In the late 1950s, computer engineer Arthur Samuel created a program to play checkers, using a simple algorithm to discover the best moves to win. Samuel trained the computer with a copy of himself the self-play and with a database in which hundreds of parties were registered. It was the beginning of machine learning a branch of artificial intelligence (AI) that allows machines to learn without being explicitly programmed. Almost seven decades after that game, that technology has applications as diverse as the diagnosis of cancer or the construction of autonomous cars.

Technology that teaches

It is a technology applicable to virtually every field where data is available Thomas Dietetic one of the fathers of machine learning as a research field. The expert mentions examples ranging from algorithms used in the business world to identify potential buyers of a product to the systems used by governments to solve problems in infrastructure such as highways and hydro. Other closer examples are automatic translation systems on Skype, face recognition of mobile cameras and virtual assistants, bets from companies like Google and Microsoft to bring the technology to the end user.

Microsoft’s virtual assistant has 145 million users and the company aims to develop it to the point where it communicates directly with other IAs to offer the user any kind of information or service, from the purchase of a shoe to the Delivery of a pizza at home leader of the company’s Technical Evangelism team. Our goal is to democratize access to machine learning he says. The focus in this regard is the automated learning platform in Azure, a cloud analysis service that allows creating and implementing machine models according to the needs of each user.

Google focuses its strategy on Tensor Flow, a store of experiences and results of experiments that it uses to make its applications make better decisions, and that has data open since 2015. However, Google wants more: We are working on robots that can take care of dangerous situations and reach places that humans cannot reach like in the Fukushima nuclear power plant says Andres Leonardo Martinez, computer engineer of the company.

Risks and errors

In addition to making projections for the future, experts also wonder what are the risks of a world in which robots adapt and learn from experience (such as humans). They discard, yes, a scenario of science fiction where the machines annihilate humanity. “We create and program computers because they allow us to do things better. I imagine a future in which a person and an AI system work together as a team. In virtually every field, the combination of robots and people is more powerful.

The researcher sees at least two important roles for humans in the future: performing tasks that require empathy and deep understanding of another human being and ensuring that robots do not make mistakes. High-risk decision-making problems often involve unique factors. Automatic learning only works well on stable problems, when the world is highly predictable and it is easy to collect large amounts of training data. In problems where every situation is unique, it is unlikely that technology will succeed he explains. The expert defends the creation of a regulation that determines security tests and a specific certification to mitigate those risks. Sebastian Farquhar, a researcher at the Institute for the Future of Humanity, thinks it is too early for that.