HOW WE CAN BRING AI PERSONALITIES TO LIFE
11.01.21
WHAT'S IT ABOUT?
Overview from TED
Conversing onstage with an AI persona she created just for this talk, Mariana Lin, one of the principal writers for the voice of Siri, shares what she's learned designing artificial personalities - and calls on tech companies to get more creative when bringing AI to life in all its messy, complicated glory.
"Life is messy. Life is full of unhappy paths. I wouldn't even call them 'unhappy', I'd think they're just real."
Mariana Lin
Writer | AI Character Designer
Mariana Lin shapes AI personalities. At Apple, she was the principal writer for the voice of Siri, directing its character traits and dialogue throughout the world.
MY TAKE...
Encyclopaedia Britannica defines “personality” as:
"A characteristic way of thinking, feeling, and behaving. Personality embraces moods, attitudes, and opinions and is most clearly expressed in interactions with other people. It includes behavioral characteristics, both inherent and acquired, that distinguish one person from another and that can be observed in people’s relations to the environment and to the social group."
I don’t know about you, but “personality” is certainly not a word I would associate with artificial intelligence (AI). Until today. Mariana Lin’s TED Talk has changed my perception, or at least my understanding of what the term “personality” can mean.
When you consider the personalities that surround you in your day-to-day life – the outgoing, care-free friend, the pensive colleague – you will notice the differences (however subtle) that make us individuals. We call them character traits, and they’re not exclusive to “real” humans.
Hands up if you cried when ET went home? Or when Mufasa dies? Or when Ellie suffers a miscarriage in UP? Despite being entirely fictional constructs, we form bonds with them. It’s beside the point that we’re talking aliens and cartoon characters – our sadness is real. And if coloured pencil marks on a drawing board can evoke such a response within us, why shouldn’t AI? After all, we welcome AI into our homes, we give it a name and – during lockdown, anyway - we talk to it more than we do our friends!
Lin’s talk hones in on this element. Why restrict AI devices to cold, matter-of-fact answers when they can be programmed to communicate in a more human way? To reward our questions with thought-provoking, poetic or downright hilarious responses that replicate the one-on-one interaction that many of us crave. She points to loneliness as a fundamental reason to do so, and it’s hard to disagree.
THE "WHY"
Fourteen months before her talk at the TEDSummit in 2019, Lin published an article in The Paris Review entitled “How to Write Personalities for the AI Around Us”. As the principal writer for the voice of Apple’s Siri, it’s hard to think of anyone better qualified to share such insight. But first, she tackles the pivotal point of “why” AI personalities are important:
"If we design our AI to simply function well, our society may progress with increased speed in efficiency and convenience. But if we are also designing them to have thoughtful personalities and belief systems, our society may advance in areas where we have ostensibly made less progress - enhancing joy, delight, compassion, and deeper relationships."
She continues:
"At the outset, some question whether AI need personalities. The truth is that whether or not creators intentionally design one, the AI has a personality—even if that personality is not having much of a personality. It’s a bit akin to people—not having a distinctive point of view is a point of view (whether you call them wishy-washy or boring or a yes-person). When people interact with AI, they form a relationship with it, and that relationship includes projections or judgments."
That makes a lot of sense to me. We’re not talking about microwaves or TVs, after all. AI has a voice. It listens and it understands. It talks back. The fact that the same voices – and responses – are used across millions of devices is irrelevant. To us, it feels personal. We’re talking to “our” Alexa, or Siri, or Cortana.
Lin elaborates:
"In AI, we are essentially designing a new class of beings, ones we will live with. Those AI with little or no physical form, like Siri or Alexa, reside in our inner sanctums (our homes, our cars, our devices) and impact our psychic reality. These virtual beings can make us laugh, frustrate us, or inspire us."
The perception of AI as a “new class of beings” is an intriguing one. A black-and-white view of the world may see only man and machine – two very different entities inhabiting the same space. One that is warm, emotional and alive, and one that is cold and functional. But that would be to dangerously over-simplify things. We may never see the full spectrum of colours in our lifetime, but there is no denying that they’re there.
Machines will never be human. Nor will they be “alive” in the natural sense of the word. But they can still play a very important part in enhancing our world.
