What is artificial intelligence?

What is artificial intelligence? It’s an easy question to ask and a hard one to answer—for two reasons. First, there’s little agreement about what intelligence is. Second, there’s scant reason to believe that machine intelligence bears much relationship to human intelligence, at least so far.

There are many proposed definitions of artificial intelligence (AI), each with its own slant, but most are roughly aligned around the concept of creating computer programs or machines capable of behavior we would regard as intelligent if exhibited by humans. John McCarthy, a founding father of the discipline, described the process in 1955 as “that of making a machine behave in ways that would be called intelligent if a human were so behaving.”

But this seemingly sensible approach to characterizing AI is deeply flawed. Consider, for instance, the difficulty of defining, much less measuring, human intelligence. Our cultural predilection for reducing things to numeric measurements that facilitate direct comparison often creates a false patina of objectivity and precision, and attempts to quantify something as subjective and abstract as intelligence is clearly in this category. Young Sally’s IQ is seven points higher than Johnny’s? Please—find some fairer way to decide who gets that precious last slot in kindergarten. For just one example of attempts to tease this oversimplification apart, consider the controversial framework of developmental psychologist Howard Gardner, who proposes an eight-dimensional theory of intelligence ranging from “musical–rhythmic” through “bodily–kinesthetic” to “naturalistic.”

Nonetheless, it’s meaningful to say that one person is smarter than another, at least within many contexts. And there are certain markers of intelligence that are widely accepted and highly correlated with other indicators. For instance, how quickly and accurately students can add and subtract lists of numbers is extensively used as a measure of logical and quantitative abilities, not to mention attention to detail. But does it make any sense to apply this standard to a machine? A $1 calculator will beat any human being at this task hands down, even without hands. Prior to World War II, a “calculator” was a skilled professional. So is speed of calculation an indicator that machines possess superior intelligence? Of course not.

Complicating the task of comparing human and machine intelligence is that most AI researchers would agree that how you approach the problem is as important as whether you solve it. To understand why, consider a simple computer program that plays the game of tic-tac-toe (you may know this as noughts and crosses), where players alternate placing X’s and O’s on a three-by-three grid until one player completes three in a row, column, or diagonal (or all spaces are filled, in which case the game is a draw).

There are exactly 255,168 unique games of tic-tac-toe, and in today’s world of computers, it’s a fairly simple matter to generate all possible game sequences, mark the ones that are wins, and play a perfect game just by looking up each move in a table. But most people wouldn’t accept such a trivial program as artificially intelligent. Now imagine a different approach: a computer program with no preconceived notion of what the rules are, that observes humans playing the game and learns not only what it means to win but what strategies are most successful. For instance, it might learn that after one player gets two in a row, the other player should always make a blocking move, or that occupying three corners with blanks between them frequently results in a win. Most people would credit the program with AI, particularly since it was able to acquire the needed expertise without any guidance or instruction.

So what is artificial intelligence? The truth is, it’s not a science in the common sense of the word. Rather, it’s an aspirational concept that doesn’t quite fit today’s realities. Modern artificial intelligence is a grab bag of computational tools for dealing with a variety of problems that people approach using their native intelligence. But that doesn’t mean the problems require intelligence to solve or that the machines are “intelligent”– it simply means that there may be other ways to solve these problems.

Featured image credit: “Hand, Robot, Human, Machine” by geralt. CC0 Public Domain via Pixabay.