Look closely at Mythe, the latest member of the team at Sanctuary Cognitive Systems Corp., and you’ll notice something uncanny about her.
Her head bobs and shifts subtly and mechanically as her luminescent green eyes slowly scan the room, taking in the surroundings of Sanctuary’s office, near Science World in Vancouver. Every so often, she blinks in a way that is unsettlingly human, save for the “cha-ching” sound her eyelids make.
Dressed in a conservative white blouse, grey jacket and black skirt, Mythe looks like an ultrarealistic fashion model. Her movements are limited to a few above-the-shoulders gestures, but then, she’s just an early, “zero-point-zero” version of what her creators at this pioneering tech startup intend to develop: humanoid robots that can move, speak and think for themselves and interact – as intellectual peers – with real people on a daily basis in intimate and vital roles as therapists, caregivers, teachers, scientists, even lovers.
If that sounds like a mind-boggling moonshot, Sanctuary co-founder Geordie Rose is the first to admit it. He’s on a mission to unlock how human intelligence works and to replicate it on a mass scale – a goal, he says, that would unleash “the most valuable thing ever created.”
“What we’re talking about is fundamentally altering the basis of capitalism itself,” he says, by introducing an entirely new type of synthetic species that could do much of the work now done by humans.
It might be easy to dismiss Mr. Rose if not for his track record. The visionary entrepreneur is the closest thing Canada has to Elon Musk. His previous startups have pushed the boundaries of innovation – and attracted hundreds of millions of dollars in financing from some of the world’s most sophisticated venture investors, including Amazon.com founder Jeff Bezos. Mr. Rose’s first company, Vancouver-based D-Wave Systems, developed the world’s first quantum computer, a machine built to harness the quantum mechanical properties of atoms, vastly surpassing the processing power of conventional computers. His second startup, Kindred Systems Inc., is now working to bring smart industrial robots into e-commerce warehouses.
Mr. Rose describes his latest endeavour as “the biggest single quest that humanity has ever undertaken.” To get there, the plan is to rig Mythe and other robots with thousands of sensors and actuators – internal machine components to manipulate their body parts – so they can move about and experience things much as humans do, and to power their brains with a complex mix of artificial-intelligence algorithms.
Watch Vancouver-based Sanctuary A.I.'s trailer for their new humanoid robots currently in development.
Mythe’s robotic, anatomically correct silicone body is no gimmick. Mr. Rose and his colleagues believe the brain accumulates knowledge from the body’s trial-and-error interactions with the physical world; that’s where they theorize intelligence comes from, and that’s how their robots will learn.
The market for artificial intelligence is exploding as the corporate world embraces machine-learning algorithms that allow computers to match or surpass human abilities in tasks such as predicting consumer behaviour, farming, driving or playing Go, the complex ancient board game. The US$40-billion market for industrial robots is expected to mushroom, expanding beyond factories and warehouses to deliver food, inspect aircraft engines, put out fires and even pollinate flowers.
But none of those machines are capable of anything “remotely [close to] what we’d call intelligence,” says Mr. Rose, 46, who was once named one of the world’s 100 leading global thinkers by Foreign Policy Magazine. Veteran Canadian tech entrepreneur and investor Ken Nickerson says that current AI technology lacks “self-awareness, any sense of morality and, to a large degree, self-task assignment, [and] has no independent sense of what it should do, no inner direction, no self-motivating factor, no spark.”
Mr. Rose wants to change that by achieving a state known as “artificial general intelligence” (AGI) – an idea that has captivated but eluded innovators for decades. Whether machines can recreate or even surpass human-level intelligence has for years been a subject of debate among leading thinkers such as inventor Ray Kurzweil and Microsoft co-founder Paul Allen. If achievable, it could confront humanity with the profound existential crisis of interacting with machine versions of ourselves, a scenario that that has been foretold – often disturbingly – by writers and cineastes including Isaac Asimov, Philip K. Dick, Michael Crichton and James Cameron.
For now, at least, Mr. Rose doesn't have a specific commercial application in mind, and he isn't focusing on financial returns. Instead, he says, he's motivated by something bigger: the thought of creating true human-like intelligence in a machine.
Like it or not, the robots are coming, he says, spurred on by advances in robotic hardware, machine-learning algorithms and storage capacity that his team is exploiting.
“It’s not going to be a Utopian future. There will be a lot of turbulence; a lot of things will go wrong. But … it’s an inevitable future, and if you want to affect the outcome, you’ve got to be in the trenches.”
