
Big data, artificial intelligence, nanotechnology and robotics will undergo an exponential change that’s already in sight. The future is here.
The three laws of robotics are:
-A robot will not harm a human being or, by inaction, allow a human being to be harmed.
-A robot will follow the orders of human beings, except when they are in conflict with the first law.
-A robot will protect itself as long as this protection does not conflict with the first or second law.
When Isaac Asimov announced these laws for the first time three quarters of a century ago, they were a kind of moral code for the fictional world inhabited by the characters in his novels and stories. But no longer. The future is here and the three laws of robotics are the simplified answer to dilemmas that are being posed now or will be soon.
If not, what are the accidents involving driverless cars? How will an automatic car resolve the dilemma between driving over a human or slamming against a wall with passengers on board? What limits will be imposed on autonomous machines designed for war? What can machines deduce about our personality? What possibilities does artificial intelligence open in the field of health? The simple answers to these questions is in algorithms. Nevertheless, societies are not run by mathematical formulas but by moral and judicial norms.
The new electricity
Andrew Ng, professor of Computing Sciences and Electrical Engineering at Stanford University, has a phrase that’s often repeated: “Artificial intelligence is the new electricity.” It’s his way of warning that the industry, business and health of the future will revolve around artificial intelligence.
Enrique Puertas, professor of Artificial Intelligence and director of the University Master in Big Data Analytics at Universidad Europea, explains that artificial intelligence is a 50-year-old concept that has undergone a surprising boom in the past three or four years. Why? Because of big data: “We can process huge volumes of information that was limited before. With the capacities of the Cloud it’s easy to put 200 computers to work to process huge quantities of data.”
Another detail has democratized access to artificial intelligence: graphics cards, which are optimized so as to make mathematical calculations and complex operations with matrices, are already used to program algorithms of artificial intelligence and automatic learning. “This makes it possible to considerably reduce the times for constructing learning models,” says the professor.
Deep Blue, the IBM supercomputer, beat world chess champion Garry Kasparov en 1997 because of the brute force of its computing powers: 200 million positions per second, while Watson, which beat humans in the Jeopardy competition in 2011, based its victory on the treatment of massive amounts of data. Neither machine was based on learning. Yet today it’s not enough to apply stored data to merit the distinction of artificial intelligence: this concept is usually associated with the capacity for automatic or ‘machine learning.’
“I tell my students that artificial intelligence is like the Enterprise in Star Trek: it explores new worlds that humans still have not reached,” says Puertas to illustrate that we’re faced with a changing concept.
The term ‘intelligent’ implies that the algorithm evolves and learns. “Algorithms develop and become more intelligent, in quotation marks. They can carry out certain functions but are incapable of others,” says Diego Gachet, professor of Informatics Languages and Systems at Universidad Europea. He says Google recently used a robot to make a reservation by phone in a restaurant and the employee who took the order didn’t even realize it was a robot. The robot adapted its conversation to what the employee was saying, based on millions of training conversations. Just like the T-1000 in Terminator 2.
“They can construct knowledge and find new patterns,” the professor goes on. There is a kind of learning, non-supervised learning or by reinforcement, that functions more or less like the conditioned reflexes Pavlov applied to dogs: reward and penalization. You ask the algorithm to do something, and if it does so correctly, you reward it. This means that it changes the value of some variables.”
So are they really intelligent? Gachet still thinks it’s a stretch to compare computerized technology, which is based on mathematics, with natural intelligence, which stems from the capacity to survive in one’s surroundings.
He wrote his thesis about a mobile robot that had to find its own path through trial and error. That was in the 1990s. Combined with the massive use of data, the computational learning technique has uses that were previously unimaginable.
Models to predict behavior
Artificial intelligence applied to large amounts of data make it possible to establish a behavior model that a person could not detect on his own. For example, it can be used to anticipate when a machine is going to stop functioning. Predictive models based on the relationships of multiple sources of data (company information, social networks, demographic information…) are also the basis for a discipline that is on the rise: Business Analytics, which makes it possible to predict commercial trends.
They are only data but their power is so great that not even anonymity can resist them. According to a recent study by University College London and the Alan Turing Institute, an algorithm of automatic learning applied to meta-data makes it possible to identify an anonymous user of Twitter with 96.7% reliability. Enrique Puertas cites another example: with just the information on social networks, it is possible to predict the political party a person will vote for and his or her sexual orientation and religion, with an extremely high degree of accuracy.
