Thanks to AI, the startup Iris AI is about to disrupt scientific research by putting in some order. It’s a first step before it transforms its intelligent assistant into a virtual scientist.
Moffett Field Campus, Mountain View. June 15, 2015. Anita Schjøll Brede listens attentively to a welcome address delivered by Peter Diamandis, the co-founder of Singularity University. Every face surrounding her in the big conference room is focused on what’s happening on stage. 80 students were selected from more than 3,000 applicants to participate in this successful Global Solutions Program at Singularity University. Among them are two people, unknown to her then, with whom, in a matter of months, she will launch a startup that will revolutionize scientific research.
Just like every summer since its creation in 2009, Singularity University offers a summer training course focused on innovation, new technologies and creating new companies whose objectives are to positively change the future of humanity. This Silicon Valley satellite was not set up just anywhere: Mountain View is the seat of Google. This small town in the outskirts of San Francisco is also known to have housed one of NASA’s largest research centers for many years. In fact, it was right on Moffett Field’s grounds, where the “SU” campus is located.
With her hair shaved on the sides, tattoo-covered arms and impeccable suit jacket, Anita wears her punk look with pride. Having “never had any definite career plans,” she says she’d never imagined working here one day. She studied theater. After finishing her studies in Oslo, the young Norwegian decided to launch her own theater company to present shows for children. But the process was a revelation to Anita, who was “quickly drawn by the business side” of the undertaking. Six weeks later, she was flying to the US to study entrepreneurship at Berkeley and management at Stanford. In 2015, she applied to Singularity University’s summer camp.
During the intensive 10-week program, Anita became friends with Maria Ritola and Jacobo Elosua. For their group work, the three young entrepreneurs chose to look into why scientific progress is so slow. They believed this was partly due to the complexity of research work. This is where the idea of Iris AI, a smart research tool that uses artificial intelligence’s analytical capacity to connect “anyone trying to find accurate information and the huge amount of scientific articles available on the web.” Today, Iris AI’s clients are essentially “science geeks, students and researchers.”
When a user enters a scientific article’s URL into Iris’ interface, AI quickly produces a visual map showing the article’s key themes, then proposes a series of other scientific articles related to each theme. More precisely, Iris analyzes a given text, identifies its keywords, then searches those words’ synonyms. “We’re running an open source database with over 50 million items,” says Anita.
The map works like a Voronoi diagram, a mathematical method that breaks information down into sets and subsets. Each set represents a key word and contains several subsets allowing the user to narrow his or her research. Anita and her co-founders created a simple and colorful interface to make it more pleasant to use. “People told us we didn’t need to make our product pretty because it was for researchers,” she recalls. “We don’t agree.”
At this stage, Iris allows researchers to simply advance in their research and, with just a few clicks, have access to all articles on their subject of interest. The platform’s aesthetic contributes to its clarity, but the way it operates could be improved.
For the time being, Iris is only able to read abstracts. That’s because most scientific publications use archaic hosting systems. When you enter a URL into Iris’ searchbar, the AI goes to the given address to locate the relevant keywords. Unfortunately, scientific publications are often presented in summary form, and oblige you to download the full article as a PDF. So right now, Iris can’t read the full articles. “It’s the only format that does not work right now. If we insert the URL of a .pdf file in Iris, it runs into a wall.” For Anita, “the next step in development is having Iris read entire articles instead of abstracts only.”
But Anita Schjøll Brede wants to go even further. She hopes that, “artificial intelligence will be able to produce abstracts by itself within two or three years.” The ultimate goal is to convert this virtual intelligence into a super scientist of its own. It’ll take some time, but that’s the project’s logical next step. Thanks to its processing power, Iris will, in the next few years, be able to effectively get rid of the the confusion that currently clouds research. And once this is done, machine learning – the ambitious technique that allows machines to understand and improve by analyzing data – should make Iris able to reach its own conclusions, drawing from the huge body of existing research.
