by Olaf Groth
Qure.ai helps compress that diagnostic process into a single day, Warier explains. Its platform can scan an X-ray image for abnormalities and return results in seconds. If the image reveals signs of tuberculosis, then the doctor can order the appropriate but costly microbiology tests to make the conclusive diagnosis. The whole process gets compressed down to a matter of a few hours, and sick patients can go home with the medicines they need to cure what ails them. Qure.ai focuses specifically on chest X-rays and head CT scans, automating the image analysis, helping identify abnormal pathologies, and then prioritizing critical cases for radiologists to review on an expedited basis.
They focus on those two core imaging procedures, but the platform is opening up an array of other possibilities, Warier says, including a move to other markets, such as the United States, and audits of a hospital’s radiological diagnoses. “We can process all the X-rays a hospital does in a year, and do so in a few hours,” he says. “And then we can compare that with the diagnoses on the written report using natural language processing. So, we can immediately compare and tell the hospital that there are these, say, 100 X-rays that were incorrectly reported.”
Warier already sees a rapidly growing community of AI developers in India who will take these technologies in new directions. While still definitely behind China and the United States, the widespread and open availability of AI models and computer technologies makes the development of a vibrant industry a matter of talent and data. Qure.ai worked so well because it could access about 5 million relevant medical images and then tweak them, rotating them or cropping them in different ways to expand its training set even further. “We have done a lot of cutting-edge research on model architectures because our problems are different,” Warier says. “Interpreting radiology images is a much more challenging task than decoding an image of a cat or a dog.” Almost all the current AI research focuses on images that are about 100 times smaller than a typical chest X-ray, he explains. However, several open source repositories exist, and almost everyone in the field wants to publish their work, so there’s a lot of open literature. Warier and many others will continue to build and innovate on top of what’s already available. “For India, people here can work on the bleeding edge with the latest technologies and be up and running in no time,” he says. “There’s a lot of democratization and a huge amount of talent that can take advantage of this opportunity.”
Not unlike Qure.ai, the talented developers at SigTuple also used AI systems to tackle a major time gap in rural health care. The company’s Manthana AI engine can digitize and analyze certain pathology tests, returning reports in minutes, passing them along to pathologists to review and then on to doctors to make treatment decisions in far less time. That speed can have profound effects on rural care for acute cases of Dengue fever or other diseases. More than 100 million Indians live more than 100 kilometers from a hospital, with access to bare-bones clinics that don’t have diagnostic equipment. Up to four or five hours can elapse while an ambulance is dispatched, arrives, and transports the patient back to a hospital. SigTuple aims to put its equipment in the clinics, allowing for a rapid enough diagnosis that ambulances can be dispatched with medicines to help treat the patient while on the return trip, essentially halving the time to care.
As in Nigeria or Barbados, the country still faces enough basic infrastructural problems that could hinder the application of AI solutions. In many cases, rural clinics don’t have the connectivity or proper training to make use of technologies. While mobile connectivity is widely available, broadband speeds are not. The Indian national government doubled its budget for AI, 3-D printing, and other advanced technologies to $477 million in 2017,*** and its national digital identification system, called Aadhaar, has opened possibilities for a range of data-driven financial and health care services that require online verification of individuals. But Indian states have significant control over the application of policy on the ground, so top-down initiatives don’t fly as well as they do in, say, China. And, despite rapid educational and economic gains across much of Indian society over the past fifty years, about 300 million of its 1.2 billion residents still live in poverty, says Srikanth Nadhamuni, CEO of Khosla Labs in Bangalore. Any use of AI or other advanced technologies needs to focus on “the bottom of the pyramid where there are significant challenges and compelling needs,” Nadhamuni says. “Making quality health care affordable to the rural poor through AI-enabled diagnostics with smartphone-based sensors and delivered at kirana (mom-and-pop stores) could transform the country’s health care.”
The same rural and impoverished chasms that technology has started to bridge remain barriers for sustainability. Almost two-thirds of the Indian population lives in rural areas, but 70 percent of the country’s GDP churns out of its mega-cities, including Mumbai, New Delhi, Bangalore, and Hyderabad. AI-powered systems can overcome some of these barriers, but how can the companies developing the platforms make money from a population that has so little? And since these places likely won’t be primary markets, how will companies adapt and adjust their applications to accommodate different cultural contexts, such as the dozens of different dialects spoken across the country? But the question isn’t whether rural or poor Indians will accept artificial intelligence, Nadhamuni says: “People will take to solutions that solve their problems.”
The concern with any technology, though, is its potential to create as many problems as it does solutions. New AI systems can democratize access to resources and amplify voices across class and income boundaries, but their benefits won’t accrue equitably to everyone they touch. Rural Kenya’s smallholder farmers have little to their names other than a tiny plot of land, a few tools, and a mobile phone to keep tabs on market prices, weather reports, and crop conditions. That digital access has given them far more control over their livelihoods. However, it also extends the widening advantage of the global technology elite who supply it. The digital intelligentsia that creates and sells technology accumulates disproportionate power, and that accumulation will only accelerate as the capabilities of these products and platforms increase, as is the case with AI.