The potential uses of artificial intelligence have raised concern among legislators and experts. Article 22 of the General Data Protection Regulation, applicable throughout the European Union, states that a person “shall have the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her.” And yet the same article foresees exceptions.
Another case illustrates the power of data when it receives the correct treatment: some programmers designed an algorithm capable of detecting the sexual orientation of a person starting from images of that person’s face. The degree of correct forecast was 91% and 83% for men and women, respectively. The study is titled Deep Neural Networks Are More Accurate Than Humans at Detecting Sexual Orientation From Facial Images and is authored by Yilun Wang and Michal Kosinski, professors at Stanford University.
For years Kosinski has been doing research on how to predict personality based on data taken from the social networks, something he began studying for his thesis at the Psychometrics Centre at Cambridge University. He recently received wide attention for warning about the Cambridge Analytica scandal revealed after the 2016 United States presidential election.
In addition to moral and legal dilemmas, there are purely technical questions: to produce good answers, it is necessary that the original data be correct and pertinent. As professor Puerta reminds us, artificial intelligence lacks empathy, and thus the behavioral models that are generated lack a priori the social factor. In an accident involving a driverless car, who is responsible? The driver? The manufacturer? The programmer? Can an algorithm be held responsible for a catastrophe?
From banking to medical diagnosis: a range of possibilities
The future will depend on how we manage the infinite possibilities of artificial intelligence. Banking is one of the pioneering sectors in the use of predictive computer models. For years bankers ignored the alerts about their reckless home mortgage loans. When the housing bubble burst, it was shown that greed had led the managers to grant loans in spite of the high risk of default –something which computer models had been warning about, based on millions of pieces of data.
The automotive industry will also change from top to bottom. The vehicles of the future (and some in the present) will carry cameras to recognize the surroundings, determine the route to follow, and detect and avoid obstacles. And this is not the only application for artificial vision. “Instead of a human inspector, quality will be controlled by an algorithm vision system that analyzes images, processes them, and determines if the piece is correctly made or not and is able to be sold,” explains Javier Fernández, professor and director of the SIC (Systems and Control Engineering) research group at Universidad Europea.
The possible applications of artificial intelligence will revolutionize the health sector.
“We have a health system that is unsustainable because of an ageing population, budget restrictions, and because we are increasingly demanding… Technology plays an essential role in resolving these problems and is going to be necessary in the near future,” according to Mari Luz Morales, director of the Master in Digital Health at Universidad Europea, which is specifically designed for engineers who want to specialize in the health sector.
The possibility of making an automatic diagnosis –by processing millions of medical records– opens the door to a kind of automation of medical practice. It’s a question of “training” a neural network to process million of facts so that its analysis results in a predictive algorithm.
These algorithms based on deep learning are thus devised in two steps: a first training phase, in which the algorithm discovers the relationships between the data and the output variables (diagnosis); and another one for prediction, so that, faced with a new case, it will respond with a diagnosis. This is similar to what is applied by the Savana Med system, created by a Spaniard, Ignacio Medrano, and already present in dozens of hospitals.
The future of health care will also include nanomaterials. Arisbel Cerpa, a professor and researcher in the Department of Electromechanics and Materials in the School of Architecture, Engineering and Design at Universidad Europea, heads research about the application of carbon nanotubes to health. “These materials have revolutionized the scientific world because they have excellent mechanical and resistant properties,” she explains.
Cerpa has studied their application in biomedicine. For example: the nanotubes are an ideal contrast agent for obtaining clearer and sharper images in magnetic resonance scans. But that’s not the only use for these structures. At present the professor is directing a line of research at Universidad Europea about applying graphene oxide with metal for controlling microbes. She reveals that the results to date have been “marvelous” and adds: “Spain is in the forefront in biomedical applications.” Nanotechnology could even serve to accelerate the healing of wounds.
The future is yet to be written
The fact that we’ve been fantasizing about artificial intelligence for a century now adds another dimension –and a bit of déjà vu. We’ve seen so many films about intelligent robots that we’re hardly aware they have become a reality. Maybe the day will come when we’ll again have to classify movies, because 2001: A Space Odyssey or Blade Runner have ceased to be science fiction and have become historical documents.
HAL 9000 and the Nexus 6 replicants are right around the corner. If there’s anything clear from what we’ve imagined after so many books and films, it’s that the rewards that await us in the future are on a par with the threats. There are as many utopias as there are future dystopias. Although many of the health applications of artificial intelligence, robotics and nanotechnology are impossible to anticipate, they still depend on what we humans decide. If we’re capable of controlling this revolution and safely directing it, we’ll find a different future. And of not, it will also be a different future.