But we’ll have to wait a decade before Iris can present its own scientific hypotheses, Anita says. The reason for this is that the machine learning process of artificial intelligence is not fully developed yet. “I tested Iris’ skills on a study that looked at skin infections,” says Anita. “It gave me very varied but not very relevant conclusions, with articles ranging from elliptical galaxies to the effects of sustainable flushing in buildings…”
Iris shouldn’t feel bad about this though, because all forms of artificial intelligence are in the same boat. Creators aren’t too worried either, when they look at the spectacular advances they’ve made these last few years and the exponential growth of technological progress. Artificial intelligence is set to be a game changer in scientific research. For this reason, researchers around the world are closely following its evolution and turning their attention to Silicon Valley.
At first glance, nothing distinguishes Palo Alto from the other wealthy cities that line San Francisco Bay. Located north of Silicon Valley, it is nevertheless one of the most advanced technological hubs on the planet. It’s home to corporate headquarters like Tesla Motors and Facebook, as well as Stanford University, the third largest university in the world this year, according to the Times Higher Education. At the edge of a busy city road and a few blocks from the McLaren and Tesla dealers, there is a building with a clear exterior and a flat roof. It’s halfway between a high school and a sales depot. You wouldn’t expect it to be the scene of anything important.
But between September 30 and October 2, 2016, Palo Alto’s Elks Lodge was the center of an extraordinary meeting. No one even realized it. In a large room with bare walls, located on the second floor of the building, 40 of the greatest contemporary scientists gathered to discuss the promises of artificial intelligence. In fact, if you look at the building more closely, you’ll notice one element suggesting it’s not a regular place: the big letters of the word “Elks” that appears on the exterior. The Order of Elks was a 19th century American brotherhood from in New York, a private club that now has more than 1 million members worldwide. Some of the most famous were five American presidents – including Roosevelt and JFK – as well as celebrities like Clint Eastwood and the magician Harry Houdini. The California lodge is familiar with elite meetings of this kind.
The event, which was organized by the Foresight Institute – a non-profit that supports technological advances – was dedicated to exploring the artificial intelligence’s potential for scientific progress. Its president, Julia Bossman, explained that it aimed to “generate new reflections, new projects and new funding at the crossroads of these fields, to apply the most advanced computer science technology to the challenges of atomic precision.” Since the creation of the institute, the team’s mission is to bring together researchers from previously remote fields to facilitate potentially revolutionary collaborations.
Away from the swimming pool, the fitness center and the bar-restaurant, the small “Palo Alto” room brought together big names in artificial intelligence, robotics, bioengineering and nanoscience during intensive workshops. Researchers present at the scene worked for Google and Samsung as well as major universities such as Stanford or Polytechnique de Lausanne. DARPA, a US government agency specializing in research on new technologies for military use, sent an emissary. The agency is famous for its research on robot soldiers and drones. An engineer from Zee.Aero, the company secretly founded by Google co-founder Larry Page to design flying cars, also took part.
Finally, the leading specialists in artificial intelligence – including Dr. Ben Goertzel, the director of Hanson Robotics’ AI department – were also there. For three days, science and intelligence converged in a series of discussions. But nothing was leaked to the media. The institute is still compiling the results of these meetings and has promised to publish them soon on its website.
It is no surprise that such skilled scientists are so interested in artificial intelligence. According to Ben Goertzel, by taking charge of the most laborious aspects of scientists’ work, AI will allow researchers to focus exclusively on their vision and raise the bar of their ambitions. AI’s extraordinary learning, analysis and compilation capabilities will allow scientists to make giant leaps in their field. Over the next decade, smart tools such as Iris AI will dramatically decrease the research time required to come up with new hypotheses. Artificial intelligences could be transformed from super assistants into super scientists, as Anita hopes.
In fact, that process has already begun. Medicine is one scientific domain that most benefits from the latest advances in AI. In this crucial field, those advances don’t just make life easier, they save it, too.
“How do we deal with a society in which Google tends to our health instead of the local doctor?” Laurent Alexandre, the founder of Doctissimo, asked himself during a recent debate on the links between disease and society. This businessman and surgeon in training has long been interested in transhumanism and the issues that plague the medical field, notably those concerning artificial intelligence.
