Hayley Atwell as Martha in Black Mirror
James Vlahos can no longer sit across from his father, hold his hand or give him a hug. But he can ask him for advice when he’s feeling blue and let his children ask questions about his family’s life in Greece or listen to him sing “Me and My Shadow.”
When his father, John James Vlahos, was diagnosed with advanced lung cancer, at the age of 80, James began racing to record his life stories. For months, he sat across from his dad with an audio recorder, asking questions and recording long answers and jokes he’d heard “a hundred times.” In the end, he recorded 91,970 words.
What began as an oral-history project quickly evolved into a quest to give his dad virtual immortality. Vlahos, a journalist and author, had long been interested in Artificial Intelligence (AI) and voice applications. Shortly after his father’s cancer diagnosis, he learned the company PullString was releasing a software that would allow the public to create what are called conversational agents—like messenger chatbots. He wondered: What if he took the thousands of words and audio clips he’d collected and built a virtual replica—a Facebook Messenger chatbot—of his dad, with whom he could converse?
With his father’s and family’s blessings, Vlahos began building the Dadbot. His parents had never interacted with a chatbot before, but they liked the idea of being able to pass on his father’s stories in an interactive way.
“He had a way of sassing people in the friendliest kind of way, like if he thought you were getting a little too high on your horse, he had sayings like, ‘well, hot dribbling spit!’” Vlahos says. “I love hearing those words but I like when they come out at a surprising time. That is one thing I am proud of with the programming—that I got him.”
The Dadbot is designed to respond to questions driven by the people interacting with it, such as “What was your favourite class in college?” And every so often, the bot will decide to tell a story or joke.
Vlahos says building the Dadbot was part of his grieving process, and he still enjoys interacting with the bot.
“It’s comforting for me to talk to the Dadbot…It was neat to kind of step through his life, chapter by chapter, but then also to step into his mind, to a degree. How would he respond if somebody sort of teased him in this sort of situation?”
Digital avatars of the deceased are a rising trend, and go by different names, such as memorial bots or griefbots. When Vlahos detailed his experiences in a 2017 article for Wired magazine, he unintentionally positioned himself as a leader in this virtual immortality space.
“A lot of people wanted to talk about it,” he says. “One man, he was dying, and he wanted to create something for his kids to have, another woman had lost her son in an automobile accident and she wanted something to help remember him. It really resonated with people in a way that no project that I’ve been involved with before has.”
THE QUEST FOR IMMORTALITY IS nothing new but the ways we live and die have changed dramatically. In the 2018 version of the Vanier Institute report, Family Perspectives: Death and Dying in Canada, the first desire listed is “We want to live forever.” As the report points out, contemporary Canadian culture is death defying and death denying. We seek out anti-aging products and praise those who make it to the senior years with youthful lifestyles still blazing.
In Canada, sudden deaths are rare—the majority of Canadians get a heads-up that they, or their loved one, is dying. But in general, we are living longer than ever before. According to the 2016 census, the fastest-growing age group in Canada is actually “centenarians,” those aged 100 or older. The number of Canadians aged 85 and older now represents 2.2 percent of the total population (Japan, the country with the highest, is at just 4 percent).
Could this culmination of longer lives and deaths be making us think more about our personal legacies? Could it be driving our desire to achieve immortality in the most likely way we can—in the virtual world?
ANDREW LOUIS, A TORONTO computer scientist, has accumulated a lot of data. As a Millennial and tech entrepreneur, you’d expect his digital footprint to be large but it’s Louis’ personal archiving project, “Building a Memex,” that takes it to the next level.
For more than 15 years, Louis has programmed the devices in his life to track, record and store his digital footprint, resulting in a massive database that includes everything from what he’s typed and tweeted to the websites he’s browsed and what he’s eaten. He’s created a digital time capsule that can be searched, curated and, potentially, gifted to his children some day. In theory, someone could recreate a day in his life—or even recreate him.
