Reinforcement Learning

Chapter 1. Why Reinforcement Learning?

How do humans learn? This deceivingly simple question has baffled thinkers for millennia. The Greek philosopher Plato and his student Aristotle asked themselves: are truth and knowledge found within us (rationalism) or are they experienced (empiricism)? Even today, 2,500 years later, humans are still trying to answer this perpetual question.

If humans already knew everything, then they would not need to experience any more of life. Humans could spend the rest of their time on Earth improving lives by making good decisions and pondering big questions like “Where are my keys?” and “Did I lock the front door?” But how do humans gain that knowledge in the first place? You can teach knowledge. And higher levels of average education lead to a better society. But you cannot teach everything. Both in the classroom and in life, the student must experience.

Young children inspire this process. They need to experience a range of situations and outcomes. In the long term, they begin to seek rewarding experiences and avoid detrimental ones (you hope). They actively make decisions and assess the results. But a child’s life is puzzling and the rewards are often misleading. The immediate reward of climbing into a cupboard and eating a cookie is great, but the punishment is greater.

Learning by reinforcement combines two tasks. The first is exploring new situations. The second is using that experience to make better decisions. Given time, this results in ...