Some new technology players want to plug our digital data into artificial intelligence to create web copies of ourselves. Will digital immortality soon be just a click away?
Sundar Pichai mounts the state to applause from the crowd. The Google CEO wears a khaki jacket and thin rectangular glasses, his movements simple and measured. As usual, he opts for a sober presentation. But what he’s preparing to unveil tonight, at 2018’s Google I/O, the company’s annual summit, is perfectly astounding. Sundar Pichai is here to demonstrate the abilities of Google Duplex, a virtual assistant designed to converse naturally with humans. For an astonished audience, the artificial intelligence calls a barber, then a restaurant to make a reservation. In addition to sounding like a human voice, the assistant measures pauses in its speech, affects the various tics of human conversation, and deftly manages disagreements and issues in the exchange. In other words, it’s impossible to tell it’s a robot and not a person on the other end of the line, and its conversation partners are obviously convinced they’re speaking with a flesh-and-blood person. The demonstration comes with loud applause, followed by a flood of articles in the media.
A software that with such fluent and natural oral conversation skills is a big first. In the scheme of things, though, it’s yet another step in the stunning recent progress in conversational artificial intelligences. Virtual assistants like Siri, Alexa and Ok Google can already handle simple requests made by users’ voices. When it comes to writing, the progress is even more spectacular. The French startup Julie Desk designed an impressive virtual assistant responsible for organizing email appointments. Other conversational agents are employed in customer service and marketing.
Excited by this technology’s potential, some are already imagining its next steps. That could mean a virtual assistant that learns to know us flawlessly, serving as representatives in the virtual world to help with basic chores, from managing our emails to making hotel reservations – a super-assistant capable of accompanying us throughout our lives and beyond. The potential is twofold: as ambassadors of our daily lives, these digital secretaries could become virtual memorials after we die, eternal and digital representations of ourselves that would give us a sort of online immortality.
That’s the goal of the online startup Augmented Eternity. The company takes the massive quantities of data we generate each day in our searches and social media shares to create digital copies of ourselves in the form of an artificial intelligence. This virtual double might take the form of a chatbot, an oral conversationalist àla Siri, or even a three-dimensional person inhabiting a virtual reality universe. It might seem like a shocking concept, but it already has its believers: Augmented Eternity is already working with the CEO of a major financial company to design a virtual copy that will be able to represent them and to advise the company after their death.
The project does evoke the great science-fiction writer Philip K. Dick’s novella What the Dead Men Say. In the story, a businessman’s conscience survives his body’s death. The man, the head of a powerful international conglomerate, remains entrenched in communication networks, communicating with old associates via radio and television. Augmented Eternity is not the first company to tread this ground. In Silicon Valley’s Mountain View, Eternime wants to gather the personal information of Internet users to construct intelligent avatars in their image. Those avatars would live forever online, a personal legacy for future generations.
Talking with the dead men
However, the idea of an artificial intelligence crafting a true copy of ourselves with only web-based data raises the eyebrows of experts. “I’m skeptical of the idea of replicating a personality with data,” says Anders Sandberg, researcher at the University of Oxford and the Future of Humanity Institute. “The brain has 90 billion neurons, each one with around 8,000 synapses, and each synapse has at least two states. That leaves us with, at the very least, 700 trillion degrees of freedom, which need to be replicated as data. Of course, in principle, we could produce enough terabytes of data to do that, but most of the videos and emails we make reveal just a tiny fraction of what happens inside us. For example, think of all the thoughts you’ve never shared with anyone.”
But even if we can’t create a perfect copy, it’s possible to create an approximate duplicate of our social selves. “I do think, however, that it’s possible to design a rough approximation of what we say and do, and probably better than most people would think. We’re not so creative in our daily lives. So I wouldn’t be surprised if we could design believable chatbots for most people. But they’d still be a long way from the profound and original self.” Still, an approximate copy could prove to be largely sufficient. Hossein Rahnama, creator of Augmented Eternity, says that the sole responsibility of the CEO’s avatar is to help with decision making, not to run the whole company from the web, as in Philip K. Dick’s imagined scenario.
For others, erecting a virtual memorial can help with the grieving process. When her best friend Roman Mazurenko died in a car accident in Moscow, Eugenia Kuyda, a young Russian entrepreneur in San Francisco, collected all the messages she’d shared with him over text message and social networks. She wanted to build a chatbot with her friend’s personality that she could talk to in order to lessen the weight of grief. Since then she has designed the chatbot Replika, which serves as a virtual friend for users and weaves their digital copy through their exchanges. Fenix, a Swedish funeral home, is exploring the idea of creating virtual doubles of deceased people that could talk to loved ones. Japanese roboticist Hiroshi Ishiguro, famous for his humanlike androids, likes the general idea. He defines two kinds of immortality: immortality of the conscience, and immortality of the social person. According to him, technology can provide social immortality, creating a space for interaction between the living and the dead. However, it remains impossible to recreate the conscience, and it will remain so, he says, forever. Ishiguro created an anthropomorphic copy of himself, which will survive him once he’s passed on.
