A short while ago someone sent me the link to an article titled Teaching Me Softly… How intriguing, I thought. As I read on, I became aware of the connotations in the title, which piqued my curiosity even more. The subtitle read: How machine learning is teaching us the secret to teaching. So, this article promised to give me new insights into teaching and learning from recent studies in artificial intelligence (AI). I couldn’t wait!
Artificial intelligence uses humans’ intelligence as a model for building machines with a mind of their own. The resulting intelligent machines can carry out complex tasks as well or even better than humans. A famous example is Deep Blue, the computer that plays chess so well that it beat the world chess champion Garry Kasparov. Since the advent of Deep Blue, computer science came a very long way, and the field of AI followed along closely. Present-day uses of intelligent machines are too many to count – and expanding – while new ways of simulating human brain processes are being explored. The robots are here…
Machine intelligence mirrors human intelligence. The artificial neural network used by an intelligent machine functions similarly to our brain’s neural network. And like our brain, an intelligent machine learns and adapts by reconfiguring its neural network as a result of new experiences and information. If machines learn like people, what can they teach us about the learning process? More importantly, can we improve our own learning using some of the lessons gleaned from working with intelligent machines?
For people as much as for machines, what makes learning effective is access to “privileged information“, the author of the above article says. What does privileged information mean, in relation to learning? The concept seemed to me slippery at first. “Privileged information encodes knowledge derived from experience”, the author explained later. And it makes the difference between brute force learning and smart learning. Aha!
Machines are becoming smarter by learning smarter. New algorithms give machines the ability to interpret unfamiliar situations, to solve problems, and to learn from their mistakes. Robots that were programmed to follow rules which had no meaning to them are now evolving into “learning cognitive agents”. These new machines don’t need to be given rules because, given the right context, they can deduce them themselves. They can extract meaning from experience. Which implies an ability to understand. And understanding is key to effective learning. Even for robots, it seems! This is what I found striking and got me thinking…
Although understanding is often equated with knowledge, I see them as fundamentally distinct. One can have plenty of knowledge but grasp very little of its meaning. Rote learning, with its emphasis on memorization and repetition, can be helpful in acquiring knowledge but isn’t conducive to understanding. Rote learning is the equivalent of brute force learning, which is by now considered primitive and inefficient, at least in the world of intelligent machines. I think the same holds true for us, intelligent beings. Despite what some say, we learn best when knowledge is relevant to us, and we discover it ourselves within a real-world context. When we can make sense of it, and understand its meaning.
I realize that people achieve understanding differently than machines. Some people think that, no matter how intelligent they are, machines can acquire knowledge but aren’t capable of understanding. Although they can make connections and learn, machines aren’t able to grasp “the physical and causal relations between things and people” says Michael Stevens here. “Grasping those relations is what understanding consists in”. As a result, machines – in his opinion – are unable to achieve understanding. I would debate this, but I think it’s only semantics.
Isn’t machine learning showing that experiential, active learning is superior to rote learning? It seems it is. Is it showing that the best way to learn how to perform complex tasks is by doing and troubleshooting? Again, the answer seems to be yes. And isn’t understanding what is gained through direct experience? I think it is. Direct experience then provides a short cut to understanding, and therefore it is the smart way to learn. I rest my case.