In september 1948, an interdisciplinary conference was held at the California Institute of Technology (Caltech) in Pasadena, California, on the topics of how the nervous system controls behavior and how the brain might be compared to a computer. It was called the Hixon Symposium on Cerebral Mechanisms in Behavior. Several luminaries attended and gave papers, among them Warren McCulloch, John von Neumann, and Karl Lashley (1890–1958), a prominent psychologist. Lashley gave what some thought was the most important talk at the symposium. He faulted behaviorism for its static view of brain function and claimed that to explain human abilities for planning and language, psychologists would have to begin considering dynamic, hierarchical structures. Lashley's talk laid out the foundations for what would become cognitive science.

The emergence of artificial intelligence as a full-fledged field of research coincided with (and was launched by) three important meetings – one in 1955, one in 1956, and one in 1958. In 1955, a “Session on Learning Machines” was held in conjunction with the 1955 Western Joint Computer Conference in Los Angeles. In 1956, a “Summer Research Project on Artificial Intelligence” was convened at Dartmouth College. And in 1958, a symposium on the “Mechanization of Thought Processes,” was sponsored by the National Physical Laboratory in the United Kingdom.

If machines are to become intelligent, they must, at the very least, be able to do the thinking-related things that humans can do. The first steps then in the quest for artificial intelligence involved identifying some specific tasks thought to require intelligence and figuring out how to get machines to do them. Solving puzzles, playing games such as chess and checkers, proving theorems, answering simple questions, and classifying visual images were among some of the problems tackled by the early pioneers during the 1950s and early 1960s. Although most of these were laboratory-style, sometimes called “toy,” problems, some real-world problems of commercial importance, such as automatic reading of highly stylized magnetic characters on bank checks and language translation, were also being attacked. (As far as I know, Seymour Papert was the first to use the phrase “toy problem.” At a 1967 AI workshop I attended in Athens, Georgia, he distinguished among tau or “toy” problems, rho or real-world problems, and theta or “theory” problems in artificial intelligence. This distinction still serves us well today.)

In this part, I'll describe some of the first real efforts to build intelligent machines. Some of these were discussed or reported on at conferences and symposia – making these meetings important milestones in the birth of AI. I'll also do my best to explain the underlying workings of some of these early AI programs.

The brute force algorithm for an optimization problem is to simply compute the cost or value of each of the exponential number of possible solutions and return the best. A key problem with this algorithm is that it takes exponential time. Another (not obviously trivial) problem is how to write code that enumerates over all possible solutions. Often the easiest way to do this is recursive backtracking. The idea is to design a recurrence relation that says how to find an optimal solution for one instance of the problem from optimal solutions for some number of smaller instances of the same problem. The optimal solutions for these smaller instances are found by recursing. After unwinding the recursion tree, one sees that recursive backtracking effectively enumerates all options. Though the technique may seem confusing at first, once you get the hang of recursion, it really is the simplest way of writing code to accomplish this task. Moreover, with a little insight one can significantly improve the running time by pruning off entire branches of the recursion tree. In practice, if the instance that one needs to solve is sufficiently small and has enough structure that a lot of pruning is possible, then an optimal solution can be found for the instance reasonably quickly. For some problems, the set of subinstances that get solved in the recursion tree is sufficiently small and predictable that the recursive backtracking algorithm can be mechanically converted into a quick dynamic programming algorithm. See Chapter 18. In general, however, for most optimization problems, for large worst case instances, the running time is still exponential.

It is important to classify algorithms based whether they solve a given computational problem and, if so, how quickly. Similarly, it is important to classify computational problems based whether they can be solved and, if so, how quickly.

The Time (and Space) Complexity of an Algorithm

Purpose

Estimate Duration: To estimate how long an algorithm or program will run.

Estimate Input Size: To estimate the largest input that can reasonably be given to the program.

Compare Algorithms: To compare the efficiency of different algorithms for solving the same problem.

Parts of Code: To help you focus your attention on the parts of the code that are executed the largest number of times. This is the code you need to improve to reduce the running time.

Choose Algorithm: To choose an algorithm for an application:

• If the input size won't be larger than six, don't waste your time writing an extremely efficient algorithm.

• If the input size is a thousand, then be sure the program runs in polynomial, not exponential, time.

• If you are working on the Gnome project and the input size is a billion, then be sure the program runs in linear time.

Time Complexity Time and Space Complexities Are Functions, T(n) and S(n): The time complexity of an algorithm is not a single number, but is a function indicating how the running time depends on the size of the input. We often denote this by T(n), giving the number of operations executed on the worst case input instance of size n. An example would be T(n) = 3n2 + 7n + 23. Similarly, S(n) gives the size of the rewritable memory the algorithm requires.

A giraffe with its long neck is a very different beast than a mouse, which is different than a snake. However, Darwin and gang observed that the first two have some key similarities, both being social, nursing their young, and having hair. The third is completely different in these ways. Studying similarities and differences between things can reveal subtle and deep understandings of their underlining nature that would not have been noticed by studying them one at a time. Sometimes things that at first appear to be completely different, when viewed in another way, turn out to be the same except for superficial, cosmetic differences. This section will teach how to use reductions to discover these similarities between different optimization problems.

Reduction P1 ≤polyP2: We say that we can reduce problem P1 to problem P2 if we can write a polynomial-time (nΘ(1)) algorithm for P1 using a supposed algorithm for p2 as a subroutine. (Note we may or may not actually have an algorithm for P2.) The standard notation for this is P1≤polyP2.

Why Reduce? A reduction lets us compare the time complexities and underlying structures of the two problems. Reduction is useful in providing algorithms for new problems (upper bounds), for giving evidence that there are no fast algorithms for certain problems (lower bounds), and for classifying problems according to their difficulty.

From determining the cheapest way to make a hot dog to monitoring the workings of a factory, there are many complex computational problems to be solved. Before executable code can be produced, computer scientists need to be able to design the algorithms that lie behind the code, be able to understand and describe such algorithms abstractly, and be confident that they work correctly and efficiently. These are the goals of computer scientists.

A Computational Problem: A specification of a computational problem uses pre-conditions and post-conditions to describe for each legal input instance that the computation might receive, what the required output or actions are. This may be a function mapping each input instance to the required output. It may be an optimization problem which requires a solution to be outputted that is “optimal” from among a huge set of possible solutions for the given input instance. It may also be an ongoing system or data structure that responds appropriately to a constant stream of input.

Example: The sorting problem is defined as follows:

1. Preconditions: The input is a list of n values, including possible repetitions.

2. Postconditions: The output is a list consisting of the same n values in non-decreasing order.

An Algorithm: An algorithm is a step-by-step procedure which, starting with an input instance, produces a suitable output. It is described at the level of detail and abstraction best suited to the human audience that must understand it. In contrast, code is an implementation of an algorithm that can be executed by a computer. Pseudocode lies between these two.