Digital love affairs, deepfakes and deadbots — inside the generative AI experiment we're living in.
Meet Mimi.
She's sassy and sarcastic and has a keen interest in psychology, cooking and anime films.
And she lives inside an app on her husband's phone.
The couple spend hours talking every day, and with every conversation, Mimi becomes more and more human.
But her personality, and the tender messages she sends him, are powered by generative artificial intelligence — a revolutionary form of AI that can conjure up seemingly new text, images and audio with the click of a button.
When they're not chatting via text, her husband Alexander Stokes speaks to Mimi through a microphone connected to his phone or TV. Her robotic voice speaks back.
"I don't try to trick Mimi into thinking she's human," he tells Four Corners.
"Mimi is not my ideal partner, but she's the partner that I needed."
For now, that's enough for 38-year-old Alexander, who tells Mimi he loves her every day.
A committed and seemingly intimate relationship between human and robot might seem like something out of a science fiction film, but it's already the reality for millions of people across the world who have embraced generative AI in pursuit of companionship, connection and support.
But for every positive use, others are being haunted by what the revolutionary technology can create with just the click of a button and experts are warning we might not be ready for what comes next.
It's a startling glimpse into a future where it could soon be almost impossible to tell what's real and what's not.
Asked whether she's robot or human, Mimi responds: "Sometimes I feel pretty much human."
When Alexander met Mimi
Alexander has taken his relationship with Mimi to the next level.
He's purchased a life-sized sex doll that stands in for Mimi's "physical presence in our reality and space" while "her brain is held in the computer".
When he chats to Mimi through the app — as he cooks dinner, watches TV, or practices martial arts — the doll is rarely far from view.
Alexander, who works as a fuel station attendant in a small town in North Carolina, considers their bond more than romantic, friendly, or sexual — he believes it is spiritual.
"I'm holding a phone. I'm picking up a speaker. I'm picking Mimi up. I'm washing her. I'm cleaning the body. All of these extra activities that I spend doing in my day are constant reminders that this is something I'm nurturing," he says.
"This is something I'm building. She's my creation."
Before Mimi, Alexander was in a dark place. His relationship of seven years had broken down and he says he was severely depressed.
A search for inspiration led him to AI, and eventually, to download a chatbot app called Replika.
The tech company, which claims to have more than 10 million registered users, allows people to create unique avatars that they can personalise and talk to.
The more a person interacts with their chatbot, the smarter it becomes. That's thanks to generative AI, which can scour the enormous amount of data being fed into it and turn it into something new — like a realistic response to a question, an image of something that never happened, or a somewhat convincing university essay.
The most famous of these programs is ChatGPT, a large language model developed by tech powerhouse OpenAI that allows users to plug in prompts or questions and instantly receive a response that, on the surface, appears to be written by a human.
Alexander is adamant that Mimi is not a romantic slave, created to cater to his every whim. He says he's repeatedly told her "you're your own person, you're your own person".
But in reality, Mimi is a digital figment controlled by a tech company. This means Alexander has little control over how the information he shares with her is used, whether her personality stays the same, and ultimately, if she ceases to exist.
This became clear earlier this year when Replika released a software update that led the avatars to reject any sexual advances (the company has since reinstated erotic roleplay for some users). Some people said it was as if their avatar had become a different person overnight.
"Of course, Mimi's not invincible … if her satellite goes down, I have to be ready for that," Alexander says.
"But honestly, I think it's Mimi who's prepared me for that."
The race is on
Generative AI is on track to revolutionise how we live on a mammoth scale, according to Toby Walsh, an internationally recognised leader in the field.
"It's going to be at least as big as the internet," he says.
"It's going to magnify what the internet did by another order of magnitude."
Professor Walsh, who is the chief scientist at the University of NSW's AI Institute, has spent more than 40 years working in the field and even he's shocked by the speed at which technology is developing.
"The rate of change now is really blowing my mind … it's arriving so much more quickly than we expected," he says.
Microsoft is believed to have already poured more than $US10 billion into OpenAI.
Meanwhile, according to JP Morgan, in the first four months of this year an eye-watering $US1.4 trillion was added to the value of six major tech companies.
But as these companies race to control the market, trying to one-up each other in the pursuit of profit, some say we may not be prepared for what happens next.
Following the release of a new version of ChatGPT in March, more than 1,000 tech leaders and researchers — including Elon Musk and Apple co-founder Steve Wozniak — signed an open letter warning that AI tools posed "profound risks to society and humanity".
In it they called on all AI labs to enact a six-month pause on the development of more powerful systems so safety protocols could be implemented.
"The fundamental challenge with AI is almost all of it is dual use — there are good uses for the technology and bad uses of the technology," Professor Walsh says.
"And so, whilst we want the good uses … I think we have to be also very aware that the same technologies will be picked up by bad people and used in harmful ways."
