"There are compelling technical problems in understanding how a neural network can solve crucial computational problems that arise in reasoning about and acting in the world. A basic problem, and the one that we will focus on, is the "variable binding" problem. As a first example, consider your ability to pick up objects. Depending on the object, its current position, and your goals, you have a very wide range of ways of grasping and manipulating things, all realized by the neural networks of your brain. This is an instance of the variable binding problem because choosing values for the three variables object, position, and goal has consequences throughout your brain on how the action is carried out. In conventional computing, we assume that different modules all have access to the values of (global) variables and can modify their behavior appropriately. Any theory of neural computation needs some mechanism for achieving this kind of global effect."