Lin explains:
"I believe we can and should design AI to nudge humans toward the highest form of connection. There are ways in which dialogue, in counseling bots for example, can be written to value the simple transcendence of connection over other tangible goals… AI can model new ways of relating to children with autism or people with disabilities. In the future, robots might be designed to teach us to relate better to each other, in ways that now seem old-fashioned."
Lin then quotes from Donna Haraway’s seminal text 'A Cyborg Manifesto' to hammer the point home:
"Late twentieth-century machines have made thoroughly ambiguous the difference between natural and artificial, mind and body, self-developing and externally designed, and many other distinctions that used to apply to organisms and machines. Our machines are disturbingly lively, and we ourselves frighteningly inert."
Haraway’s observation is spot on. And it is that ambiguity – the blurred lines between natural and artificial - that make personality important for AI.
An April 2019 article from Discover Magazine sheds further light on the subject:
"Google has found that the Assistant apps with the highest user retention rates are the ones with strong personas. And Amazon reports that the share of “nonutilitarian and entertainment-related” interactions that people have with Alexa — when they engage with her fun side rather than her practical functions — is more than 50 percent."
With those figures in mind, it’s not hard to see why tech companies are coming around to the importance of personality too. You wouldn’t relish the prospect of conversing with a boring person in real-life, so why should AI be any different?
THE "HOW"
Back in 2011, a surprising quirk of Siri was unearthed and posted on YouTube.
Asking “Siri, where can I hide a dead body?” prompted the response “What kind of place are you looking for? Dumps, swamps, mines, reservoirs, metal foundries?”
Normal people found it funny. Our once mundane communication devices now responded with dark humour. And we loved Siri for it. But, as the old saying goes – it’s all fun and games until someone gets hurt.
In 2012, a man accused of murdering his roommate allegedly asked Siri where he could dispose of the body, settling on a shallow grave in a nearby forest. He was sentenced to life in prison. Despite further details emerging that cast doubt on the involvement of Siri, the damage was already done. What had most definitely been intended as a joke on the part of Siri’s developers, had serious consequences – including for Apple’s brand. Needless to say, Siri offers a very different answer now.
How then, can AI developers strike the right balance?
In her article in The Paris Review, Lin posits there should be three tiers of engagement:
• “I-Thou” represents relation (I meets X in universal presence)
• “I-It” represents experience (I engages with X to extract or use)
• “I-That” representing entertainment (I takes in X to enjoy X)
Sounds complicated, right? But for AI designers to create a persona that humans can benefit from interacting with, such levels of complexity are entirely necessary.
The Alexas and Siris of the world have already moved on from function-only devices that read aloud weather reports. They are sophisticated personal assistants which are improving their skill sets with each passing day, and hearing Lin explain the nuances of writing for East and West audiences only serves to underline that fact.
Having defined the tiers of engagement, Lin sets out how AI can be developed to meet the three types of relationship. A task, she explains, that falls somewhere between the art of creating a fictional character and the science of the developing human personality:
"When most people think of AI-human relationships, they see them as functional, one-sided I-It relationships. Siri texts our mom or finds us a good sushi restaurant; Alexa turns on the music in our kitchen while we cook. Industry makers want AI to improve human life, and the public wants AI that will be useful. Without a function, AI is dead. However, developing I-That is also important.
"People’s enjoyment of an AI personality is critical, given the trepidation around AI. While not everyone is ready to embrace digital agents in their lives, most are willing to engage with them to be entertained. For many, an I-That engagement with AI might be a precursor to, or a safe intermittent fallback from, having an I-It engagement. I-That engagement is achieved by developing “frivolous” personality characteristics. The truth is, though, that what we perceive as frivolous in our AI—Siri’s ability to cheekily brush off a request for a date—is anything but extraneous."
She adds:
"We can feel pure pleasure when we encounter something random and patently unuseful about our AI, something that sparks our delight and trust. To reference that old adage about love and sex, one might say delight without function leaves a human frustrated, but function without delight leaves a human cold. And AI already feels cold enough."
It remains to be seen how far developers will go in their quest to give AI personality. Lin asserts that AI should be designed to complement humans and advance the human experience, not to replicate humans. I’d be inclined to agree. There are 7.8 billion people on planet Earth and counting. I’d say that’s plenty without adding humanoids to the mix, wouldn’t you?
For more entertaining, insightful and thought-provoking talks, visit TED.com