If we understood how babies develop mentally, this knowledge could help every child to reach their full potential. However, considering them simply self-learning machines, we will not achieve this goal.
On a lovely July day in 2005, Deb Roy and Rupal Patel drove up to their house. Sleepy faces shone with smiles: today Roy and Patel became parents. Stopping in the hallway for grandfather to take a picture, they chatted happily, hugging their precious firstborn.
This seemingly ordinary provincial couple was somewhat different from other families. Roy is an expert in artificial intelligence and robotics at the Massachusetts Institute of Technology (MIT), and Patel is a distinguished expert on speech impairments at nearby Northeastern University. A few years ago, they decided to put together the largest collection of family videos.
On the ceiling in the hallway were two subtle black dots, each about the size of a coin. Copies covered the entire living room and kitchen. There were twenty five of them in total: 14 microphones and 11 fisheye cameras. And this is just part of the system, which was planned to launch upon arrival from the hospital. Its purpose: to capture every movement of the baby.
It all started 10 years ago in Canada, although Roy started assembling his first robots back in the 70s in Winnipeg, when he was only 6 years old, and has not stopped since then. His hobby grew into work, and he became interested in android robots. I began to think about how to teach them to think and talk. “I thought it would be enough to dig into the literature to understand how this mechanism works in infants. And then I can create a learning system for robots,”Roy recalls.
Patel was pursuing a Ph. D. in speech correction at the time, and Roy once boasted to her at dinner that he had created a robot that could learn like babies. Roy was confident that if the robot perceives information in the same way as children, it will be able to learn and develop.
Toko's robot had an uncomplicated design and quirky appearance: a microphone and a camera were attached to a movable frame, and the image was complemented by eyes made of ping-pong balls, red feather bangs and a curved yellow beak. However, the robot was smart. Using algorithms for voice and image recognition, Roy taught Toko to highlight words and concepts in everyday speech. Previously, computers perceived language in digital form and saw only the connection of some words with others. Roy managed to create a machine that could relate words to images. Upon receiving the voice command, Toko was able to recognize the red ball among other objects and pick it up.
Patel had his own research laboratory in Toronto, and Roy went there hoping to better understand how babies learn. As he watched mothers play with their children, he realized that he had been ineffective in teaching Toko. In 2007, Roy told Wired magazine, “I did the wrong learning algorithm. When communicating with an 11-month-old baby, parents behave consistently and do not jump from one topic to another. For example, if the conversation is about a cup, all words and gestures in one way or another refer to it. And so on until the child's interest disappears and he switches to something else."
Promotional video:
Before that, when meeting a new object, Toko compared it with all the phonemes in his memory. Now Roy has changed the algorithm so that the machine gives more importance to newly acquired knowledge. After listening to Toko's tapes from Patel's lab, Roy was surprised to find that the robot's vocabulary was growing at an unprecedented rate. The dream of creating a robot that can evolve using what he saw and heard was closer than ever. But this required audio materials that were not easy to obtain.
No one has ever thoroughly studied the life of a child in the first years, the most important for learning. Typically, researchers conduct one hour of observation once a week: this is how Patel studied the communication between mothers and children in her laboratory. But in order to really understand how children learn language, you need to decide on something extraordinary: for example, place hidden cameras and microphones around the house.
I first heard about the Roy and Patel experiment when I was working as a school teacher in London. Most of my students entered school at the age of 11. They lagged behind in speech development, and I tried my best to help them catch up. In his research, Roy uses a scientific approach that makes all the methods I've tried seem hopelessly outdated. I would like to believe that his discoveries will help create a methodology that will help children reach their full potential. The hope has arisen that by creating machines that can learn in the same way as humans, we will be able to improve the mechanism of human learning.
Before starting, Roy and Patel laid down a few rules. Firstly, only a narrow circle of specialists, whom the family trusts most, will be able to view the records. Second, filming stops if a family member becomes uncomfortable with continuing the experiment. Third, the surveillance system can be temporarily turned off if one of them needs a little personal space. They didn't know what would come of it, but they decided it was worth a try. This experiment had every chance of shedding light on how children's brains work.