“Pretty much every email or chat message I’ve ever sent is in there,” he says, joking that most of it “isn’t worth ever reading again.”
Last year, Louis decided to put his database to the test. He took the extensive MSN chat logs he’d saved as a teenager and resurrected his teenage self into a chatbot, using Recurrent Neural Networks (RNNs), a class of AI that creates vocabulary and speech recognition patterns. RNNs process all words in a text database, then develop a sequence or speech pattern based on those words.
This is no easy process. Chatbots require training to work and can require months of tweaking and training to create ever more complex and original sentences. The more times they run through a database of conversations, the more words can be stored for future retrieval. It’s also not a flawless process. As Louis points out, “All your typos and autocorrects will come back to haunt you.”
After 10 training sessions, a conversational structure emerged but it was only after 100 training sessions that his teenage replica started to sound more human and was able to understand and ask basic questions. Louis demonstrated the chatbot during a conference in New York City last year.
“Hello, are you there?” types the real Louis.
“Ya,” replies his teenage replica.
“Do you know how to use words?” asks the real Louis.
After more training sessions, the common “ya” responses became a series of “lol” answers, even when inappropriate. Louis explains that without extensive training, bots choose the safest and easiest routes.
“There’s no general, structured knowledge that a bot can rely on,” he added. “Stuff like short-term memory is really difficult so if I say ‘I’m feeling sad’ and then continue on with the conversation, there’s no way for the bot to remember and work it into a future chat.”
Louis had high hopes going into the experiment but came away feeling decades away from realistic, human-like interaction with bots.
“Even if we can get a chatbot that has really good conversations, I don’t think there’s enough history yet that a single person has typed into a computer to really make a replica of a person that sounds like them and has all the general knowledge,” he says. “My dataset would be a best-case scenario and it’s minuscule compared to the amount of data you would probably need to do this properly.”
He also points out that who we are and how we speak isn’t always best captured in our correspondence.
“If a bot is just being trained on things I’ve said, there’s a whole set of experiences that have never made their way into text messages,” he explains. “At some point, you’re going to be really disappointed or frustrated by something the bot doesn’t remember.”
SCI-FI MOVIES AND LITERATURE have long inspired real-life tech innovation. A 2013 episode of the Netflix series Black Mirror, titled “Be Right Back,” is often referenced in articles and debates about digital immortality. When character Martha loses her partner Ash in an accident, her friend signs her up for a service to turn his digital life into a bot. She quickly becomes dependent on virtual Ash, alienating herself from the real people in her life. But as she trains the bot version and upgrades her account to phone conversations with his voice, then a life-sized robot replica of him that lives alongside her, things become too real—and unrealistic. The more he fails to remember important details about their life or react in the exact way Ash would have, the more he disappoints her.
“You’re just a few ripples of you. There’s no history to you,” she tells the bot. “You’re just a performance of stuff that he performed without thinking, and it’s not enough.”
Several start-ups are experimenting with memorial bots. One of the more well-known companies, Eternime, touts itself as “an artificial intelligence platform that collects your thoughts, stories and memories and stores them forever into an intelligent digital avatar that looks and talks like you.” It’s still in private beta but the company says more than 40,000 users have signed up for access when it’s available. The current version tracks daily personal habits, social media interactions and even physical movements to create an automated biography and generate an AI avatar.
But with bots still far from feeling emotions, and replicating personalities, are we limited to building griefbots that are more reminiscent of bodies without souls?
For Vlahos, the Dadbot fulfilled his needs. He hadn’t hoped for a virtual copy of his dad, but rather a storytelling device to help serve up his voice and stories. But Vlahos had a lifetime of knowledge about his dad to draw from. Companies are only as good as the data their users contribute.
