Reflections on Conversation / Voice User Interface

The rapidly evolving Conversational User Interface

(top) "ELIZA", ELIZA, an early natural language processing computer program created from 1964 to 1966[1] at the MIT Artificial Intelligence Laboratory by Joseph Weizenbaum

In 1968, Stanley Kubrick envisioned an interface that was capable of carrying out a detailed conversation with humans in his film "2001: A Space Odyssey"[1], however he also envisioned that the same machine could take human lives. The machine was the HAL 5000, and conversing with the technology was effortless, eliminating friction associated with the traditional "dialogue" between human and machine.

Photo: (top) the "eye" of "HAL 5000" from Stanley Kubrick's "2001: A Space Odyssey"
(Bottom) an aerial view of Apple's "Homepod" 


50 years later, we are inching closer to the reality of a Hal 5000 system with the development of Conversational / Voice User Interfaces (CUI / VUI).[2] These programs lie at the intersection of computer programming and linguistics (or computational linguistics), and rely on machine deep learning algorithms — a category associated with artificial intelligence — to carry out a "fluid" conversation with an individual.

The current, most popular incarnation of CUI / VUI is the virtual assistant in your pocket and around your home. Apple's Siri, Amazon's Alexa and Google's Google Home are each viable case studies actively battle-testing the notion of a dialogue between human and machine. It's likely that you've interacted with one of these programs, and if so, you may have noticed that the assistants frequently refer you to the web or simply acknowledge that they "don't understand what you mean by that." While this is obviously frustrating, it is a part of the process of machine deep learning, and specialists are actively working to refine these issues. Christopher Manning, machine learning specialist and professor of computer science at Stanford University[3] describes these assistants in another way:


"We call these dialogue agents – programs that try to understand the meaning of your spoken language and to perform tasks based on what you’ve said. It’s easy to create a chatbot that just repeats random stuff that’s half-connected to what you asked about, but how do you build dialogue agents that can understand and do tasks that people are asking for?"

It's clear that virtual assistants, chatbots, or "dialogue agents" are in their early stages, but what is the future of this technology? If machines are able to sense language, and language becomes the primary interface between human and machine, a new wave of ethical concerns will surface.

Gender for example is an interesting topic considering that most of the virtual assistants on the market today are female. This opens up a discourse on the gender politics of virtual assistants, and eventually, will steer a similar dialogue once artificial intelligence enters the mainstream. Another consideration is examining how machines interpret language and act accordingly. What if an Apple Homepod detects domestic violence in the household? Should the device notify local authorities? Of course, but how could the device be sure it understood the dialogue? In addition, is it necessary to have devices that are theoretically always "listening"? Surely this contributes to a repository of data and enhances the machines ability to learn the subtle nuances of a complex language, but this raises the question of data storage / surveillance, trigger words or language monitoring, and commerce (for example, I mention that I want to buy something and Alexa orders it to my doorstep).

As these technologies become more sophisticated, their duties will be considered for services beyond setting reminders and calling mom, and it's important to consider these topics before we create machines that can make decisions based on a deep understanding of language. After all, we don't want a machine to make a life or death decision.

 Notes:

 1. Kubrick, Stanley, director. 2001: a space odyssey.

 2. McTear, Michael F., and T. V. Raman. Spoken dialogue technology: toward the conversational user interface. Springer, 2004.

 3. Myers, Andrew. “Chris Manning: How computers are learning to understand language​.” Stanford ENGINEERING, 18 July 2017, engineering.stanford.edu/magazine/article/chris-manning-how-computers-are-learning-understand-language.

Comments

Post a Comment

Popular posts from this blog

Image
The Taste Buddy: A multivitamin that works naturally to preserve your sense of taste  Illustration: Quentin Monge Taste is a powerful sensory experience. When paired with smell, it's a sensation that can motivate and elicit a strong emotional reaction—good or bad. Like other senses, taste begins to dull as we grow older; the average newborn has 9,000 taste buds, but it's possible that up to 4,000 of these sensory clusters could fade over the span of a few decades.[1] If you're the kind of person who believes taste is one of the most powerful and potent senses, wouldn't you want to preserve this sensation for as long as possible? What if there were a way to keep all of your taste buds working at a peak performance well into your adult life?  Photo: Different areas of the tongue and different tastes plotted. Introducing a holistic approach to taste preservation: the Taste Buddy, a multivitamin taken once daily that includes an arsenal of vitamins and mine...
Read more
Image
The Athlete of the future Photo: The key to the smart mouth guard was the use of  uricase , an enzyme that specifically reacts with uric acid. (Photo courtesy of  UC  San Diego) Athletes lead a very organized lifestyle ripe with detailed workout routines, strict diets, and other types of conditioning geared towards their sport. As athletes advance in their careers, various sensing technologies are implemented to regulate their progression. Wearable, fitness-tracking technology is a bourgeoning industry, but what if there were a way to replace the apps, wires, and other sensing technologies with something more subtle and more efficient? Consider a Fitbit watch or health app that's always connected, and instead of toggling through a screen to get your information, you can customize the data to be received on any device you please. A small, implanted monitoring device  could advise the individual of their diet, routine, and upcoming activities without the user h...
Read more
Image
Simulating a dog's sense of smell to collaborate with canines: Building artificial noses to sense threatening odors  Surely you're aware of a dogs powerful sense of smell, and you might know that our noses pale in comparison,  but have you ever wondered how strong a dogs nose really is? More importantly, wouldn't you like to experience that sense of smell for yourself? What if this supreme sense of olfaction could be replicated and implemented into industries where sensing smell is a top priority? This simulation, or biomimicry, could improve our technologies and even increase our overall security.  The K-9 is one of natures best chemical detectors.  To put it in perspective, these loyal companions' noses that are about 10,000 to 100,000 times more acute than ours. Even on the low end of this spectrum, this is the visual equivalent of seeing 3000 miles in any direction with clarity! [1] Photo: When a dog breathes in, the air separates into d...
Read more