As companies analyze data on global weather patterns, agricultural production, market prices, and infrastructure conditions, they can quickly shift global resources from one market to another. Few rural farmers and other digitally disadvantaged populations fully understand the global machinations that investment banks, global food conglomerates, and high-tech firms will play. So, while the poor Kenyan farmer or the rural Indian mother of four might benefit from the technology in their hands, they have far less opportunity to maximize their power or give voice to concerns about the ethics of food justice or global income distribution.
ASTRO BOY
Astro Boy made his manga debut back in 1952, but he lived in a future, science-fiction world where humans and robots coexisted in harmony. An android with human emotions, Astro was created by the head of the Ministry of Science to replace his lost son, who died in a self-driving car accident. The protagonist, known as “Mighty Atom” to the many Japanese who followed the series over the next sixteen years in Weekly Shōnen Magazine, would soon disappoint the inimitable Dr. Umataro Tenma, who realized the android could not replace the void of his lost boy. Astro would be sold to a robot circus, saved by a magnanimous professor, become part of a robotic family, and set off on various adventures.
The 112 chapters of the series, along with the subsequent remakes and spin-offs it inspired, now rank among the most influential forces in the history of Japanese manga and anime. But Astro Boy also provides one of the earliest pop-culture references for the more-symbiotic mindset that Japanese citizens hold for human-robot interaction. The greater affinity for androids and robots has tangible roots in demographics and economics—primarily as a replacement for a shrinking labor pool—but it also grows out of a philosophical tradition that doesn’t consider humans exceptional in the ways that Western traditions do. It’s no surprise, then, that the leadin
g edge of Japanese AI tends to revolve around humanoid and other robotics, with development of both leading toward what some Japanese experts call “Society 5.0.”
The urgency today stems primarily from the demographic cliff the country faces. Low rates of procreation, and an aversion to immigration, have flipped Japan’s age pyramid onto its point, leaving companies in a scramble to replace retiring workers. “Labor saving technology of any kind is critical,” says Kenji Kushida, a Japan Program Research Scholar at Stanford University’s Walter H. Shorenstein Asia-Pacific Research Center. “That’s where AI and robotics really come in.” However, the latest generation of robots was born into a limbo, caught between two popular conceptions. While robots do far more in 2018 than the traditional factory machines programmed to do one task repeatedly, they remain a far cry from the sci-fi humanoids of Hollywood and Japanese anime.
Still, in this middle ground, developers have made significant gains on the twin tasks of robot perception and grasping, thanks largely to new applications of machine learning. These advances have pushed robotics near a breakthrough in the grasping of arbitrarily shaped objects, such as pine cones, pencils, or wine glasses. “I’ve been working on the grasping problem for 35 years, and now, with cloud robotics collectively learning from millions of examples, I feel we are getting close to solving it,” says Ken Goldberg, head of UC-Berkeley’s Center for People and Robots.
Solving the grasping problem would move us one step closer to creating a humanoid robot, but neither grasping nor perception help much in everyday situations if the robot doesn’t also possess some basic temporal and causal sense about the environment in which it operates. That sort of cognition includes the kind of reasoning that lets one know that a full mug of coffee needs to remain upright and steady when carried from tray to table, while an empty mug can be flipped and moved swiftly. This level of understanding heuristics—the sort of common sense concepts humans grasp without explicit instruction, or with just one or two tries at a young age—could have broad implications across AI. Yet, they’ve slipped out of the spotlight, often replaced by brute-force processing on deep neural networks.
The wealth of Japanese research on ways to imbue greater environmental awareness in machines has yet to produce groundbreaking discoveries, but that shortcoming hasn’t stopped the country from widespread adoption of automation throughout the country. Stores that normally operated twenty-four hours a day have curbed hours because they can’t find enough workers, Kushida says. Many restaurants have replaced cashiers with machines customers use to order and pay. Factories and other skilled trade shops are seeking ways to collect retiring workers’ craftsmanship and knowledge and embed it in an AI system that can hold that institutional knowledge and, eventually, pass it along to a new generation of workers.
Fortunately, the integration of more AI-powered robotics and automation didn’t require a massive cultural shift in the country, thanks partly to early signals like Astro Boy. “When I talk to European colleagues, robots are conceived as some monstrous existential threat that may destroy human society,” says Junichi Tsujii, director of the Artificial Intelligence Research Center at Japan’s National Institute of Advanced Industrial Science and Technology. “It’s a monster type of image there, but in Japan the robot is a protector of human beings or some kind of friend.” Because it can do so in a culture that more readily embraces artificial intelligence, AIST focuses most of its efforts on the development of AI systems that will integrate directly in the physical world, particularly in manufacturing, health care, and elderly care. There are limits, Tsujii notes, as Japan clings to its traditions. Yet, even those aren’t always sacred. Developers made waves in the country when they taught robots how to perform a traditional dance that was starting to fade away.