“Imagine being a company, and they have never met the person and they are trying to create something that captures the essence—that’s hard to do,” Vlahos says. “Despite what you might see in the movies, there’s no just ‘oh, we’ll dump all the emails and text messages into a computer program and the computer magically recreates the persona—that tech is not actually here yet. And even if it were, it would be a very distorted and incomplete replica of a person. Think of what you put in text messages—there’s a lot of ‘I’ll be home at six, can you pick up some milk?’”
Toronto grief counsellor Lysa Toye says it is important to keep in mind people present different facets of themselves in online environments, and some of them aren’t necessarily meant to be public.
“Depending on where information is culled from, it can be really problematic,” she says. “We have different identities that we share with different people in our lives, strategically sometimes. There are reasons why my romantic or sexual relationship is a different kind of relationship then the relationship I have with my kids, which is different from the ones I have with my friends or my parents. I don’t know how you navigate those different kinds of identities and self-states in one glommed-together version of a person.”
Dr. Hossein Rahnama envisions a future where we are not trying to replicate entire beings, but rather memorializing and “borrowing” parts of people’s identities—specifically, their knowledge. Rahnama is a professor at Ryerson University in Toronto and CEO of Flybits, a company that uses AI to create micro-personalized experiences.
His Augmented Eternity project is making waves in the digital identity space. He has spent the past few years researching what he calls “swappable identities” at the Massachusetts Institute of Technology (MIT). Rahnama and his team of computer scientists are working on ways to use digital tools to democratize and socialize philanthropy.
Rahnama thinks the future of digital immortality is in allowing people to preserve and gift specific parts of their identities, like an entire body of professional knowledge, in the form of a digital avatar—or some other digital representation—that can be activated and accessed by communities. Similar to how chatbots create vocabulary and learn to interact by drawing on massive databases of previous text conversations, Augmented Eternity could mine someone’s body of knowledge and then create patterns and predict what advice they would give. Instead of paying a human big bucks for their expertise, someone could activate their digital avatar for free or a low cost. This could help people in sectors where knowledge is pattern-based, like medicine or law, ensure their professional legacies outlive them.
“When we started this project about three years ago, the first use cases we were hearing from our focus groups were very much around ‘I want to stay in touch with my loved ones, I want to leverage their expertise and I want to remember them,’” he says. “As we continued, one thing we noticed was there’s a philanthropist in every single individual. Everyone wants to leave a legacy behind, but based on the physical world and what we have seen so far, philanthropy is very limited for certain groups of people, based on their wealth or their ability.”
Rahnama believes society is tracking toward a more decentralized data economy, where our digital lives are not all stored by Google or Facebook, but housed in decentralized systems like the blockchain, allowing people to maintain more control over their digital footprints and decide how it’s used—or what it becomes—in life and the afterlife.
“[Millennials] are generating gigabytes of data on a very regular basis,” he says. “We said, what if we can allow them to own that data and turn that into an expertise? Let’s say after about 50 or 60 years worth of developing insights and gathering knowledge, they can now pass their expertise to a loved one or family member, and be able to kind of make their digital identity more sentient.”
But if we’re closer to being able to borrow someone’s identity and could turn their correspondence into a service or product, it raises the question: Can your identity or brainpower not just be borrowed, but stolen if in the wrong hands? Who owns your digital footprint when you die?
“That question is a very new question, even for the legal system. If you look at the terms and conditions that you find with Facebook, it’s very scary. They basically own all of your data, like whatever footprint you leave there is owned by them,” Rahnama says.
Ann Cavoukian is one of the world’s leading privacy experts and the former Information and Privacy Commissioner of Ontario. Cavoukian says she isn’t aware of any concrete ethical or legal guidelines around who can own and use your data after you die, and she has “been in this area a long time.”
“There’s a hope that the family members, etc. will honour the wishes of the individual, assuming that they had any wishes declared to their family relating to their personal information, but it’s really soft.”
And, she adds, “How do you ensure that the control you want or your expectations are met?”
For now, it seems the virtual afterlife remains as mysterious as the physical one.