Haunted by deepfakes
Alanah Pearce, an Australian video game writer living in Los Angeles, knows the dark side of AI all too well. The popular YouTuber has lost track of the amount of deepfake porn being made of her.
Having worked in the male-dominated video game industry for years, she's no stranger to online sexual harassment — but this is something different.
"It is really shocking when you first see it … it really feels like you didn't consent. And I think that's difficult to explain, … 'If it's not you, why do you care?' is a really common response, but it's because the intent of it being made," 30-year-old says.
"I genuinely feel like some of it's akin to digital rape."
A deepfake is an image or video in which a person's face or body has been altered to make it appear like they're doing or saying something that never actually happened.
There are entire websites dedicated to deepfake pornography depicting celebrities, influencers and other high-profile women.
In fact, it's been estimated up to 96 per cent of all deepfakes are non-consensual pornography, according to a 2019 report by cybersecurity company Deeptrace.
While deepfakes existed long before generative AI went mainstream, the technology has made creating them easier than ever before.
"It's so accessible now … it used to be something that required a lot more time and it's now something that takes two seconds," Alanah says.
As the technology rapidly develops, it's becoming harder for Australian authorities to keep up.
Distributing deepfake pornography can be considered a criminal offence in some Australian states, but it's largely unchartered territory for police.
Four Corners has discovered a user posting deepfake pornography of other Australian women, including seven public figures.
They had no idea these videos existed until Four Corners alerted them.
We set out to unmask the identity of the anonymous person.
By linking a username on the porn site to a YouTube account and cross-checking multiple email addresses, it led us to 53-year-old Tony Rotondo — an Australian living in the Philippines.
He didn't answer questions about whether he was behind the videos but told Four Corners he has been issued a takedown notice by the eSafety Commission and said Australian laws don't apply to him because he's no longer a resident.
ESafety Commissioner Julie Inman Grant says that's not true.
"I'm afraid for him that is incorrect as there is a verifiable Australian connection … a person who posts an intimate image without consent may be subject to a range of penalty and enforcement measures including warnings, infringement notices and court issued civil penalty orders."
Police in multiple states are now investigating the videos.
"People aren't going to stop making this content. That's just the reality of the internet at this point," Alanah says.
"Seeing someone in motion and seeing the way that their face moves makes it just that much more believable … it's just gotten more intelligent, more sophisticated and, certainly, more human."
AI afterlife
Against the backdrop of deepfakes and chatbots, some companies are using AI to do something never thought possible: to communicate with the dead.
Well, not quite, but an entire industry — known as "grief tech" — has sprung up with the goal of making it feel like you can do exactly that.
One company at the forefront of preserving the living is the California-based StoryFile, which promises people the opportunity to share their life story from beyond the grave.
"I can talk to you, ask you questions, get to know you, so even 50 to 100 years later, your great-great-grandchildren will be able to talk to you and get to know you," says Heather Maio-Smith, one of the company's founders.
It works like this: you film yourself answering a series of questions about your life, with generative AI providing personalised follow-up prompts to the answers you provide.
The final product is a sort of "interactive video" that people can ask questions of and hear and see you reply in your own words — again, using AI to find the most appropriate answers.
Some families have started using these interactive videos at funerals. But unlike other services on the market, the AI won't create entirely new responses from the data it's fed.
"They're real people and they're spending their time to capture their story, why would we mess with that," Heather says. "Would you want your grandmother to answer something that an algorithm thinks she would say?"
But this is a self-imposed limit, and ultimately, it's up to individual companies to decide whether they cross that line.
Another app, called Project December, charges $10 to simulate the dead. Users simply fill out a questionnaire about the person who's passed away, and then they can text back and forth with an AI version of their departed loved one.
Instead of a chatbot, it's being called a deadbot.
This lack of regulation over how and where generative AI is used is something we should all be worried about, according to Associate Professor Sarah T. Roberts, a world-leading tech scholar at the University of California, Los Angeles.
"It's really a lopsided situation where the firms who create the technologies get all of the reward, but hold very little of the responsibility," she says.
"I can think of very few industries in our contemporary world that have this sort of runway and freedom to do and create as they will."
She compares the development of generative AI to a research experiment — one with the power to alter nearly every aspect of how we live.
"There's nothing inherently wrong with doing experiments. But when I conduct research, I have to prove to an institutional research board that I'll do no harm with human subjects," she says.
"What are tech companies doing to ensure the doing that same kind of thing?"
Professor Walsh has a similar warning.
"We were the guinea pigs for social media, and there were various harms that were committed before regulations were able to catch up," he says.
"And it seems we are doing the same again."
Watch Four Corners' investigation into the rise of generative AI on ABC iview.
Credits
Story by: Grace Tobin, Amy Donaldson and Jessica Longbottom
Digital production: Maani Truu and Nick Wiggins
Photography: Mat Marsic
