Once viewed as science fiction, Artificial Intelligence (AI) has become one of the most widely discussed topics in today’s world. Even if you don’t fully know what this term means and how it will help your business, you have surely watched numerous movies that scratch the surface of this topic.
A great example is The Matrix. There, characters that look, walk and talk like humans. But they are actually programs powered by artificial intelligence. Another example is the film I, Robot. The central character here is a robot that is so intelligent so that he can have feelings and act consciously.
But what is Artificial Intelligence? In this article, we explore what AI is and why is it relevant.
What is Artificial Intelligence?
In the Handbook of Artificial Intelligence, Avron Barr and Edward Feigenbaum describe Artificial Intelligence as a
branch of computer science – the study of the relation between computation and cognition. Research in AI involves writing programs that attempt to achieve some kind of intelligent behavior.
Nils Nilsson gives a more broad definition of Artificial Intelligence, as
concerned with intelligent behaviour in artifacts
in his book Artificial Intelligence: A New Synthesis.
The term artificial intelligence is coined of two words, artificial and intelligence. The first part is easy to understand. In this context, it just means something that was created by humans.
You might argue that artificial intelligence can be created by machines as well. However, these machines would have been created by humans. We could add layers to this logic ad infinitum, however, the prime creator of artificial intelligence will be humans.
The intelligence part is a bit more challenging to define. This is the case because human intelligence is hard to define in itself. What is human intelligence? When can we say that a being is intelligent? There are countless questions like this – and these surely fall outside of technical and scientific domains.
However, without continuing too deep into philosophical questions, we can simplify this discussion by taking human intelligence as a reference point. Think of any kind of human behavior that you consider intelligent. Would it be easy for a machine to act in the same way?
If we create a machine or a program that can simulate human behavior that we consider intelligent, then we have created artificial intelligence. Or, stated differently:
Artificial intelligence is a discipline studying how to create machines which behave as if they are intelligent.
Strong vs Weak AI
You might notice that this definition doesn’t actually say whether machines have intelligence. Acting as if you have something does not necessarily imply that you don’t have it. The important part of this definition is that artificially intelligent machines produce behavior comparable to that produced by human intelligence.
Of course, we can easily extend this discussion to the question whether machines are capable of having real, human-like intelligence. This immediately begs the question whether machines are capable of having consciousness, feeling pain, and so on.
This is probably the most important philosophical question within the area of artificial intelligence. It divides artificial intelligence in two types. The two types are called weak and strong AI.
Weak AI is a form of AI that is merely behaving intelligently. However, it isn’t actually capable of intelligence comparable to human intelligence.
It is not necessarily weak in the sense of tasks it can complete. Today’s weak AI can solve many types of tasks. In many, it’s doing a lot better than humans. For example, state-of-the-art AI programs are able to beat world champions in chess or Go. Weak AI can also drive cars, give intelligent recommendations, create art, understand language and many other things.
On the other hand, strong AI is a form of AI that actually is intelligent. This means that it possesses actual intelligence in the same sense you do.
Strong AI is currently only a theoretical concept, and it is not even clear whether it is conceptually possible. An intelligent machine will have to be created by someone or something, and if we go tracing back the line of creators, we’ll arrive at the human creator(s). So for the strong AI to exist, a human would have to imbue intelligence into a machine, or create a machine that is able to create intelligence. Whether this is possible is not only unknown, but also raises various ethical and philosophical issues that are out of scope of this article.
Strong AI is a very popular concept in movies and books. Emergent consciousness in machines may easily lead machines to have their own will, and perhaps even act against the interest of humanity. Many experts are concerned with its dangers – for example, Elon Musk famously quoted AI can become more dangerous than nukes. Only time will tell whether any of that will eventually materialize.
However, since strong AI is only a theoretical concept at the time of writing, this article will focus on the weak variant.
How does AI work
After covering the philosophical part of AI, we can explore how AI works on the high level. First of all, it is important to note that the line between “intelligent” computer systems and non-intelligent ones is not well defined. Therefore, a certain system or program can be considered intelligent or not, depending on whom you ask.
However, the most common trait people associate with AI is the ability of the machine to learn. The discipline studying that is called, unsurprisingly, machine learning. Although the term artificial intelligence is somewhat broader in meaning than machine learning is, the most common usage of artificial intelligence actually refers to machine learning. Other than machine learning, artificial intelligence may involve any other kind of smart behavior that need not be learned.
But how can a machine learn? What does it even mean for a machine to learn? Isn’t a machine strictly following its program?
These are very valid questions. Put simply, a program can be programmed to be able to change itself, based on the data that’s fed into it. Usually, these changes manifest themselves as changes to internal coefficients within the program. And they are triggered by feeding the desired data to the program.
Interestingly, the way a program can change itself has to be defined by the programmer. Of course, some ways of changing are more productive than others. Instructing a program how to change based on the incoming data, as well as figuring out what data to feed to the program is what machine learning is about.