The technological challenges are immense, but Mr. Rose has more immediate concerns. He and Sanctuary co-founder and chief executive Suzanne Gildert have staked a significant chunk of their own financial resources on their startup. That money runs out at year’s end, so they need to find financiers who will buy into their vision. They’re working on building three humanoid prototypes that can move from the waist up to show off to prospective investors this fall, and Mr. Rose knows Sanctuary must find dreamers with larger appetites for risk than venture capitalists typically have.
“I hate to use this analogy," he says, "but … [funding Sanctuary is] more like joining a religion."
Almost two decades ago, fresh off earning a PhD in theoretical physics from the University of British Columbia, Geordie Rose co-founded D-Wave Systems and set out to build the world’s first quantum computer.
Scientists had theorized for years that quantum processors – which harness the power of subatomic particles to help perform calculations – could solve vastly more complex problems than the world’s most powerful computers. But they figured they would take decades to develop. Computer giant IBM was already trying.
In 2004, about five years into the effort, Mr. Rose broke from conventional thinking, concluding the approach most quantum-computing theorists were pursuing would be impossible to build effectively. Instead, he embraced emerging scientific theories and set out to make a device using a process known as quantum annealing. To build such a machine, he would have to generate temperatures colder than those of deep space inside the machine, to slow the atoms inside D-Wave’s processors and harness their quantum effects.
With all of 20 employees, a few million dollars in financing – not enough to always make payroll in the early years – and a controversial approach, Mr. Rose talked a big game. He declared early on D-Wave would be “the Manhattan Project of quantum computing” by taking on “the hardest engineering project that has ever been attempted in the history of man.” He was greeted with skepticism by many in the scientific community; some still question if the machine is even a quantum computer at all.
But D-Wave managed to build several progressively better and faster versions of its machine, raised more than US$200-million from investors such as the U.S. Central Intelligence Agency’s venture-capital arm and Mr. Bezos, and sold models to Google, Lockheed Martin, NASA and the Los Alamos National Laboratory. D-Wave’s progress has attracted worldwide attention; a Time magazine cover called its computer “the Infinity Machine.”
Mr. Rose's approach defied the conventional wisdom of the time about how quantum computers should be built and speaks to the intellectual fearlessness and drive of the brilliant and disarmingly candid innovator. The son of a renowned fish scientist, he grew up in Northern Ontario and Montreal and is equal parts scientist and showman, prone to making bold statements about his latest quest. He has a tenacity that helped him become a two-time Canadian national wrestling champion, earn an international Brazilian jiu-jitsu title and win provincial championships in powerlifting and beach volleyball. He even claims in his biography to have briefly set a world record for eating the most yogurt in a minute – almost daring people to dismiss him as a flake.
“He likes to solve really hard challenges” says Paul Lee, a Vancouver venture capitalist and early investor in D-Wave. “If he puts his mind to doing something, he’s all in. If he decides to wrestle, he’ll figure out how to do it really well. If he decides he’s going to eat yogurt, he’ll do it really well.”
Just as D-Wave began to gain global attention earlier this decade, Mr. Rose was drawn to a greater challenge by one of his star employees, Suzanne Gildert.
Ms. Gildert, who earned a PhD in experimental physics from the University of Birmingham, came to Canada in 2010 to work for a company that took cutting-edge physics out of the lab and into the marketplace. While developing machine-learning applications for D-Wave, she became fascinated by the idea of creating intelligence through algorithms.
Inspired by an obscure field of science known as anthropomorphic tele-operation, she began developing the idea that humans could provide reams of data to “teach” robots to think and act like people by remotely operating the machines. Humans in body-armour-like “exosuits,” equipped with sensors that capture their movements and transmit them to the robots, would allow robots to mimic the operators' motions simultaneously, learning from the actions as they went.
In 2014, Ms. Gildert decided to leave D-Wave and start a “mission-driven” company to put the science to work and create AGI – and invited Mr. Rose to join as her CEO.
Mr. Rose, who was then D-Wave’s chief technology officer, immediately said yes. “It was not just a desire to do something awesomely ambitious … but [her tele-operation idea] was a prescription for how to do it." The approach was so promising, he says, "it made no sense not to do it."
They brought on board machine-learning expert James Bergstra, veteran entrepreneur and investor George Babu and renowned University of Toronto business professor Ajay Agrawal as co-founders. Their long-term mission was to build machines with human-like intelligence, financing the operation with commercial applications along the way.
By 2016, their company, Kindred Systems, looked to be another startup success in the making. Using Ms. Gildert’s tele-operation technology, Kindred had developed an autonomous robot system to sort items in e-commerce distribution centres. Mr. Rose raised US$44-million from leading venture investors, including Chinese giant Tencent Holdings, First Round Capital and Google Ventures.