Deb Roy and his robot Toko are somewhat similar to Papa Carlo and Buratino. During the experiment, Roy tried to understand what robots can learn from children. I was interested in whether it is possible to develop a methodology that will help children learn more effectively, using the conclusions drawn from these frames from the family archive.
In 1995, two scientists, Betty Hart and Todd Risley, published a study of 42 Kansas City families to compare the development of children from poorer families to those of wealthier peers. The study lasted two and a half years and covered a developmental period from nine months to three years: scientists visited the family every week for an hour, recorded and transcribed the speech of children and their parents. The findings were disappointing. The more words a child under three hears, the higher his school performance at nine will be. There was a huge difference between the groups: Scientists calculated that by age four, the richest children heard 30 million more words than the poorest.
“The developmental gap between preschool children turned out to be a much more important and complex issue,” Hart and Risley argue. Their research showed that it is necessary to intervene in the development of the child as early as possible. "The longer you delay, the less chance you leave the child."
The solution seemed to be on the surface. Children do not have enough words - we will show more. Findings Hart and Risley launched a "word fever": all parents in English-speaking countries rushed to buy flashcards with words and other toys for early development for their babies.
From my experience in school, this interpretation seems a bit simplistic. Teaching children should not be equated with entering data into a computer: the number of words heard is not the only factor in the development of mental abilities.
My opinion is shared by Professor Katie Hirsch-Pasek, who studies preschool development at Temple University in Pennsylvania. According to her, “everyone knows that the fast food industry feeds us empty calories - the same way the learning industry works. Memorization of information is not the only component of successful learning and a happy life.” Moreover, Hirsch-Pasek is the author of the popular book “Einstein Learned no cards,”in which she outlined her thoughts on“word fever.”Perhaps this book will help to understand why children need to play more and remember less.
Hirsch-Pasek is one of the most influential experts in early childhood development, the author of 12 books and hundreds of academic articles, and she also founded Temple's Infant and Child Laboratory, whose motto is: "Children teach adults here."
In the laboratory, scientists test what little people are capable of. Researchers have developed a unique technique: they measure the heart rate to find out what babies understand at the age of eight months. “They know that the toys hanging over the bed will not fall on them,” says Hirsch-Pasek. “They know that if you put a plate on the table, it won't fail. This is amazing. They understand that the lower part of the body does not disappear anywhere, even if the person is sitting at the table and cannot be seen."
Until recently, scientists believed that babies' thinking was irrational, illogical, and self-centered. In 1890, in his book Principles of Psychology, William James described sensory overload in infants: "A toddler's eyes, ears, nose, skin, and intestines experience the world as one booming, muddy mess." This discovery gave rise to a mechanistic view of learning: the constant repetition of words was believed to be the most important ingredient in successful learning. However, it is not.
Children begin to learn already in the womb. At this stage, they learn to recognize sounds. The child can distinguish the voice of the mother from the voice of another person within an hour after birth. The brain of a newborn is well adapted to learning with the help of the senses. All people are by nature researchers, ready to make scientific discoveries. Only when we understand this thought, we realize how great our ability to learn.
“From birth, we are able to 'perceive precise external stimuli,' says Hirsch-Pasek. I remembered the robot Toko, which can read environmental signals: it receives data from cameras and microphones that replace its eyes and ears. However, robots are limited in their perception: they are programmed to perceive and use only the signals selected by a person for their training, so the range of their experience and behavior is limited. Such training is completely unsuitable for people. Babies learn through communication.
“We are naturally ready to interact with other people and our culture,” says Hirsch-Pasek. The peculiarity of children is not that they learn by studying the environment. Baby animals are also capable of this, but unlike them, children learn to understand the people around them and their intentions.
In the process of evolution, we have developed socio-cultural ways of transmitting information. This became possible thanks to language, the ability of two creatures to ascribe a common meaning to a certain abstract concept or symbol. How can we not notice the beginnings of communication in the behavior of even the smallest children? Babies under the age of one year enter into a dialogue with their caregivers in proto-language. They mumble, maintain eye contact, exchange objects with them, and imitate their actions and facial expressions. They experiment with various objects: they put them in their mouths or hit them on other things.
