AI Research at Amazon: Brand Voice, Entanglement Frontier, Humor Recognition  

By John P. Desmond, AI Trends Editor    The most popular articles, blog posts, and downloaded publications of 2020 published on the Amazon Science website can serve as an update on AI work at Amazon.  The most popular article of 2020 was an account of advances in text-to-speech technologies. Amazon’s Alexa voice service has been on the market for […]

Nov 30, -0001 - 00:00
 0
AI Research at Amazon: Brand Voice, Entanglement Frontier, Humor Recognition  
Techatty Supportive Club
Techatty Supportive Club

By John P. Desmond, AI Trends Editor   

The most popular articles, blog posts, and downloaded publications of 2020 published on the Amazon Science website can serve as an update on AI work at Amazon. 

The most popular article of 2020 was an account of advances in text-to-speech technologies. Amazon’s Alexa voice service has been on the market for more than five years and is now available on hundreds of millions of devices from Amazon and other device manufacturers.   

Introducing Google Cloud

Advances in text-to-speech (TTS) technologies—the ability of computers to convert sequences of words into natural sounding, intelligible audio responses—have made it possible for computers to sound more human-like. 

Amazon scientists and engineers are working on predicting how the sentiment of an utterance might sound to an average listener, for example, and respond with human-like intonations.  

An advance in the field occurred in 2016, when researchers at DeepMind, the London-based AI firm owned by Google, introduced WaveNet, which could generate realistic voices using a neural network trained with recordings of real speech.  

Andrew Breen, senior manager of the TTS research team, Amazon

“This early research suggested that a new machine learning method offered equal or greater quality and the potential for more flexibility,” stated Andrew Breen, senior manager of the TTS research team in Cambridge, UK. Breen has long worked on the problem of making computerized speech more responsive and authentic. Before joining Amazon in 2018, he was director of TTS research for Nuance, a Massachusetts-based company that develops conversational artificial intelligence solutions. 

Amazon Polly is a cloud service from AWS that converts text into spoken audio. Amazon recently announced a new feature called Brand Voice, which provides the opportunity for organizations to work with the Amazon Polly team of AI research scientists and linguists to build an exclusive, high-quality, neural TTS voice that represents their brand’s persona. Early adopters include Kentucky Fried Chicken (KFC), which features Colonel Sanders in its advertising.   

“The ability for Alexa to adapt her speaking style based on the context of a customer’s request opens the possibility to deliver new and delightful experiences that were previously unthinkable,” stated Breen. 

 

Progress Seen on Challenges of Quantum Computing  

John Preskill, Physics Professor, Caltech, advisor to the National Quantum Initiative

Among the top blog posts on Amazon Science in 2020, was an account on the challenges of quantum computing with thoughts from John Preskill, a physics professor at the California Institute of Technology, and an advisor to the National Quantum Initiative, who joined Amazon as a scholar in 2020.   

Introducing Google Cloud
Introducing Google Cloud

A quantum bit, or qubit, can take on the values 0, 1, or, in a state known as superposition, a combination of the two. Quantum computing depends on preserving both superposition and entanglement, a fragile condition in which the qubits’ quantum states are dependent on each other.  

Amazon’s goal with the AWS Center for Quantum Computing on the Caltech campus, is to develop and build quantum computing technologies and deliver them onto the AWS cloud.  

Asked why quantum computing is so hard, Preskill stated, “What makes it so hard is we want our hardware to simultaneously satisfy a set of criteria that are nearly incompatible. On the one hand, we need to keep the qubits almost perfectly isolated from the outside world. But not really, because we want to control the computation. Eventually, we’ve got to measure the qubits, and we’ve got to be able to tell them what to do. We’re going to need some control circuitry that determines what actual algorithm we’re running.”  

Furthermore, “in the computation, we don’t want to look at the state until the very end, when we’re going to read it out. But even if we’re not looking at it ourselves, the environment is looking at it. If the environment is interacting with the quantum system that encodes the information that we’re processing, then there’s some leakage of information to the outside, and that means some disturbance of the quantum state that we’re trying to process.” 

He added, “The essence of the idea is that if you want to protect the quantum information, you have to store it in a very nonlocal way by means of what we call entanglement. Which is, of course, the origin of the quantum computer’s magic to begin with.” 

Asked what is the appeal of working on a project whose goal is to develop new technologies, Preskill stated, “This is the new frontier of the physical sciences, exploring these more and more complex systems of many particles interacting quantum mechanically, becoming highly entangled. Sometimes I call it the entanglement frontier. And I’m excited about what we can learn about physics by exploring that. I really think in AWS we are looking ahead to the big challenges.” 

 

Model To Teach Computer to Recognize Humor is Serious 

Another popular blog post on Amazon Science in 2020 involved teaching computers to recognize humor. In a paper presented (virtually) by lead author David Carmel, principal applied scientist at Amazon in Haifa, Israel, to this year’s Special Interest Group on Information Retrieval (SIGIR) conference of the ACM, described a new approach to humor detection when the system answers questions about products. 

A percentage of people try to be funny. “Our system leverages two insights from humor theory,” the authors stated, which were “incongruity” and a “subjective tone.” The team built a model, then tried to train it. Questions on associated product titles were extracted from an Amazon product page and passed through an “incongruity detection module,” which scored them. Then they passed through a “subjectivity module, which also scored them. Then they pass through a classifier the team built that tries to figure out if the question was an attempt at humor.   

The team went through a number of steps to try to produce data needed to train the model. “Past research has shown that using machine learning to train humor recognition models runs the risk of domain bias,” the authors stated. Allowing the model to recognize products that invite comic questions led to better results in initial testing.   

“Recognizing humor is a difficult AI challenge, but meeting it will ensure that the Amazon Store remains a place where customers can find useful product information quickly and have some fun while they’re at it,” the authors state. 

On the Amazon Science website, see the article on advances in text-to-speech technologies, the blog post on the challenges of quantum computing and the blog post on  teaching computers to recognize humor. 

Techatty Connecting the world of tech differently! Read, Learn, Thrive, and Make an informed decision without distractions. We are building tech media and publication networks to connect YOU and everyone to reliable information, opportunities, and resources to achieve greater success.
Web and Cloud LLC - talk to us and let's discuss your needs.
Let's help transform your business