Unlike other scientific fields, medicine is less inclined to accept changes drummed up by individuals outside the medical field. But these days, many prominent individuals with no medical background are suggesting tectonic shifts: Microsoft plans to end cancer by 2026, Mark Zuckerberg plans to eradicate all human disease by 2099, and Google is working hard to extend life expectancy by 25 years. “In 10 or 20 years, artificial intelligence will be our doctor,” says Laurent Alexandre. “For the medical community, this promises a shocking shift in power.” Not everyone agrees with this forecast.
If medical artificial intelligence scares many specialists, others are excited by the prospect of being freed from tasks that take their attention away from patients. For Jurgi Camblong, the founder of Sophia Genetics – a world leader in data-driven medicine – “artificial intelligence will further humanize medicine.” A doctor freed from the technical constraints of his profession will be able to get closer to his or her patients. “Once replaced by the machine, the doctor of tomorrow will have the role of a consultant,” predicts Guy Vallancien, a French surgeon, professor at the University Paris Descartes and member of the National Academy of Medicine.
Beyond the debate surrounding these developments, AI is already being applied in many medical settings. Their success forces us to admit that so far, it all looks encouraging. Last August, Watson, IBM’s AI platform, used Big Data to detect the first signs of leukemia of a patient in Japan, at a stage when it usually goes unnoticed by human doctors. Watson’s interpretation of all of the patient’s medical data, as shown in his neural networks, allowed the platform to accurately predict how the condition would evolve.
Google researchers decided to try their hand with the medical startup Verily, a subsidiary of Alphabet. In collaboration with the French laboratory Sanofi, Verily announced last September the creation of a joint venture called Onduo. It was funded to the tune of 500 million and destined to fight diabetes. Verily’s artificial intelligence specialists developed an intelligent algorithm that can detect eye diseases in nearly half of diabetes patients. After analyzing a multitude of photographs of human retinas, their artificial intelligence learned through machine learning to recognize if the patient had diabetic retinopathy.
These examples are all encouraging, but they’re nothing compared to the advances made by Sentient Technologies. The Silicon Valley company was founded eight years ago by the Frenchman Antoine Blondeau, its current CEO. Sentient is best known for its work in e-marketing and trading, but it’s also the brainchild behind the world’s largest AI applications. The California-based company recently developed a medical AI platform in partnership with MIT. “Drawing from our work in trading, we created an application to prevent septic shock,” he says. Sepsis is a violent infection triggered by a germ that spreads throughout the body via the blood. It is now the leading cause of death in emergency room throughout the United States. “If you get it, there’s a high chance you’ll die,” says Antoine Blondeau.
Before Sentient and MIT created their application, there was no way to prevent the arrival of infection. Their AI is able to predict sepsis 30 minutes in advance, a crucial time frame for action. The system examines blood pressure and detects warning signs before shock sets in.
“We have a 91% success rate. Health is a highly regulated field, so it will take time before the product is born. But it is an application that will save many lives,” Blondeau concludes. Sepsis was discovered in the 19th century. It’s a serious blood infection that has defeated generations of doctors and researchers, who were unable to predict its arrival. It took less than a year for Sentient to obtain these miraculous results.
Even though Iris AI’s artificial intelligence may be far from the three founders’ ultimate goal, it works well enough now to help researchers conduct their work. Last October 30, the company collaborated with the Future Earth Media Lab and the Stockholm Resilience Center to organize a scithon in Stockholm.
Scithons, or science hackathons, are timed events during which participants are challenged to solve complex scientific problems by using one innovation. Four teams were randomly assembled. Their mission was to determine how long global urbanization would take to make a positive impact on the biosphere, and in what way. They had four hours to figure it out. To carry out their research, two of the teams used Iris AI’s interface.
The jury has not yet revealed the winner of the competition, but the teams who used Iris told Maria Ritola, one of the company’s co-founders, that thanks to the technology, they were able to see the problem both more broadly and more precisely. Iris spared them hours of tedious research by compiling all the relevant data quickly, leaving them the time to conduct the rest of the work. Iris may be in a development phase, but what it’s capable of accomplishing is already revolutionary. It can do what scientific research needs above all else: eliminate the time that man is forced to waste before he can actually solve a problem.