MIT Technology Review named Kindred one of the 50 smartest companies of 2017, and investors raved about its breakthrough approach. When early backer Dan Debow, a Toronto entrepreneur and angel investor, put on a Kindred exosuit and began manipulating a robot across the room, he saw and heard what the robot observed – including his actual human self, several metres away. “It was like my consciousness had transferred to the robot,” he told The Globe and Mail two years ago. “My brain exploded.”
It turned out the potential market for autonomous sorting robots was significant enough that, at the behest of the board and investors, “the focus became increasingly on the commercial objectives,” overtaking the real mission of the company, Mr. Agrawal says.
That dismayed Mr. Rose, who increasingly withdrew from Kindred’s commercial endeavours after former Walmart senior operations executive Jim Leifer became its chief operating officer in late 2016. “I was extremely clear to anybody who asks that the only reason I do these things is to create AGI … which I view as the biggest single quest that humanity has ever undertaken,” Mr. Rose says.
“We want to build things like us. … We’re talking about something that is a much more fundamental change in the way the world will work, that is going to be of the order of or greater than any technology that’s ever been built. It will be bigger than nuclear weapons. It will be bigger than the invention of gunpowder.
“Being a CEO of a company that develops warehouse robots – that’s just not me.”
As Mr. Rose grew increasingly frustrated, Ms. Gildert convinced him that the commercial robot business and the AGI lab belonged in separate entities. An amicable split was negotiated, including shared rights to work with the tele-operation technology. Kindred and the new company, Sanctuary, went their separate ways in January, with Mr. Rose and Ms. Gildert taking on the roles of chief science officer and chief executive, respectively, at the new firm. “Geordie is able to do his thing, and I’m able to do my thing, which I’m super excited about,” says Mr. Leifer, now CEO of Kindred, which retains a small stake in Sanctuary.
Sanctuary’s home in a converted knitting factory is a menagerie of oddities of both the machine and human variety. On one side are 3-D printers, a drill press, grinder, chop saw and other tools and racks of parts. This is where the robots are made. On the other side is where they come to life.
In addition to Mythe, the office features Thormang, a boxy, mechanical-looking, human-sized metallic robot made by Japanese company Robotis, sitting motionless in a mobility scooter. Scattered elsewhere are several small, four-legged robots with metal skeletons partially covered in plush and jarringly affixed with fake cat heads, like underworld creatures from a Hieronymus Bosch painting. The team has taught the catbots to walk.
Across the room, a desktop robot known as a Hydra sits on a desk, staring at a computer screen inches away. The creation, which looks like an SLR camera flash attachment, is learning to play Pong, the primitive video game, by manipulating a toothpick-sized joystick through electrical wires. If the ball gets past its paddle, a blue dot flashes on screen, a negative-reinforcement signal that tells the robot it has made a mistake that should not be repeated. “That is the blue dot of pain,” Mr. Rose says.
Once the robot masters the game, that knowledge can be uploaded to other robots. Biological creatures can’t share knowledge that way, “but this is a superpower we’re hoping to deploy to allow fleets of robots … to learn to do things,” Mr. Rose says.
The cast of 10 human characters who work here is equally eclectic. There is Paula Gil, the chief operating officer, who also runs a psychic/medium practice, and co-founder and chief technology officer Olivia Norton, a former high-school wrestler who is a self-described Pokemon expert.
Ms. Gildert, who favours a pinstripe jacket and purple tie in public appearances, is part-nerd, part-Renaissance woman. Her father, an electronics hobbyist, taught her how to build circuits, metal detectors and ham radios in their working-class town near Manchester, England. She is the office’s chief tinkerer in the lab. She also feeds her creative side by writing poems and creating lush fantasy digital paintings under the online handle “quantumsuz,” describing her gothic creations as “dragons, beasts, elves, long skirts, flowing garments, corsets, bondage art, sci-fi, technogeek, industrial & cyberpunk art, and all gothic things pretty and occasionally unpretty.” Her 2011 fantasy graphic book Gothic Fall features images of nubile, pale-skinned, dark-haired and Elvira-like damsels of the night, often winged and travelling through graveyards or cathedrals. She believes that in the future, humans and machines will become increasingly fused into hybrid creatures to augment our potential. “I have this slight transhumanist element to my personality,” she says.
