from a technical point of view. There were probably more than 1,000 people working on Echo and Alexa.

CHAKRABARTI: Data, in this case, meant human speech. People talking.

SLIFKA: The notion of what would delight customers would be that you don’t have to go find a device, and unlock it, and bring it up and speak to it, which certainly was the model I had been working on prior to joining. It was this nice, big idea about, Look, I just want to talk. I just want to speak out in the open air and have things happen. Like, that would be magical.

CHAKRABARTI: So, the team handed out Alexa-enabled prototypes to Amazon employees. Including Jeff Bezos, who tested Alexa at his home. It did not go well.

As Brad Stone writes in Amazon Unbound, engineers reviewing the data heard their boss get so frustrated by Alexa’s lack of comprehension, they can hear him telling the machine, ‘Go shoot yourself in the head.’ But, I mean, when you want the Star Trek computer …

CHAKRABARTI: But then Janet Slifka came in with an idea. It became known internally as AMPED.

SLIFKA: [Laughs] Um I was very involved in that.

CHAKRABARTI: No longer satisfied with employee testing at home — that didn’t give them enough data — the Alexa team went into the field.

SLIFKA: We would call this, in the lingo, a supervised data collection.

CHAKRABARTI: They rented out different homes and apartments initially in the Boston area … where they paid a constant stream of random people to come in and read a script.

SLIFKA: We can’t just use a studio room that we’ve set up in a lab and think that that’s going to be good enough. Right? We wanted to say, let’s get out into real spaces that, as you know, can be quite quirky. So you want to have confidence that you’re meeting the expectations of all of your customers.

CHAKRABARTI: But of course, the people reading the script had no idea why they were doing it, or that Alexa prototypes, hidden under cloth, were listening. So, in addition to the scripted speech, engineers collected a lot of … let’s call it, colorful, frustrated language as well.

CHAKRABARTI: But, by 2014, as Brad Stone reports, AMPED had done its work. Amazon had increased its store of speech data by a factor of 10,000. While Bezos didn’t know about AMPED ahead of or during the program, he was happy with the results.

CHAKRABARTI: The Amazon Echo, enabled with Alexa, launched on November 6, 2014. Janet Slifka remembers the moment.

SLIFKA: The moment of launch you don’t know what the first feedback is going to be. And of course, first, feedback’s all over the place. But I remember a couple comments were a little bit like, oh, well, what is that? And why would I want that? [Laughs] There was a bit of puzzlement that very quickly turned into delight, like, oh, wait, this is actually awesome.

CHAKRABARTI: Amazon sold about 5 million devices in the first two years. In 2016, Bezos paid $10 million for a Super Bowl advertisement featuring football great Dan Marino and actor Alec Baldwin.

[Tape] 2016 SUPERBOWL AD:

BALDWIN: Alexa, how many championships has Dan Marino won?

ALEXA: Dan Marino has won zero championships.

MARINO: Alexa, how many Oscars has Alec Baldwin won?

ALEXA: Alec Baldwin has won zero … 

BALDWIN: Alexa stop. Well played, Marino.

CHAKRABARTI: Alexa is one of Amazon’s most successful products. But what, fundamentally, is its purpose? Well, ask, Alexa:

[Tape] ALEXA: I was made to play music, answer questions and be useful.

CHAKRABARTI: Janet Slifka says it’s that last attribute that defines how Amazon corporate culture, and Jeff Bezos’s relentless drive flows into Alexa’s development.

SLIFKA: I think fundamentally that phrase about be useful, there’s so much in that … because it does imply a personalization. Be useful to me as a customer. And that means deliver me the experiences I want, with the language I want, change with me as what useful means for me as a customer. So it might look like a fairly short set of words to be useful, but there’s so much in that. And I think that’s what we keep striving to deliver.

CHAKRABARTI: For example, right after launch in 2014, Bezos instructed the team to develop new useful Alexa skill sets every week. They pretty much still stick to that, seven years later.

SLIFKA: The way that innovation at Amazon thrives, or at least in my experience thrives, is because of the way the leadership principles capture that philosophy, that proven track record of success for fostering innovation. And I think [that] in a way that’s so baked in, it’s so integral. And that’s just the fabric of every day and is a constant part of Amazon. Independent of how people shift their focus.

CHAKRABARTI: Recall, Janet Slifka is the director of research for Alexa AI So what does she want Alexa to be able to do next?

SLIFKA: Addressing how creative people are with all languages, and mixing languages.

CHAKRABARTI: It sounds so simple. But human language is incredibly complex. Slifka is describing technology vastly more sophisticated than a directive-processing machine. For Alexa to be able to understand how, when, and why people mix languages, the AI will have to understand things like context, culture, cues, idiom, innuendo and meaning.

CHAKRABARTI: Huh. Imagine that capability, that power, leveraged across one of the largest corporations on Earth.

[Tape] BEZOS: We as humanity, as a civilization, as a technological civilization, are still quite a ways away from being about to be as magical and amazing as the Star Trek computer. But I think, I don’t know, when is Star Trek? So, we still have a couple of centuries. I don’t think we’ll need that much time, actually.


From The Reading List

Bloomberg Businessweek: “The Untold Story of How Jeff Bezos Beat the Tabloids” — “‘Raise your hand if you think you’ve had a harder week than I’ve had.’ It was Feb. 14, 2019, in the early afternoon, and for perhaps the first time in the 25-year history of Amazon.com Inc., Jeff Bezos was prepared to explain himself to his employees.”

New York Times: “Inside Amazon: Wrestling Big Ideas in a Bruising Workplace” — ” On Monday mornings, fresh recruits line up for an orientation intended to catapult them into Amazon’s singular way of working.”

This article was originally published on WBUR.org.

Copyright 2021 NPR. To see more, visit https://www.npr.org.