For a proper demystification of Artificial Intelligence we should best look at the basic definition first. So let’s start exactly there and build on that:
“Artificial intelligence (AI) is intelligence exhibited by machines. In computer science, the field of AI research defines itself as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of success at some goal.” [Source: Wikipedia. The intelligent agent definition is based on: Russell and Norvig (2003)]
There’s a lot of information in these few lines. The first part can be translated into: “…computers do things that require intelligence when done by humans” [Copeland (2000)]. So it’s easy to crack the “Artificial” in AI, this obviously means machines/computers are doing something, not humans. The tricky bit is the intelligence. We’ll get to that.
In the very early days of AI, actually before John McCarthy held the famous “Dartmouth summer research project on artificial intelligence” in 1956, considered the first academic conference on Artificial Intelligence, Alan Turing wrote a wonderful paper under the visionary title “Computing Machinery and Intelligence” [Turing (1950)]. On a side note, Turing uses the terms “computing machinery” in the title and later in the paper “digital computers” and “electronic computers” to refer to non-human computers. Keep in mind that back then most computers were actually human beings.
The first two sentences of Turing’s paper are remarkable – and they give us a solid clue:
“I propose to consider the question, “Can machines think?” This should begin with definitions of the meaning of the terms “machine” and “think”.” [Turing (1950)]
Let’s assume that the definition of “machine” – or these modern “digital computers” Turing refers to – is something we know by heart. This leaves us with the “think” or “intelligence” component. What constitutes intelligence or in other words, how can a machine possibly think? Human beings think all the time but it’s not easy to distinguish thinking from purely mechanical processing. There’s no point in drifting into a philosophical discussion on this, so let’s focus on the required ingredients of the “think” component, representing the “intelligence” in AI. We’ll have to embrace the fact that many of them heavily depend on the specific problem at hand.
The more or less application-independent requirements are:
Depending on the specific application we’ll need a lot more. Here are a few additional, clearly not mutually exclusive, requirements:
This is by no means a comprehensive list. Mastering some of the individual core requirements is already a major achievement. Some of the above-mentioned fields are still considered unsolved, with creativity, social intelligence and self-awareness being the most obvious ones that are yet to be cracked.
Turing ends his paper with a brief reflection on what challenge we might want to choose when building machines that think:
“We may hope that machines will eventually compete with men in all purely intellectual fields. But which are the best ones to start with? […] Many people think that a very abstract activity, like the playing of chess, would be best. It can also be maintained that it is best to provide the machine with the best sense organs that money can buy, then teach it to understand and speak English. […]“ [Turing (1950)]
He proposes narrow and clearly defined problems that would represent good examples for “thinking machines”. Each one of them must have been considered a major challenge in 1950. AI has obviously evolved over the decades and many problems have been solved. In fact in today’s terminology we’d distinguish different evolutionary stages of AI, ranging from application specific, or weak AI to Artificial General Intelligence. In the next post we’ll shed some light on these which will help distinguish solutions that are available today from science fiction.