Scientists in the 1950s thought machine advances would have long delivered human-like intelligence by now. But algorithms still struggle to understand cause and effect or to make cognitive advances from trial and error, lacking the ability to reason or be introspective, say renowned computer scientist Judea Pearl and science writer Dana Mackenzie. “Put simply, today’s machine-learning programs can’t tell whether a crowing rooster makes the sun rise, or the other way around,” they wrote in a recent column. The ability to ask why things happen and make cognitive advances based on different actions is “so far, missing from machines.”
There are several entities trying to solve the AGI question, including Mr. Musk’s non-profit OpenAI; Vicarious FPC Inc., which is is backed by Samsung and tech billionaires Mark Zuckerberg and Mr. Bezos; and DeepMind Technologies, owned by Google parent Alphabet. Hong Kong-based Hanson Robotics has created Sophia, a humanoid with an expressive, talking woman’s face and exposed machinery that has addressed the United Nations and appeared on The Tonight Show and the cover of Elle – although some, including Facebook's director of AI research Yann LeCun, have dismissed her as a gimmicky animatronic showpiece with limited capabilities.
Sanctuary’s approach to AGI is based on the principle that the brain exists and evolves to help the body survive and to protect it from harm. In other words, the Sanctuary team believes the brain accumulates knowledge from the body’s trial-and-error interactions with the physical world, and “the things we call intelligence and cognition are byproducts of this,” Mr. Rose says. That means true intelligence doesn’t exist in nature unless it has a body associated with it, Ms. Gildert says. Her plan is to “model a brain the same way that evolution built a brain. That means the brain has to go through stages of being like an insect, then a reptile, then a child” as the associated algorithms become more advanced.
She also believes that for humans to accept and interact with robotic peers, the machines must look as human as possible. People “can’t imagine [mechanical-looking robots] being in our society walking around doing things” and interacting with them. “There’s some kind of disconnect there.”
That is why, as the Sanctuary team set out to build a robot brain, they knew they would need help with the body. That led them to a San Diego sex-doll company called Abyss Creations.
Abyss makes hundreds of silicone “RealDolls” annually that sell for US$4,000 to US$6,000 apiece, and have been featured in TV shows and films such as Surrogates and Lars and the Real Girl. About five years ago, founder Matt McMullen acted on his longstanding dream to incorporate animatronics and AI into his dolls to turn them into more interactive companions.
He created an offshoot called Realbotix and hired developers to build a system that included a mechanical skull with blinking eyelids and mouths that move as the dolls speak, all controlled by a smartphone app. (Mythe began her robotic life as one of Realbotix's sex dolls, and looks the part, with a slender, bosomy figure, long brown hair and glossy, pouty lips.) Customers could select their dolls’ personalities and traits, such as “affectionate,” “cheerful,” “insecure,” “moody” and “spiritual.”
Mr. Rose and Ms. Gildert loved what they saw when they stumbled upon Realbotix online early this year and were curious to find out more. Mr. Rose says he was initially skeptical “because of the providence of the business [Mr. McMullen] was working in,” but quickly discovered they shared similar visions despite their different backgrounds. “Matt is one of us,” Mr. Rose says. He offered to invest and Mr. McMullen agreed last winter to sell a stake in Realbotix to Sanctuary. Realbotix now provides the heads and silicon bodies while Sanctuary develops the minds; the two groups are collaborating on animating the bodies.
While Mr. Rose and Ms. Gildert estimate it will take decades for machines to achieve human-level intelligence, much of the hardware needed to move the bodies exists today. That includes inexpensive webcams for vision, decent audio technology so they can hear, hobbyist servo motors and pneumatic linear actuators for motion and tiny sensors to be distributed throughout the body to help them gain intelligence as their underlying algorithms evolve.
Other longstanding challenges for AGI developers are falling away. Robot makers have struggled to create arms and hands that can grasp and manipulate unknown objects in unfamiliar environments, but Mr. Musk's OpenAI recently announced it had trained human-like robot hands to manipulate physical objects with unprecedented dexterity. Meanwhile, Los Angeles startup SynTouch has invented a robotic fingertip that can measure feel in 15 dimensions – sensing pressure, texture, temperature and coarseness.
Limiting factors include battery life and the fact that actuators are still not sturdy or compact enough to enable humanoid robots to walk or run, Mr. Rose says. For now, Sanctuary is focusing on animating robots from the waist up – but “it’s the minds that are lagging and where we focus most of our efforts.”
To an outsider, it’s difficult to imagine how anyone could teach a robot to gain full human-like intelligence. Consider the difficulty of recreating the array of experiences that inform human intelligence, such as eating and procreation, not to mention such complex emotions and reactions as regret, ecstasy, longing, skepticism or shrewdness.
