Artificial Intelligence: Teaching Machines to Learn Like Children

We’ve experienced startling successes with artificial intelligence over the last decade, but many researchers view the current state of the field as having plateaued. Part of the problem is that the current AI algorithms require massive amounts of data for training. That data can be hard to acquire and the learning period can take a very long time to train artificial neural network models.

Lorijn Zaadnoordijk, researcher at Trinity College, said that “progress is stalling in many areas because the datasets that machines learn from must be painstakingly curated by humans. But we know that learning can be done much more efficiently because infants don’t learn this way. They learn by experiencing the world, sometimes by seeing something just once.”

Does the current approach to how AI works need to change? Alan Turing, mathematician and computer scientist, wrote in a paper that “instead of trying to produce a program to simulate the adult mind. Why not rather try to produce one which simulates the child’s?”

Tarek R. Besold, researcher at TU Eindhoven, said that “as AI researchers, we often draw metaphorical parallels between our systems and the mental development of human babies and children. It is high time to take these analogies more seriously and look at the rich knowledge of infant development from psychology and neuroscience, which may help us overcome the most pressing limitations of machine learning.”

Alison Gopnik, professor at UC Berkeley, said that “even with a lot of supervised data, AIs can’t make the same kinds of generalizations that human children can. Their knowledge is much narrower and more limited, and they are easily fooled. Current AIs are like children with super-helicopter-tiger moms—programs that hover over the learner dictating whether it is right or wrong at every step. The helicoptered AI children can be very good at learning to do specific things well, but they fall apart when it comes to resilience and creativity. A small change in the learning problem means that they have to start all over again.”

Yann LeCun, chief AI scientist at Facebook, said that “most of what we learn as humans and most of what animals learn is in a self-supervised mode, not a reinforcement mode. It’s basically observing the world and interacting with it a little bit, mostly by observation in a test-independent way. This is the type of learning that we don’t know how to reproduce with machines.”