Michael Tomasello, professor at the Max Planck Institute for Evolutionary Anthropology in Leipzig, wrote that babies learn "in an environment where new objects and situations are constantly emerging, and at any given moment this environment resembles the collective experience of an entire social group throughout the development of its culture."
Each of us will unleash our learning potential only when we understand how we can create such an environment. The human brain is specially designed for learning. A long period of human immaturity is a risky evolutionary strategy, in which we are at an early stage vulnerable to predators and diseases, and the ability to reproduce appears after many years. But in the end it fully justifies itself due to the ability to assimilate gigantic volumes of the latest information coming from the environment or from a social group.
Scientists have long recognized that the discussion about the role of upbringing and natural factors in the formation of a person is meaningless. A significant part of our brain development occurs in the first three years of life. During this time, the brain is formed in interaction with the environment and in connection with sensory perception. As Hart and Risley's study of the difference in the number of words heard has shown, perception can have a profound effect on the kind of person a child becomes.
Evolution has given us everything we need to learn and teach. Our ability to understand other people arises at nine months, when the child begins to draw the attention of others to the objects that he is pointing at or holding in his hands. In a year, children can follow the attention: watch, listen to, or touch what other people are watching, listening to and touching. At 15 months, children can direct their own attention: “Listen to that! Look at it! For meaningful human learning, it is necessary that the teacher and the learner direct their attention to the same object. This is why children cannot learn to speak through video, audio, or listening to parent conversations. We developed differently. This is why it is important to talk to children. And that's why we can't learn from robots. At least for now.
The conclusion suggests itself: each generation must do everything so that the next one at the earliest stages will master the entire toolkit, all the symbols and social practices of the current culture.
In search of an educational environment that could help develop our inclinations, I traveled to Corby, Northamptonshire, to visit the Pengreen Early Childhood Center, which specializes in early childhood development. It is cold and gloomy on the site near the center, but this does not frighten the children. Next to a bamboo bush, two little boys splash water from an open tap. "Don't urinate me!" they shriek happily. The teacher calms the kid in a T-shirt with the words "Hurry up, otherwise you will lose." Four girls are enthusiastically engaged in a dialogue, automatically pouring sand into colorful buckets.
The Pengreen Center is world renowned for its achievements in early childhood development and family support. The work of the center became the basis for the creation of government projects in the field of early development in Britain, in particular, the "Sure Start" and "Early Success" programs. I spoke with the director of the center, Angela Prodger. She took office recently, succeeding Margie Whalley, who founded the center in 1983. In the 80s Corby was one of the poorest cities in Great Britain. After the closure of steel mills, the number of migrants from Scotland who moved to the south for work has dropped significantly: the city has lost about 11 thousand people. The center was conceived as a lifeline for the next generation. Today it serves 1,400 of the poorest families in the UK.
I asked Prodger about how children learn language. We already know that one cannot do without words, but on the platform of conversations something is not heard. “If the first step is not to address the issue of personal, social and emotional development, it will be early to start learning,” she said. The director of the center explained that before children have the opportunity to master the tools of language and speech, it is necessary to make sure that they feel “present and connected”. In her opinion, we often overlook this when developing approaches to early learning. I thought it would be nice to do this, but not necessarily, but research suggests otherwise.
In the 1950s, British psychoanalyst John Bowlby put forward the theory of "attachment." He suggested that children who cannot control their feelings are more likely to get upset when they feel hungry, sad, or lonely. A caregiver is needed to help children "regulate" their feelings. If the child discovers effective techniques, over time he will begin to use them independently. But if the child's negative experiences are not smoothed out due to parental love, they can subsequently take root.
Growing up in a poor family or traumatic environment has enormous consequences for children. This is why Pen Green considers it important to prioritize “presence and involvement”. This idea also explains the behavior of some of the children in the school where I taught. I did not notice that children somehow reacted in a special way to the stressful environment in which they grow up. But at Pen Green, they work closely with educators to ensure the creation of strong, caring relationships that will help children develop dynamically, first in kindergarten and then at school. I always thought that you don't feed the children with bread - let me do something wrong. It never occurred to me that the environment placed them in such conditions, that they could not behave otherwise. “Through their behavior, the child is always trying to communicate something,” said Prodger.