Downloading your brain
The television series Altered Carbon depicts a futuristic universe in which every human being has a digital file, or a “cortical stack,” implanted in his or her vertebrae. The file contains a copy of the person’s conscience and memories, making it possible to effectively resuscitate them after their death. Hardcore transhumanists, not content with creating half-baked approximations of ourselves, share this idea. According to them, the role of artificial intelligence is to give man access to true digital immortality, an ambition they believe they’ll attain by allowing them to download their consciences to a computer. This somewhat confusing perspective would be made possible by a technique whereby a human brain would be scanned to create a precise and exhaustive map, then reproducing it in software. Although still a highly speculative idea, it nevertheless has several believers, including the well-known Ray Kurzweil, entrepreneur and media author, bitter defender of the theory of the Singularity, according to which artificial intelligence will eventually overtake the capacity of the human brain.
The scientist has collected an exhaustive quantity of data on his father (photographs, letters, disks, films, and even electricity bills), who died more than 50 years ago, in the hope of assembling a database with enough data to give him a second, virtual existence in 2029, the year he believes the technological Singularity will arrive. In his books, where he attempts to predict the future of new technologies (more or less successfully), he announces the imminent arrival of techniques making it possible to scan the human brain. The 2045 Initiative, launched by Russian billionaire Dmitry Itskov – a transhumanist – gathers scientists and technology experts to work on the technology in order to reach “cybernetic immortality.”
Are these crazy science-fiction fanatics mistaking their dreams for reality? Some of them truly believe in it, to the point of basing their business models on it. That’s the case for the startup Nectome. In March, the startup launched by MIT graduates and accepted by Y Combinator created media buzz by announcing they were working on a system that could scan the human brain. The only catch: the technique would be 100% lethal. It turns out it’s easier to map the brain when it’s come to a complete stop. After the media frenzy subsided, the startup opted for a less sensational marketing strategy, presenting itself now as a research institute focusing on memory and how to preserve it, in particular with brain-scanning techniques. The nonprofit CarbonCopies, founded by Dutch neuroscientist and former 2045 Initiative member Randal Koene, is also studying how to recreate the human brain artificially.
Reproducing the brain’s inner workings
For Anders Sandberg, who is on Nectome’s ethics committee, the idea is worth exploring. “Nectome’s approach consists of immobilizing the brain so that only a fragment of information is lost at the time of storage,” he says. “They’re also trying to prove that the most relevant memory fragments can be localized and stored. For me that’s an interesting idea, and potentially a useful one. It seems realistic to me in that the goal is to ensure effective preservation of the brain, without the pretense of being able to then recreate it in digital form. That part is for future scientists.”
According to him, the brain-scanning concept is unrealistic at present. However, in the long term, it’s not only possible, it would actually be more efficient to create a digital copy of ourselves than to simply collect data of ourselves from the web. “As I see it, the most obvious way to make a digital copy of someone would be to scan their brain (lethally, at least for now). That way you could read all the brain’s connections and chemical states in hopes of reconstructing its functional architecture. Obviously, that wouldn’t be easy, and it’s still very hypothetical right now, but as a long-term goal in computational neuroscience, I think it could be done.”
Not everyone shares Sandberg’s optimism. For Richard Jones, physics professor at the University of Sheffield, author of the nanotechnologies blog Soft Machines and the book Against Transhumanism, many transhumanists tend to underestimate the human brain’s complexity. “We often hear that neurons are the brain’s computing units, like transistors in computers,” he says. “So to reproduce the workings of the brain and its tens of billions of neurons, you’d need some tens of billions of transistors. But the brain’s computing units are not neurons, they’re molecules, which makes the brain even more complex. With this level of complexity, we still don’t even understand very well how the brain functions, which makes the process of replicating it artificially pretty complicated. Even the parts of the brain we understand best are still impossible to reproduce, given their complexity. The idea that we could capture our memories and thoughts and reproduce them on a computer seems extremely speculative to me.”
Whether his project works or not, Hossein Rahnama, CEO of Augmented Reality, hopes to stimulate public debate around ownership of our digital data. Basically, whether we can or can’t create virtual avatars using our online data, the data stays there well after our deaths. Who, then, does the data belong to? How could it be exploited? And who will be allowed to delete it? There are plenty of ethical and legal questions worth posing today.