Artificial intelligence does not understand Hemingway, but personalizes the English class

Is it easier to learn a language online, in front of a machine, than when you have a real teacher in front of you? "If you had asked me a year ago, I would have answered that it is clearly better to have a person in front ...", responds to INNOVADORES from London Brian O'Sullivan, head of evaluation, research and development of the British Council, the great teaching institute of the English language, homologous of the Spanish Instituto Cervantes.

But the answer itself already indicates a change of opinion "after having worked with some technology companies, especially in China, and after evaluating a study on the efficiency of a British university, the evidence suggests that the average learns faster with its app".

O'Sullivan has immersed himself in the world of artificial intelligence applied to the teaching of languages and fights a certain skepticism with the facts. And vice versa. "The day a computer is able to write as Gabriel García Márquez, I will raise my hands and say 'I was wrong'," he confesses, but underlining the great change that comes "like a tsunami".

"Neural networks are used to go more towards teaching with personalized follow-up and diagnosis," he explains. "It's not just saying 'you've done a good job' or 'you've made mistakes' but indicating where to find more information. The system is learning from you all the time, on a personal level." A report from Stanford University notes that the The main objective of education with AI is personalization, the question is not whether it will happen, but when.

Language learning is based on the student's own interests. "The AI ??is guided by the personal information that the student introduces, and then by the decisions makes when interacting with the system: what kind of texts reads, what subjects, the level of difficulty of those texts and how fast they read them; how he reacts on them with other students in the minichat ... It continually observes where his interest leads him to accelerate it more, it is almost like the Big Brother. " Even the student can define his program: "I can create my own learning with the machine, to write an article for an engineering publication: what things should I know, what is the writing style, how the pronoun is used in the first person, 'I'? ... In engineering, I should never use it, in social sciences, always".

O'Sullivan also cites examples in general education: "There is a company linked to Dublin's Trinity College, Learnosity, which produces apps for the Irish educational system, such as personalization in geography learning, the most attractive being its scale. Applications are expensive to develop and distribute and it is not easy to introduce them into the formal education system, the AI ??will have an impact that it does not yet have". On the other side of the scale, questions arise about the ethics of using an intelligence whose decisions are made without knowing very well how and why, in a 'black box'. And observe deficiencies of the machine with the language. "In terms of structure, vocabulary and grammar, the machine is the simplest to teach and evaluate. In what is not so good is to understand the contents. Or the style. For example, if I put Hemingway's novels on an AI-controlled test, it will fail, because it uses very short phrases as a writing style. It says: 'Barry entered the room. He sat. He turned on the computer '... It is a sequence of very simple sentences, they seem like a very low level speaker. It's to get a dramatic effect and the machine does not understand that drama".

"In self-learning systems without human supervision, the machine decides on the quality of language, with some controversy," he adds. "Algorithms are developed to dig the language on the Internet, comparing what people are saying and doing, unfortunately it is not the most wonderful place in the use of language and machines are scratching very poor uses of the language and also very poor ideas. to form his judgment, it's one of the biggest ethical problems we have. "