In this chapter, we familiarize ourselves with artificial intelligence as a concept and give some examples.
As you may have noticed, artificial intelligence is a hot topic today. It is almost impossible to avoid news and discussions about artificial intelligence. You may also have noticed that artificial intelligence can mean very different things to different people. For some, artificial intelligence is about artificial life forms that can transcend human intelligence, while others would call almost all types of data processing for artificial intelligence.
We will approach the subject by discussing what is artificial intelligence, how it can be defined and what other subject areas and technologies it is related to. We will start by looking at three examples where the application of artificial intelligence highlights different parts of the subject. During the course, we return to these examples and deepen our understanding of them.
Application 1. Self-driving cars
A self-driving car requires several types of technology based on artificial intelligence. Examples of such techniques are route planning (finding the best possible route from location A to location B), computer vision to identify obstacles and decision-making ability in complex and varied environments. Each su
The same technologies are also applied in other autonomous systems, such as logistics robots, unmanned aerial vehicles (drones) and autonomous vessels.
How are our lives affected? Self-driving cars will increase road traffic safety when the system that controls the car becomes a more reliable car driver than man. Logistics chains will also be more efficient when transport can be optimized automatically. People can then switch to monitoring and checking that the machines are working properly. Since passenger and freight traffic is so central to our everyday lives, there are probably consequences of this transition that we have not thought of yet.
Application 2. Custom content
Much of the information we see in a day is personalized. This means that you and your neighbor see different information content. Examples of this are the content that appears on Facebook, Twitter, Instagram and other social media, such as advertising on the internet, music recommendations on Spotify or videos on Netflix, HBO and other streaming services. Newspaper and other media sites, as well as search services such as Google, also adapt their content for different users.
Although the printed front page of the newspapers is the same for all readers, the web version of the front page can be designed specifically for each reader. The Omni news service is a good example of a personalized news service. Algorithms based on artificial intelligence determine how the personalized content looks to each of us.
How are our lives affected? Although the details of the recommendation system’s algorithms are usually business secrets, knowledge of the basic principles of the algorithms helps us understand what the applications can lead to. Concepts related to this are filter bubbles, echo chambers, magic factories, fake news and other forms of opinion or propaganda.
Application 3. Image and video processing
Face recognition is now a common technology in both consumer applications and programs used in companies and by government agencies. Examples of application of this technique are passport control, photo albums that can be arranged by person and automatic tagging, that is, people are automatically marked on photos in social media. Similar techniques can be used in self-driving cars to identify other vehicles and obstacles or to identify wild populations.
With artificial intelligence you can also generate and process image content. Already there are techniques such as style transfer that you can use to process your own photos so that, for example, they imitate Vincent van Gogh’s paintings. Style transfer is also used to create computer animated characters based on the actors’ real movement and mines, such as in the films Avatar, Sagan om the ring and in various films from Pixar.
How are our lives affected? As these methods are further developed and become available on a larger scale, it will be easy to produce video material that looks natural and which is impossible to distinguish from real material. This challenges the assumption that we can always trust what we see.
What is and is not artificial intelligence? A challenging question!
Artificial intelligence’s popularity in the media has increased as the term has begun to be used for phenomena that have previously been called something else. With artificial intelligence you can refer to almost anything from statistics and analysis to coded if-then phrases. Why? What makes the concept of artificial intelligence so ambiguous? The causes are many.
Cause 1: There is no generally accepted definition
Not even researchers in AI use the same definition. As a research area, artificial intelligence is constantly being redefined by the fact that certain subject areas are no longer considered to fall under the concept or that new specialized areas are emerging.
According to a (slightly geeky) joke, artificial intelligence is “cool stuff that computers can’t handle”. The irony here is that, according to such a definition, artificial intelligence never comes forward. Once a problem has been solved with the help of a computer, the problem in question is no longer about artificial intelligence. There is still a grain of truth in this. 50 years ago, it was considered that automatic route optimization was artificial intelligence. Nowadays, such systems are so commonplace that they are not regarded at all as artificial intelligence but as basic computer science. Similarly, methods for processing uncertain data are now being counted in statistics or probability theory and are no longer considered, as it did a few decades ago, as part of artificial intelligence.
Cause 2: The burden of science fiction literature
The wild visions presented in science fiction literature and films further confuse the importance of artificial intelligence. These include friendly humanoid assistants who go stiff and short-cut announce hyper-correct facts. Often the robot dreams of being human, just like Pinocchio did. Another common robot type in the science fiction fairy tales is the evil robot who turns to his master in the same way as in old stories about witches’ apprentices or Golem in Prague.
