3. How do AI Machines Learn? – Just Like Humans


Without learning there is no intelligence.

Without intelligence there is no learning.

Learning is the cumulative process of acquiring information, and developing skills to figure out how to solve new problems, and deal with new situations.

Learning is the cumulative process of acquiring information, and developing skills to figure out how to solve new problems, and deal with new situations.

If learning is a prerequisite for intelligence then what is the prerequisite for learning? A rock or a plant cannot learn as it lacks this capacity. For humans and animals in general, thanks to the brain, we have the capacity to learn from our environment. Learning is critical to our survival. At birth, we possess a bootstrap set of knowledge for our basic survival needs. This resides in the limbic region?—?one of the innermost parts of the brain. Over millions of years of evolution, it has been programmed to provide us with our primal survival mechanisms and basic instincts. However, that alone is not enough. The larger part of the human brain, the cerebral cortex, with over 20 billion neurons, each connected to several thousand other neurons, is our neurally networked learning organ. It gives us our capacity to learn, to store our learning, make decisions, and to act. The large size and complexity of our brains make us the ultimate learning species on earth.

So the question is, how do we make all these neurons in our brain help us learn, acquire skills to solve new problems, and deal effectively with new ambiguous situations and uncertainty? Understanding the process of learning and decision-making in humans helps us discover techniques to make machines learn. So how do we humans learn? Here is a simplified view:

Learning is mostly about interpreting data patterns. The human brain is an incredible pattern-recognition machine. Our brains constantly process sensory inputs, label them as a dog, or a cake, and then generate a decision, or an action, as an ‘output’ (this dog is not dangerous, this cake looks delicious, I want to eat it). Our brain is a sophisticated pattern management system for recognizing, storing, and comparing data patterns. We constantly observe incoming data patterns and the resulting outcomes, and based on repeated observations our brain develops an algorithm of how data patterns correlate with outcomes. As we continue observing more and more patterns and the related outcomes, we improve our ability to deduce the right conclusion about what it means (prediction) and what needs to be done (decision). It must be pointed out that this is a statistical, and hence a probabilistic, process of arriving at conclusions ?—?which is error prone. We occasionally do make mistakes and recognize a wolf for a dog, or a plastic look-alike for a delicious cake. Our algorithmic learning becomes more refined and we are able to see subtle differences in data patterns that produce a different outcome. That is why, practice makes us better (but never perfect) and experience counts. A medical doctor who has seen thousands of cases of heart problems is able to quickly read the pattern in the symptoms much better than a less experienced doctor. A professional tennis player, with thousands of hours of practice, can predict quite accurately where the ball is going to land, just by observing the body and action of the opponent as they serve.

In order to learn, machines must have the capacity to learn. They must have a pattern recognition and management system. In modern Artificial Intelligence systems with Machine Learning capabilities, this is provided by an artificial neural network. The neural network of an AI system is analogous to the neural network of the brain in a human. The concept of machine learning using neural networks for learning is not new, but due to recent advances in neural computing hardware, access to a large number of labeled data sets for AI training, and sophisticated deep learning techniques, machine learning and AI have now become a practical reality.

Leave a Reply

Your email address will not be published. Required fields are marked *