In 1950 the pioneer of the computer Alan Turing proposed a different path to develop a non -human intelligence: build a “mathematical child” or “child machine”, a device to which they were given The basic knowledge of an infant along with learning skills and then educate it.
However, seven decades later this innovative proposal remains impossible to carry out: Why does artificial intelligence cost so much the tasks that humans can do since childhood?
It is that, in the words of the American professor Melanie Mitchell, easy things are difficult. The activities that humans do without thinking much, such as looking at the world and making sense of what we see or Maintain an interesting conversation with a personturn out to be the most complex challenges for machines.
Thus, today, algorithms sets can be very good playing chess, but they cannot compete in “Tell it with mimicry” or “I see I see”. Achieving human -like intelligence is more difficult than we thinkbecause we are not aware of the complexity of our thought processes.
In fact, the artificial intelligence models with which we can dialogue were Trained with immense sets of informationthat far exceed one billion words, much more than a human being consumes in his entire life.
Boys and girls, meanwhile, They begin to speak with very few linguistic stimuli. When Turing suggested creating a “mathematical child” thought that perhaps he could “teach to speak and write” as happened in the classrooms of schools and simply simulate humans.
The challenge is still open but it seems that we are far from achieving it.
In 2023 the American philosopher Noam Chomsky said that mass language models such as ChatgPT had limitations, including the little relevance for linguistics since They learned “too well”even dominating the so -called impossible languages.
The impossible languages are those that are not governed by the rules and structures of natural languages around the world. They are usually created by taking a real language, dividing prayers randomly and investing the order of words in some of its halves. They do not serve to communicate but to experiment.
Artificial intelligence models were created in our image and likeness, which could explain why it processes better natural languages.
A group of researchers led by the specialist in language processing Julie Kallini decided to find out if Chomsky was right And he tried to learn a dozen impossible languages.
And it turned out that these models are not as omnipotent as it was believed, since he could not learn them as English or Mandarin.
The AI models were created in our image and likeness, which could explain why it processes better natural languages. Investigate How is that learning process It could help us discover the human ability to master new languages, especially being children.
Turing’s “mathematical child” It is still far awaybut maybe the artificial helps us to better understand the human.
