American mathematicians have created a machine learning system that can "decode" the cries and cries of babies and understand what they want from their parents or nannies. Their findings were published in the Journal of Automatica Sinica.
In recent years, thanks to the development of mathematics and the growth of computing power of computers, scientists have the opportunity to create complex neural networks, artificial intelligence systems capable of performing non-trivial tasks and even "thinking" creatively, creating new examples of art and technology.
For example, in the last two years alone, scientists have created AI that can beat a person in the "uncountable" ancient Chinese game of Go, while learning from scratch to find the most important events in history from newspapers, write scripts for computer games and color photographs and videos under Van Gogh and paint your own pictures.
Early last year, scientists unveiled an AI system that can distinguish moles from skin cancer better than most experienced dermatologists can. A little earlier, Yandex programmers created neural networks that recorded music albums in the style of Nirvana and Civil Defense and painted pictures in the style of Wassily Kandinsky.
Liu and his colleagues adapted artificial intelligence to solve the most exotic problem so far - they created an algorithm that can very accurately classify the cries of babies, and used it to create a kind of "machine translation" system.
Many parents and experienced pediatricians, as scientists note, often notice that the crying pattern of babies can vary greatly depending on what exactly they are missing or what annoys them at a given time.
Liu and his team suggested that these signals obey the same rules as articulate human speech, using specific sets of sounds to denote specific phenomena and concepts.
Guided by this idea, the scientists created a kind of analogue of a speech recognition program that analyzed recordings of crying of different children made in the same situations. She tried to highlight in them a certain set of common features that form the basis for the peculiar "words" of this infant language.
Promotional video:
Using this algorithm, the scientists traveled to a maternity hospital in Illinois and began recording and analyzing the screams of babies, using experienced nannies and doctors to determine exactly what the babies wanted.
After analyzing several hundred such audio recordings, scientists have identified about a dozen signals that babies emitted when they wanted their parents to feed them, change their diaper or diaper, want to sleep or attract their attention for other reasons.
As mathematicians note, the structure of these signals, a kind of "words" of the infant language, was the same for all babies, despite the great differences in the nature of their cry. This means that machine learning systems can be used to decipher these signals without tailoring them to each individual child.
In the near future, scientists plan to test the work of their program on a wider set of children. In addition, they will compare whether there are differences in "infant language" between representatives of different ethnic groups, and also use more advanced versions of artificial intelligence to analyze the recordings and expanded "transcribing" of these screams.
