Portuguese researcher Pedro Domingos has written The Master Algorithm, a book that looks at the ways in which artificial intelligence is a part of our lives. Computers learn, but we are the ones who have to make ethical choices. Bill Gates has read the book and recommends it.
It is easier to produce emotional behaviour than intelligent behaviour. It is easier to make someone laugh, cry or love than it is to dust the house or make a meal. Over the last 50 years, artificial intelligence has developed in such a way that it has altered our perception of time and space, changed the job market, found cures for different diseases and recreated the world. However, it is still very far from human intelligence. “The idea that computers will quickly become more intelligent than us is a bit false,” states 51-year-old Portuguese Pedro Domingos, professor at the University of Washington in Seattle, researcher in the field of machine learning and author of the book The Master Algorithm: How the quest for the ultimate learning machine will remake our world, which Bill Gates, founder of Microsoft, included in his 2015 required reading list and ultimately recommended.
The book explains, in a clear manner, how we are constantly being observed by algorithms that are able to predict behaviour, satisfy desires, suggest films for us, choose jobs and guide us without our being aware that we are not making choices, but rather they are for us. In The Master Algorithm, Pedro Domingos takes us inside the brain of companies like Google, Facebook and Amazon so as to explain how it all works and how to behave in relation to this master algorithm or, as he says, to the “key that opens all doors”. “Nowadays, artificial intelligence is a vast field, and within it the area with the greatest impact today is machine learning,” points out Pedro, adding that “people come into contact with machine learning every day in both their professional and personal life but are not aware of it.” “One of the reasons I wrote this book was to make people more aware. A common reaction among people is to think that this is science fiction, but it’s present in everyday life.”
A driverless car making its way along city streets is not science fiction; mobile phones that recognise a certain voice and understand spoken commands are no longer science fiction; having Facebook or Netflix use machine learning to choose or recommend pages or programmes is a current reality. And so too is the use of machine learning by companies to choose job candidates, or by the police to select suspects that should be investigated,” says Pedro Domingos, nevertheless stating that the image of a domestic robot making beds continues to be science fiction. “Machine learning takes advantage of the data that we constantly generate, and the computer – by means of the learning algorithm – learns to do what we want all on its own. If I give the algorithm chess games, it learns; if I give it an X-ray and the diagnostics to see where the tumours are, it also learns. We, ordinary users, are continuously programming computers simply by the way we use them. We need to be aware that every time we interact with all these applications, we are using them for our immediate objectives on the one hand. However, we are also teaching them what we want and who we are. This then comes back to us in the form of suggestions and action configurations.”
And it is precisely here that a concept comes into play which is almost always associated to this type of infinite circuit, i.e. manipulation. On this topic, Pedro Domingos does not hesitate to say that “in a way we are being manipulated”, but that this shouldn’t be viewed in a fatalistic manner. “These companies provide us a service and if they no longer serve us, we will ultimately stop using them. In other words, the interests of these companies and our interests are aligned, but only up to a point. At a certain point they start to diverge. Learning is a type of taxi that takes us where we want to go, on the Internet and even in the real world. But we have to tell the taxi where we want to go. People need to demand that of companies. Nowadays, this interface doesn’t exist and not because machine learning doesn’t allow it – because it does – but rather because companies have sought not to do things that way. On the one hand, because of a question of control; on the other hand, to avoid controversy. I think this has to change,” says Pedro, bringing ethics into the conversation, since many of these things take place at an almost unconscious level. “We have to decide how we want this technology to be used, how we can use it to our benefit and not so much in benefit of Google or Facebook or whoever it is that’s making these choices for us.”
The principle of uncertainty
Pedro Domingos began writing his book towards the end of 2012, and less than three years later it had already hit the bookshelves. Many things evolved during that time, but the base changed little. “I had to rewrite very little,” he tells us. “My main concern was to write a book that set out the perpetual ideas of machine learning. Ideas are popping up all the time, but there are also ideas that appeared decades ago. The key learning ideas are the same.” The book would not become outdated, for example, if the master algorithm were to arrive at a cure for cancer, which continues to be difficult because cancer, as he says, is not exactly a disease. “There are areas where learning algorithms are better at making a diagnosis than doctors. Deciding if someone has malaria, the flu or a tumour here or there… The algorithm is able to detect lung or breast cancer sometimes months before a pathologist. This can save a life. But the problem with cancer is that it is an extremely difficult disease to cure because it isn’t a single disease. Cancers are all different, each patient has a different cancer and the same cancer mutates over time,” Pedro says, adding that there will never be one sole drug that can cure cancer. “Cancer doesn’t have a cure. The cancer cure programme is a learning programme that learns the cancer phenomenon model from patients. When faced with a new patient, and based on the genome of this patient and on the mutations of the particular cancer, as well as all the relevant factors, it predicts the right medicine for that cancer, for that patient, capable of killing cancer cells without killing healthy cells.” This is difficult with machine learning, but it would be almost impossible without it, he contends. Why? Because “the amount of knowledge involved is much greater than what any human being is able to master.” “With machine learning “we’ll get there”. When? That’s the million-dollar-question. “I think that with the research that’s taking place nowadays, it’s quite likely that we’ll have such a programme within a decade.”
The Master Algorithm is about all of this. Science, medicine, the domestic world, ethics, philosophy, everyday life, what we can and must know every time we interact with a computer and to what extent lawmakers can intervene without jeopardising progress, as well as the degree of uncertainty this all entails, every time the topic revolves around machine learning. “This is a key issue in learning. All knowledge that is induced from data is always uncertain,” says Pedro, going on to explain that “There are two types of reasoning, deductive and inductive. Deductive reasoning says, ‘Socrates is a man; all men are mortal, therefore Socrates is mortal’. There is no uncertainty. But the learning we’re carrying out is the opposite. It’s ‘Socrates is human and I also know that Socrates is mortal.’ What do I need to know in order to induce? I have to induce the rule that human beings are mortal. I’m going to induce that all human beings are mortal, but I may be wrong. There might be human beings that I haven’t seen who are immortal. We always have to carry out this uncertainty control so as to be somewhat certain that we’re learning real knowledge instead of learning hallucinations.” So what is there to praise and fear in this area? “I don’t think that we should be afraid that machines will take control of the planet and dominate or exterminate humanity. It’s a common idea in science fiction films, but not so much in reality. Since human intelligence is the only intelligence we know, we tend to draw an analogy between artificial and human intelligence, but artificial intelligence is very different from the human one. Machines have no desires, no goals of their own, no personality, no conscience… The notion of good and evil does not exist for a machine. We tell a machine what is good and what is bad. What we need to worry about and know is who is controlling the machine and who is telling the machine what the objectives are.”
by Isabel Lucas
web design & development 262media.com