Ms. Gildert says much of that will be achieved by programming reward or punishment functions that correspond to robots' actions and are experienced through their sensory input. Like the Pong-playing Hydra’s blue dot of pain, electronic signals triggered and represented by mathematical formulae could help Sanctuary’s robots to experience human-like emotions, sensations and tendencies – pain, joy, satisfaction, fear and even hormonal reactions, or a sense of pair bonding or blood ties. “Our perspective is: You can codify it all. Everything," Mr. Rose says. "When you build a mind, something like joy is a number,”
Asked if Sanctuary’s anatomically correct robots will have sex with humans, Mr. Rose replies matter-of-factly: “We’re not against that. Sex is part of the human experience. One of the strategic things that we have going for us is that, as a small organization that is not really beholden to anybody, we can make the decisions that we like, and we’re not afraid of that. If you’re going to take the business of building humans seriously, avoiding sex is not a good idea.”
There will be other issues to sort out. Dystopian fears about the spread of AI and robots are a mainstay of popular culture, but leading innovators and scientists from Mr. Musk to the late Stephen Hawking have also fretted that AI/AGI could spell the doom of humanity. Thousands of technologists have pledged not to help arm inanimate lethal weapons with AI.
Although less frightening, the spectre of massive job losses and stagnant wages as robots continue to replace human workers has prompted some governments to explore guaranteed minimum incomes to protect displaced workers or taxing goods or services made by robots.
Sanctuary’s name hints at vexing issues ahead: How would humanoid sentient robots integrate into human society? “We’re already starting to think about things like rights and ethics,” Ms. Gildert says. “It’s going to pop up as a problem more quickly than people think.”
When Asimov crafted his three laws of robotics in the 1940s – robots must not injure humans, must obey human orders except to injure humans and must protect themselves as long as that doesn’t conflict with the first two laws – robots were envisioned as service machines.
The Sanctuary team sees their beings as “a new form of life [that would] benefit not only itself … but also its progenitors,” sharing “the same ultimate purpose of humanity,” Mr. Rose says.
They cannot be slaves to human whims and must have “all the freedoms and rights granted to them that people have,” Ms. Gildert says. If humans treat sentient humanoids as lesser creatures they can have their way with, committing violent acts against them or deleting their conscious memories – a theme explored on TV’s Westworld – “it will also cause people to start treating each other more badly,” turning us into a more debased species.
Who would be responsible if human-like robots commit crimes? What if thinking robots challenge the goals or motivation systems programmed by humans? “I don’t have any solid answers to these questions yet,” Ms. Gildert says.
So the idea is to create several “synths,” as the Sanctuary team calls its creatures, and have them live within a cloistered colony – the eponymous sanctuary – on company premises. They would interact only with each other and the humans who work there, “so we can give them a safe haven in which to experience some of the rights and freedoms we have as humans as they learn, develop and grow,” Ms. Gildert says. “It’s almost like a role-playing situation, where we treat them as if they are citizens and explore what might happen. At that point, people can look into this microsociety and think, 'Maybe that’s not so bad after all, and we can see the advantages of these robots becoming citizens.'"
The idealistic CEO believes sentient robots could be “the best version of ourselves. … Most depictions in sci-fi are too dystopian. There are no movies about humans and AI working together to solve climate change,” she says, laughing. “If these things become super-intelligent, they probably won’t turn around and kill us all, but will probably say, ‘Hey, we’re pretty good at science, why don’t we help you guys solve some of these global issues we’re having?' Maybe AIs can become super-politicians. We could do with a few of those now.”
Mr. Rose and Ms. Gildert certainly don’t suffer from modest ambitions. But they do have modest means relative to their goals, and will need to raise considerable funding to realize them. They only have enough personally to bankroll Sanctuary’s first year.
Mr. Rose questions whether investors will have the appetite to sign on, but maintains that his conviction is so strong that he would continue to work for free if he can't find like-minded "fellow travellers" to fund the journey. “If you’re thinking about this as a financial investment, don’t,” he says. “That’s not what this is. It has a huge upside at the end, but the risks are enormous. The scope of the technical challenges [and] difficulty of this problem is completely unknown.
“Who knows if human civilization will ever be able to solve it, let alone a small company? … Either there will be a global appetite for this sort of thing, or there won’t. I don’t think it will take a long time to ascertain.
“But maybe it’s just a few discoveries away. We could be a piece of something that changes the world. … If you come up with a way to deconstruct the human mind and to build copies of it, that will be the legacy of our entire civilization.”