When it comes to Artificial Intelligence, often experts think that will be good or bad in the long run? What if it turns against humans? How would it learn our values and attitudes, after all, it is just a machine? Well, basically what AI does is; it self-learns. And how does it learn? It predicts different outcomes from every single approach on a given scenario. And it always goes with best-suited one.
Top 10 Famous Dota players including Dendi has been defeated by an AI bot in 1v1 play which was organized by a start-up from Elon Musk. How? It learns and implements its strategies based on probabilities in the game. Just imagine a person playing with you, predicting your next move, adapting and improvising, each second. Would definitely be difficult to win against a person like this.
But games are different than the human belief system. An AI can easily learn any game and defeat a human player. How would it learn human belief system? Well, there comes our Ordinary Reinforcement Vs Inverse Reinforcement Learning.
Ordinary Reinforcement Vs Inverse Reinforcement Learning:
Ordinary Reinforcement Learning is based on a process which involves rewards, punishment and a goal to achieve. For example, in Dota playing, your target is to kill your enemy and destroy its base. Winning is your reward and if you lose, that is your punishment. An AI bot was created for Atari-games. It was given an initial score and its goal was to score more than that. This is how ordinary reinforcement works.
Inverse Reinforcement Learning is the other way around. In this process, the AI learns from human behavior. For instance, take a real-life example. An AI would learn all your daily life tasks just by observing you. What is your routine when you wake up? What do you eat for breakfast and how do you interact with people in daily life? It will see what activities make you feel happy and what makes you down.
In no time, an AI will learn that humans tend to have a cup of tea or coffee in the morning.
Now that was a routine and habit of one person. Imagine what will an AI learns if we give it all the books and internet database on human actions and attitude to these actions. It would learn from previous court cases that learn jail belongs to evil and nasty people and genius awards are given to the people with higher IQs.