As we walked around the center, Prodger told me that the specialists at Pen Green learn to keep track of what is going on in the heads of children and interpret the behavior of toddlers as a possible signal even before the child starts using words. “The kids are in constant contact with us,” Prodger told me. “We just need to learn to understand them. You need to be able to observe. Be aware that children are interested in what they are trying to learn."
Creative games are the basis for the formation of creative thinking, success in language, mathematics and science. If it's too early to start working with cards, you can miss this stage of development. “You need to feel freedom, be able to take risks,” says Prodger.
Several times a week, the center's specialists take the children to the forest. There, they light bonfires, experiment with scissors and ride BMX bikes. I wanted to walk - they go for a walk. I wanted to return to a secluded place where you can lie around - they are returning. Learning is dictated by the environment. Adults are only trying to establish a connection with children and understand where their attention is directed. Reading and writing can wait. Educators should be as sociable as possible and follow the example of children during the game. Before children start learning, we must make sure they don't feel like strangers.
The children in the center look happy, they learn to behave in a group and through play they form the basis for future success. And yet I wondered if something else could be done to accelerate early learning. As a result of the experiment with the robot, Deb Roy came to the conclusion that every minute counts. Can we afford to leave so much to chance?
“Having a baby is the greatest source of inequality in the United States,” wrote economist James Heckman. This is equally true in today's Great Britain, where academic success is most often determined by the level of parental earnings. While two-thirds of children consistently score “fair” or higher in school leaving exams in English and math, only a third of them are from poorer families. Heckman also showed that the most effective solution to inequality is to start developing children as early as possible. Changing schools is not enough: change is needed even earlier.
Temple University's Hirsch-Pasek explained that you can't just sit the kids in front of the tablets and wait for them to learn. However, this does not mean that you cannot use smart devices. Some laboratory experiments conducted by Hirsch-Pasek are aimed at reducing the developmental gap between rich and poor children. Others relate to the development of language skills and spatial thinking. All experiments use technology in one form or another. “Computers, for all their skills, are bad conversationalists,” Hirsch-Pasek says. - They do not strive for communication. They interact, but they don't adjust."
Hirsch-Parsek's goal is to fundamentally change the way children are educated, especially the poorest. “We thought it was extremely important to give the children from poor families the basics,” the scientist continues. - We wanted to remove the changes, although we understood that physical activity helps children learn, develops the brain. We also planned to leave only reading and mathematics, and remove the humanities and all unnecessary subjects like social studies."
It was not easy for her. Politicians and amateurs have adapted science to suit their needs. No scientist believes in the effectiveness of cards. No scientist believes that you need to start teaching a child to read and write as early as possible. These are all inventions of the government. Recent research has added depth to Hart and Risley's observations of language teaching to children in Kansas. In 2003, psychologist Patricia Kuhl conducted an experimental Chinese teaching program for American toddlers. The children were divided into three groups: one learned from video, the other from audio, and the third with a flesh-and-blood teacher. Only those who studied with a live teacher managed to remember something. In 2010, scientists conducted a study on a series of incredibly popular educational DVDs Baby Einstein ("Little Einstein"), which Time magazine called "a drug for children."The study found that children who watched these films "showed no difference in understanding of words compared to children who never watched them." Children did not memorize the words and listening to the conversations of their parents or the program In Our Time ("In our time") on Radio 4, no matter how soothing and pleasant the voice of the announcer was. To learn a language, words are not enough for a child; the presence of a person is necessary. Children cannot learn just by looking at the screen.just looking at the screen.just looking at the screen.
Schools still do not pay attention to these details. Erika Christakis, parenting specialist and author of The Importance of Being Little, notes the simplification of the preschool curriculum from a versatile, idea-based approach to a two-sided, naming-based approach. Daphne Bassock from the University of Virginia wonders if it's true that kindergarten is now equated to first grade. It is widely believed that after kindergarten, that is, at the age of 5-6, a child can already read. There is no evidence of this. Scientists from Cambridge compared two groups of children: some began to learn to read and write at age five, others at seven. By the time they were eleven, there was no difference in reading ability, “however, children who were taught two years earlierdeveloped a less positive attitude towards reading, and they understood the text worse than the children from the second group."
