Getting to know you: Designing trustworthy artificial personalities

From messaging apps to voice assistants, chatbots are everywhere. Brands are calling on designers to help them develop bot personalities that gain customer trust instead of losing it. We found that bots help people create better relationships with brands when the bot is task-oriented but not transactional, and when the dialogue feels genuine.

As we’ve shifted more of our interactions online, we’ve gotten used to having simple conversations with computers. Chatbots are no longer there just to get simple tasks done quickly. They have personalities, developed by designers and writers, and are trained to get to know you.

People are starting to like them, too. In customer service, nearly half of people globally prefer to chat online — whether it’s a bot or not — than to pick up the phone. Chatbots are also cost-effective. In banking and healthcare, a chatbot enquiry shaves four minutes off a phone interaction and saves about $0.70. From nurse avatars to financial services bots, automation is coming. Here’s a reality check for the naysayers:

• Business spending on machine intelligence is forecast to reach $31.3 billion by 2019.
• 80 percent of companies said they already used chatbots or will implement them by 2020.
• 85 percent of customer interactions will be managed by machines by 2020.
• The chatbot market will reach $1.63 billion by 2025.
• By 2025, 12.7 million new jobs will involve building robots or automation software.
• AI applications could save up to $150 billion annually for the U.S health care economy by 2026.
• Facebook, Google, Microsoft, and Telegram are the big players (translation: bots are here to stay).

So what do designers need to understand about designing the personalities of bots? Experience has shown us that bots can gain or lose customer trust during conversations, so it’s critical to design trust into every exchange: Satisfy the customer’s need but also anticipate all the ways that a conversation can go belly-up.

Storytelling is key to a successful conversational interface.

When new interfaces emerge, we need new rules so we can adapt our ideas. The rules we follow for visual design do not apply to conversational interfaces. The rules that govern the art (and science) of chatbot personality design are the rules of storytelling.

Screenwriters and fiction writers understand that characters drive the narrative. What is a character like? You can start with a list or a series of mood boards, and sketch out aspects of the character, but you also have to go further and imagine how different scenarios may play out. Then you adjust for context and complexity. “It’s always important to adjust for context,” says Designit’s senior user experience designer Stine Skaarup. “Humans behave differently depending on if we are on a date, with friends, or in a job interview. To gain our trust, AI personalities should be able to know context, as well.” After all, conversations are nuanced, and each conversation tells a different story.

Remember that a chatbot represents your brand — so when you design a chatbot’s personality and back story, you reflect the brand’s personality and back story. For example, a beauty brand’s bot should reflect the brand ethos, just as a financial services bot should convey reliability and security. Just like humans, chatbots engage in complex ways with people, so their mood and attitude should vary depending on the unique exchange that they’re having with a person.

Say your please and thank-yous.

In our experience with designing chatbots, we found that the quality of the interaction and the tone can make or break a customer relationship. Bots have a chance to create a bond in every interaction — by responding in ways that are useful rather than in ways that make you crazy. How many times have you encountered a robotic-sounding call center bot that drove you up the wall or made you feel dehumanized? And how many times have you encountered an online customer service bot that, in the course of trying to solve your problem, made you so mad that you took your business elsewhere? The interaction should always be easygoing and the tone should reflect the brand but not overdo it — the bot should be empathic but not sappy, polite but not formal, friendly but not over-familiar, and knowledgeable, but not stuffy or stern.

Companies are working hard to figure all of this out. Writers, traditionally tasked with developing brand stories and identities for brands, are now being asked to do the same for computer bots. Tech companies are putting creatives on staff to help design conversations and personas for chatbots. In the same way a screenwriter or novelist develops a storyline and fills it with fictional characters, writers — “AI conversation designers” or “interactive dialogue writers” — create brand narratives and fill them with fictional chatbots. Dialogue is carefully managed, even for bots doing simple tasks, such as scheduling meetings via email. Even with limited options for dialogue, you still need to nail it.

Designing a personality is complicated.

Like humans, chatbots can have innumerable personalities. You can mix and match attributes as you build nuanced personalities geared to different mindsets and demographic characteristics — but one thing in common to all chatbots is that they’re agreeable.

But there’s a whole host of other issues to think about. In an effort to support full transparency with customers, why not make it clear that a bot is a bot? Wouldn’t that honesty help gain their trust? Another reason to make it clear it’s a bot is the difficulty of trying to help someone who is emotional. Human personalities run the gamut, from gregarious to neurotic to hostile. Should the bot ignore the user’s state of mind, or try to balance it? We believe that if the bot is not able to immediately defuse the situation, the bot should automatically transfer the conversation to a human.

You can’t write lines 2, 4, 6, and 8.

Chatbots have created a new paradigm: a two-way street. Former Pixar developer Oren Jacob offers a useful approach to this: “One can think of computer conversation kind of like interactive screenwriting. We are writing lines 1, 3, 5, and 7, and then, oddly, we have no control whatsoever over what comes back in lines 2, 4, 6, and 8.”

Remember that you are designing for two people, but you only have control over 50 percent of the conversation. Plan for uncertainty! Your bot better be equipped for it. Think of it this way: You’re not figuring out what’s missing as much as testing for scenarios in which the bot is under pressure and has no easy answer.

Humans know when to say: I don’t know. What about your bot?

Oren Jacob, offering lessons for bots from Hollywood, says the decisions that chatbots make under pressure are critical to understanding them as characters. For example, how does the bot defer a question? “Does the bot say: I don’t know, or I’m checking Wikipedia or does it try to give an answer anyway?”

The choices that your bot makes, which are outside the scope of whatever actions you’re designing for, are the most important part of communicating the tone, mood, and style of the bot you’re building, says Jacob. Those decisions, made in that uncertain moment, also affect how users will feel about the bot. (And however the user feels about the bot, they will feel about the brand.)

So you can design a personality for your chatbot, and you can write all the dialogue that you think may happen, but you also need to test for the unexpected. It is during those moments — as with humans — that you really see what’s at stake. So put your bot under pressure and see what decisions it makes.

People remember conversations.

The bread and butter of being human is interacting with other humans. Everyone in a conversation has a distinct personality and built-in biases, which shape the nature of the conversation and the dialogue. It creates serendipity in the world!

That said, we also believe that given how quickly bots have proliferated our homes, our computers, and our phones, we should try to model bot personalities responsibly. So when you’re designing a human-computer interaction, we urge designers to let the bot be a bot, and say so, but also to add a little humanity to those human-machine conversations. We prefer to design bots to imitate human behavior rather than design bots that practice technology behavior — because bots by nature are designed to serve humans. It begs the question about who is more human, the bot or the human! But that’s a question for philosophers. (Or check back in with us in 10 years.)