Often the characters in the science fiction literature are robotic just to the outside. Beneath the surface there is a doubtless human being. This is understandable, otherwise, they would feel too alien to the reader or viewer. Virtually all literature, including science fiction, can thus be interpreted as a metaphor for man’s place in the world as it appeared at the time. Robots often play roles that are in reality matched by oppressed human groups or ourselves as meaning seekers.
Cause 3: It looks easy but is difficult …
A third factor that makes it more difficult to grasp artificial intelligence is that it is very difficult to predict in advance which tasks will be easy and which will be difficult. Look around and pick up an item. Then think about everything you just did: with your eyes you watched the surroundings, you made a quick decision about which object to hold in your hand, unconsciously you planned the movement path to the object and then you performed the movement through countless muscle contractions. In addition, you managed to squeeze your fingers together with sufficient strength to hold the object.
It is difficult to estimate how complicated this action is. It is best to notice it when something goes wrong. The object is much lighter or heavier than you thought, or someone opens the door at the very moment when you grabbed the door handle to lose the balance. Normally, we are not aware of our activity because our body and nervous system are used to what we do. We have developed this habit through millions of years of evolution that has been refined by years of further training during childhood.
Although we find it childishly easy to grasp an object, the same thing is very demanding in robotics, and research is ongoing to solve the problem in a good way. Google’s project with robots that can grab things and pick cauliflower is one of the latest advances in the field. ..Or looks difficult, but in reality is easy!
In the same way, we find it difficult to play chess and solve mathematical problems. We believe that years of training, higher brain activity and deep concentration are required to master the knowledge. In the field of artificial intelligence research, they were initially interested in these types of questions because they were believed to represent the essence of intelligence.
Over the years, it has become clear that chess is really relatively simple for computers. Computers can handle huge amounts of options, which are determined by relatively simple rules. The computers victory train in chess culminated in 1997 when the famous chess computer Deep Blue won over the world champion in chess, Garri Kasparov. Would you have thought it more difficult for a computer to move a chess piece than to win against the world champion in the game itself? In Chapter 2, we will present appropriate algorithms for playing chess and chess.
Also, many tasks in school mats are based on reasoning with clear rules, and are therefore easy to calculate with the help of a computer. Applying them correctly in different situations still requires a kind of intuition and sense of the situation, which is more difficult to automate.
What would be a better definition?
Listing typical features of artificial intelligence would be a better way to define it than the joke of “stuff that computers (not yet) can handle”. Autonomy and adaptability are such qualities.
Words can be misleading
One must be vigilant when talking about artificial intelligence, since the related concepts can often be misleading. Examples of such concepts are learning, understanding and intelligence.
For example, we can describe a system as intelligent if it produces excellent driving directions when driving or recognizes signs of skin cancer based on images. The concept of intelligence can then easily lead to an assumption that the system in question can handle any task that is easy for us and our human intelligence. Examples of such tasks are shopping and cooking, washing and folding the laundry, etc.
Perhaps we say that a computer vision system understands images if it recognizes different objects in the images, such as cars, pedestrians, buildings and roads. In this case, the concept of “understanding” can be interpreted as the system also understands that it cannot run on a road depicted on a pedestrian’s t-shirt (and over the pedestrian).
It is important to note that intelligence itself is not a dimension or scale, such as e.g. temperature. We can compare yesterday’s temperature with the current or the temperature in Paris with the temperature in Rome and find the difference. We tend to to think that people can be categorized according to an intelligence quota – precisely because they have invented something like intelligence quota. Especially when it comes to artificial intelligence, things are not that simple. AI systems cannot be compared on an intelligence scale. Or could you decide which one is more intelligent: a chess algorithm or a spam filter? Spotify’s system for music recommendations or a self-driving car? Such comparisons are not meaningful as they are about narrow AI, ie even if the system is good at one thing it does not mean that it is good at other things as well. (We return to the concept of narrow AI at the end of this section.)
Why is it better to say “a pinch of artificial intelligence” than “an artificial intelligence”?
The division into the categories of artificial intelligence and non-artificial intelligence is not about unambiguous either-or. Although some methods are obviously about artificial intelligence and others certainly do not, there are also methods that contain a little AI (in the same way as a pinch of salt). It would be better to talk about artificial intelligence as a trait (such as happiness or neatness), the amount of which may vary, rather than arguing whether something is artificial intelligence or not.