In fact, Tsujii and other Japanese AI developers say they believe the general acceptance of advanced technologies derives from something even more-deeply embedded than demographics or pop culture. It emerges from a core philosophical belief that’s fairly common in Eastern traditions—that humans aren’t as exceptional as they’re made out to be in Western thinking. People, animals, plants, and even robots and AI agents reside along a continuum of existence. “We don’t have the concept of a creator like God,” Tsujii says. “Western civilization is always thinking that human beings are a copy of God and special privileges are given to human beings. Asian cultures don’t have that kind of thinking; it’s more continuous from animals to human beings.” And neither need to be perfect or flawless. In fact, the Japanese have a word for that too: wabi-sabi, which includes an appreciation for the little imperfections in everything, from humans to nature to robots.
Yasuo Kuniyoshi goes so far as to stress the importance of lifelike humanoids for the future success of artificial intelligence—for development purposes, but also for integration into society. As director of the Intelligent Systems and Informatics Laboratory at the University of Tokyo, Kuniyoshi designs systems to mimic human anatomy and neurology as closely as possible. “We are trying to build a humanlike thing,” he says. “We don’t feel it’s really an evil thing or a scary thing. . . . That’s probably the difference between Western people and Japanese people. Many Western people cannot tolerate a separate existence that’s equal to humans.”
THE CERN OF AI
Canada shares the Western view, but that mindset—along with the outsized influence of a handful of prominent AI experts—had driven it toward a far less overbearing and more cooperative approach to AI development than in most countries. In a sense, the Canadian ecosystem has become the CERN of AI, building an environment and resources that facilitate mutual discovery much like the famous European Organization for Nuclear Research (CERN) does as a global community research hub for quantum physics. With large corporate actors dominating most of the large data sets in the United States and China, an influential group of developers and investors in Montreal have set out to create an international network of data generators. Canada has a strong start-up ecosystem and commands respect around the world, as do many of its top AI minds, and their openness can help draw more international talent and cross-country connections.
It helps that some of the top AI minds in Montreal, who also happen to be among the most well-known researchers worldwide, share a similar philosophy and push the concept of nonpredatory market competition. Luminaries like Yann LeCun, Geoffrey Hinton, and Yoshua Bengio exert a major influence on the Canadian AI scene. Bengio has an especially considerable impact, says Patrick Poirier, founder and CEO of Erudite AI, which is developing an AI-enhanced peer-to-peer tutoring platform. “Whenever you have a role model who believes and promotes certain traits, you may adopt some of those traits,” Poirier says. “I think he has a quite beneficial influence on the market in that regard.” On the reverse side, there are rarely big exits for start-ups in Montreal. Maybe if the developers tasted the blood of big profits, they might push harder for it. As it stands, Poirier says, “the community rarely understands or values stock options, and motivation remains more driven by social impact.”
Bengio established Element AI, perhaps Canada’s best-known AI start-up and one that reflects much of the current mindset in the country. The firm set out to provide an alternative model to the hording of data and talent by large corporations, says Jean-Sebastien Cournoyer, one of the firm’s cofounders and a partner at Real Ventures. Before the big Internet titans could swoop in and poach the well-known AI minds and resources in Montreal, Element developed a fellowship program through which top academics could contribute to the company—and get paid quite handsomely to do so—while sustaining their academic research and training the next, critical generation of AI expertise. And with that, it sought to provide a counter to the concentration of AI power in the hands of a dozen or so companies worldwide. “Canadians are collaborative. We’re not known to dominate. We’re friends with pretty much every country,” Cournoyer says. “So, the mindset was, ‘Let’s build an AI company and build an AI platform, but pro
vide it as a service to all the companies that need it to maintain relevance.’”
By combining the Element platform and the data from its customers, it can help smaller firms create AI to run their businesses. Those clients, in turn, share the knowledge they acquire from data with companies in other industries to strengthen the AI platform as a whole—to build stronger systems for companies that can’t build their own. “Our Canadian roots had us thinking about how to build ecosystems that help the world get access to AI,” Cournoyer says. In fact, the founders initially considered launching as a not-for-profit, but realized they needed to go the for-profit route to attract top talent and sell the software that helps fund the company. Nine months after Element launched, it pulled in a $100 million round of funding. “For AI to be deployed the way we want as a society, to make us more productive and efficient and grow,” he says, “we’ll also have to evolve our social fabric and social support system.”
But that fabric—the norms implicit in the social contract—can vary widely from one country to the next. The Canadian AI scene envisions a more inclusive future. Japan works toward greater automation. The military helps power innovation in Israel, while Russia continues to grow its heritage of science, if not entrepreneurship. China and the United States connect powerful academics with even more powerful commercial sectors. And yet, the borders between countries can’t contain these diverse viewpoints, any more than they can contain the vast flow of data around the world. And so, the grand contest to influence the future of an AI-powered world begins, with the world’s powers racing to assert their values, trust, and power.
