Testing the classier requires the use of a second label data-set called the test data set. In practice, often one overall data-set is carved into a training and test set on which the classier is then trained and tested. The testing and training process may be time-consuming, but once a classier is created it can be used to quickly categorise incoming data.

Unsupervised learning is more focused on understanding data patterns and relations than on prediction. It involves methods such as principal components analysis and clustering. These are often used as exploratory precursors to supervised learning methods. Reinforcement learning is a third type of machine learning. Reinforcement learning does not focus on the labelling of data, but rather attempts to use feedback in 11 12

2 What Is AI? the form of a reinforcement function to label states of the world as more or less desirable with respect to some goal. Consider, for example, a robot attempting to move from one location to another. If the robots sensors provide feedback telling it its distance from a goal location, then the reinforcement function is simply a reection of the sensors readings. As the robot moves through the world it arrives at different locations which can be described as states of the world. Some world states are more rewarding than others. Being close to the goal location is more desirable than being further away or behind an obstacle. Reinforcement learning learns a policy, which is a mapping from the robots action to expected rewards. Hence, the policy tells the system how to act in order to achieve the reward.

2.3 What Is a Robot? Typically, an articially intelligent agent is software that operates online or in a simulated world, often generating perceptions and/or acting within this articial world. A robot, on the other hand, is situated in the real world, meaning that its existence and operation occur in the real world. Robots are also embodied, meaning that they have a physical body. The process of a robot making intelligent decisions is often described as sense-plan-act meaning that the robot must rst sense the environment, plan what to do, and then act in the world. 2.3.1 Sense-Plan-Act