Why Design a Bot with Personality?
Because no one wants to speak with a robotic robot.
Personality is the user experience of artificial intelligence
When it comes to bot design, personality is the crucial factor that allows us to interface and recognize shared values. Personality helps users understand the value of the system and how to interact with it. Right now, Alexa comes with more than 15,000 built in skills. And yet, in a recent survey we conducted for Logitech, 75% the people we talked to only asked it to play music. That’s due to a lack of personality (and perhaps because it looks like a speaker). Designing a personality promotes engagement and deepens the functionality of the service. For the end user, it extends the usefulness of the system and its usability.
A personality identifies the context of conversation we share with one another along with a shared set of values, beginning with empathy. So, how do you define empathy? For our purposes I define it as “affection symbolically provided to another person without an agenda.”
So how do you define empathy? For our purposes I define it as “affection symbolically provided to another person without an agenda.”
If that sounds like a recipe for building a bot with a “helpful” personality, that’s a good place to start. Our goal is to create a nuanced, designed, and governed bot that accommodates the end-user and functions on a genetic level of understanding. We’ll know we’ve achieved our objective when an end-user actively engages with a bot for more than rounds of question and answer. We’re getting there. For example, performance artist Stelarc created a bot so engrossing that museumgoers spoke with it for up to 20 minutes at a time.
We’ve already seen an increased predisposition to engage with bots in studies coming out of Skip Rizzo’s Institute of Creative Studies at the University of Southern California. In testing an AI healthcare app, researchers discovered 85% of the people speaking to a bot are more likely to share more information and do it more accurately than when speaking with a real doctor or nurse. Not only that, they are more compliant when it comes to taking their medication on time. It’s not only a concept, it’s already being used commercially, for example, in the healthcare company Sensely and its virtual assistant, Molly.
You have an honest face: designing a bot with a recognizable visual appearance
If you want to establish trust, design a face for your bot. The visual appearance of a bot not only demystifies and personalizes the AI, it also provides a visual component of the personality.
Cultural conditioning predisposes us to trust a human face more than a faceless bot. But to establish trust, that face has to be designed just right. Too lifelike and you introduce the concept of the “the uncanny valley,” in which a too-human representation generates feelings of revulsion. To prevent that, we make our bots imperfect. Design a cartoon-like bot with just enough imperfections and it says to the viewer, “This is not a person. It’s a representation of software.” By exposing the software we establish our honesty and our intentions, and that builds trust.
The face is familiar
On Skype, facial recognition goes two ways. Cameras enable us to see and be seen. Facial recognition software compares end-user’s expressions in real time against a categorized library of human emotions. This technique is called “multimodal affect detection” and it helps bots navigate the psychological and emotional landscape of another person. Botanic filed a patent on it.
The basics of building a bot
Before we begin building a bot, we start by finding out what’s most appropriate — by asking end-users what they want. Then, we look at the persona, the context, the need, and the functional aspect of the bot. Between form and function, we believe form is the function, when the function is social interaction. In other words, in terms of fostering social engagement, the function of the face is the form. As soon as you put a face on a robot its function becomes social.
Dialogue, interaction, and even affection can be modeled. We look at what phrases are most likely to be used by conducting probabilistic inferences of a database of recorded conversations, to determine the frequency of words and phrases.
Multimodal capability including voice, video, facial recognition, and text are key to collecting information and storing phonemes and visemes (the smallest parts of speech and visual cues that can be recognized and categorized). They’re also critical to understanding the emotional state of our end-users. Think of the sound of a syllable, or the wink of an eye.
The range of human expression is surprisingly limited. For example, in English-speaking cultures, there are just 36 different animated facial expressions required to communicate the gamut of human expression. That’s nine basic facial movements that map to the range of vocal expressions in the mouth. Chinese requires cataloging 32 different expressions. A global perspective can design for cultural biases and build a richer set of reactions. Considerations include how do we remove sexist bias in language, as well as the nuances of transitioning from one emotion to another.
From these individual notes developers can build “chords” of facial expression into a bot that can respond with a standardized repertoire of expressions and emotions. Then, we combine facial detection with voice recognition algorithms against a library of cataloged facial expressions. The overlap of those two domains equals affect detection, which gives our bot the ability to detect and respond to emotions.
Putting your best face forward
You want to make a bot that’s a reflection of yourself on your best day. Our guiding design paradigm is to build a foundation for the bot’s interaction that defaults toward a stance that provides an experience that is pleasant from the end-user’s perspective. In other words, we aim to design a bot that is programmed to meet you more than half way.
Without personality we become robots. And who wants to talk to a bot that’s brittle, chalky, and formalistic? Not us.
Ten years from now, people will have their own personal bot, complete with an avatar. The same way you have a Facebook page, a LinkedIn page, or your own website that reflects your values, your beliefs, as well as your likes and dislikes, your bot will be a reflection of your own personality. Better get to work on that.
Mark Stephen Meadows is the Founder & CEO of Botanic.io, Co-founder and Trustee of seedtoken.io (and Author, Inventor, Illustrator, Sailor).