Intelligent Machines That Learn Like Children
Machines that learn like children provide deep insights into how the mind and body act together to bootstrap knowledge and skills
Scientific American, 2018
Robots that learn like humans and program themselves: Welcome to the brave new world of artificial intelligence.
Read or listen offlineAmazon Kindle
Artificial intelligence is the decade’s new buzz word. Rapid improvements in computer science and machine learning allow scientists to create robots that have more and more human-like features. One of the newest and most intriguing developments are robots that teach themselves to mimic human development. At the same time these robots allow scientists to better understand the complicated mind and body interactions that bring about human learning, as well as learning deficits. This article will engage and inspire anyone who wants to explore the brave new world of artificial intelligence.
- Robots that teach themselves mimic human development.
- A fundamental ingredient of human learning is the trial and error loop.
- Possessing and using a physical body accelerates learning.
- Curiosity and internal motivation make each human and robot unique.
- Robots with human-like intelligence do not exist yet because crucial ingredients such as the influence of caregiving are missing.
Robots that teach themselves mimic human development.
Scientists have created robots that teach themselves to mimic human development. These robots are programmed to acquire knowledge and skills the way a human child would.
“Since the beginning of the 21st century, roboticists, neuroscientists and psychologists have been exploring ways to build machines that mimic such spontaneous development.”
At the same time the “brains” inside the robots can be examined in order to shed light on how exactly the learning came about and to better understand how humans acquire knowledge.
A fundamental ingredient of human learning is the trial and error loop.
The human senses continuously transmit information about the world to the brain. In turn, the brain sends signals to the senses, anticipating what the senses will perceive. These sets of signals interact and create a “prediction error”: The difference between what a person expects and what she or he experiences. Prediction error is a fundamental ingredient of human learning. It allows people to update predictions about the world and thereby create more and more accurate representations of the world. Recreating this trial and error loop in robots may help scientists understand disorders such as autism and Attention Deficit Hyperactivity Disorder (ADHD). In addition to the prediction error’s value for the individual (human or robot), it may play a role during social interactions. Altruistic behavior may result from an attempt to help others close the gap between expected and experienced action outcomes.
Possessing and using a physical body accelerates learning.
Using a physical body makes it easier to learn. For instance, employing the fingers for counting, instead of just abstractly representing numbers, speeds up the development of numeric skills in toddlers as well as robots. Examining the neural networks of “finger counting” robots showed a more accurate representation of numbers compared to when robots merely referred to the numbers by name.
Curiosity and internal motivation make each human and robot unique.
Going through the trial and error loop and arriving at more accurate predictions about the world is rewarding. Getting a reward in turn motivates people to keep learning.
“Remarkably, even though the robots went through similar stages of training, chance played a role in what they learned. Some explored a bit less, others a bit more – and they ended up knowing different things.”
Robots that are programmed to display such internal motivation seek out different opportunities for learning and end up with a unique set of knowledge and skills, much like humans.
Robots with human-like intelligence do not exist yet because crucial ingredients such as the influence of caregiving are missing.
Despite advances in artificial intelligence, self-taught robots do not possess human-like intelligence yet. For one, the human brain is complicated and cannot be adequately represented in robots at this point. Secondly, crucial environmental influences that shape human development cannot yet be integrated into robot learning. One example is caregiving. A robot would have to be raised like a child to truly mimic human development.
“Intelligence does not merely require the right machinery and circuitry. A long line of research has shown that caregivers are crucial to children’s development.”
Finally, human learning is incremental. What a child learns one day is the basis for more learning the next. For this reason, a robot that learns everything at the same time might not be able to acquire human-like intelligence.
About the Author
Diana Kwon is a freelance journalist who covers health and the life sciences. She is based in Berlin.
This document is restricted to personal use only.
Did you like this summary?Read the article
In our Journal
4 months ago
The Four Horsepersons of the AI Apocalypse: Part One
Why is success with Artificial Intelligence (AI) so elusive, and what can you do about it? In my three decades of working with data, mathematics, computers, and the digital Bermuda Triangle that exists between them, I have come to appreciate many of the difficulties we humans face in putting technology to work. This is particularly […]
Comment on this summary