The conclusion is obvious: if you start teaching the decoding of letters before the child understands how the story works, and learns to relate it to his experience, feelings and emotions, then the reader will turn out worse from it. In addition, this activity will be less enjoyable. If, in the early stages of development, treat a child like a robot, then he will forever lose interest in learning.
And Hirsch-Pasek wants children to enjoy learning and growing up. Besides children, she loves music. It's normal for her to start singing out of the blue, especially when she and her granddaughter are talking on the phone.
In her book, she proposed the concept of the six pillars of modern learning: trust, communication, collaboration, interesting materials, critical thinking and creative ideas. It seems to be well-known truths, however, unlike modern educational policy, they are supported by scientific evidence: "If I were asked to describe everything in one phrase, I would say that" since ancient times we have been learning from people."
Understanding of this very thought prompted one married couple to press the "record" button.
When we met at MIT, Deb Roy was dressed in black and still looked pretty young. Light gray hair served as the only evidence of 11 years of parenting. In hindsight, the Human Speechome Project (as Deb Roy and Rupal Patel called their project) seems like a quirk amid the widespread passion for artificial intelligence at the turn of the last millennium. In total, they recorded about 90,000 hours of video and 140,000 hours of audio. Records of 85% of the first three years of their son's life and 1.5 years of their youngest daughter's life take up 200 TB. But now all the materials are gathering dust on the shelf. “I don’t wash them,” says Roy. - I'm waiting for my son's wedding just to get everyone with these records.
In a way, it's also another big lost family video. Together with colleagues from MIT, Roy developed new approaches for visualizing and processing the received data. The graph of "social centers" is two lines, at the intersection of which those moments are marked when the child and one of the parents communicated, learned or explored something. On the graph of "vocabulary landscapes" there are mountain-like broken lines, the highest points of which indicate the places (in the living room and in the kitchen) where a certain word was pronounced most often. These solutions have proven to be extremely useful in analyzing Twitter communications. Roy and one of the graduate students spent 10 years building their company.
Roy is now back at MIT. He is now director of the Social Machines Lab. The scientist gave up trying to create robots that could compete with humans and instead focused on improving the learning of children. It was the upbringing of his own child that made him change the direction of his research.
His son said something conscious for the first time when they looked at the paintings. “He said ly,” explains Roy, “clearly referring to the image of a fish on the wall: we were both looking at it. It was definitely not a coincidence, because he immediately turned to me, and he had such an expression on his face as in cartoons, when the light overhead lights up, and he is like: "Oh, that's it." That's how he looked at me. He was not even a year old, but he was already aware of himself and the objects around him."
“I think this whole AI work has taught me a lesson in humility,” Roy continues. "I realized that you can't just take and solve this problem."
Roy no longer believes that it is possible (or necessary) to educate a robot as a living person. There is little value in developing a robot that would take a childhood-like period to evolve into a replica of an adult. This is how people develop. Not to mention imagination and emotion, individuality and love beyond Toko's reach. Observing his son, Roy was incredibly surprised by the "unimaginable complexity of the process of mastering the language and all the actions of the child studying it." Children do not just mechanically repeat what they have learned, they create, find new uses for words, and share emotions.
The learning process is not like decoding signals, as the scientist thought at the beginning, but a much more complex process, continuous and interactive. Roy read Helen Keller's autobiography to children and was amazed at her impressions of the first understanding of the language. After an illness in infancy, Helen lost her hearing and vision, but at the age of 7 she had an insight. “Suddenly there came a vague realization that I had forgotten something,” she wrote, “followed by delight at the returning thought. And then, by some miracle, the mysteries of the language were revealed to me. I learned that "v-o-d-a" means that pleasant cool something that flows down my hand. A living word awakened my soul, lit a light in it, gave hope and joy, liberated it. Everything has a name, and each name gives rise to a new thought. When we returned home, all the objects that I touched felt as if they were alive."
Roy recently started collaborating with Hirsch-Pasek. He liked her idea that machines can improve the learning process of a person, but they will never replace it.
He realized that a person can learn only in society, interacting with other people. For a robot, language acquisition is abstract and based on identifying patterns. We have it innate, individual, filled with emotions and life. The future of intelligence is not in making intelligent machines, but in developing our own minds.
Alex Beard
