In conclusion, let us now return to the question of intelligence. Assuming that the world is
unpredictable and yet possesses the tendency to take habits to a significant degree, let us ask:
how should a system act in order to approximately maximize the given "appropriateness"
function A? In other words, let us ask the Kantian question: how is intelligence possible? One
reasonable answer is, by repeating the following schematic steps:
1. Recognize patterns in the environment
2. Construct a model of what the future will be like, based on reasoning by induction, assuming
the strengthened Peirce’s principle and using probability theory. That is, where t denotes the
present time, construct a set of statements of the form "according to my best guess, the
probability that the environment will exhibit pattern X, with intensity K, over time interval
[t,t+v], is P(X,K,v)", where P(X,K,v) is as high as possible.
3. Estimate what strategy will maximize A according to such an assumption.
In later chapters it will be proposed that this three-step process is applied recursively — that each of the three steps will often involve the application of all three steps to certain subproblems. And the process will be embedded in a larger, subtler web of structures and processes. But nonetheless, these three steps form the core about which the ideas of the following chapters will be arranged.