What is AI really? The term is suggestive of science fiction like Terminator or Westworld, but the reality is much more prosaic. "At best, we can achieve a level of intelligence comparable to that of a snake," says Google CEO Emmanuel Mogenet, who does not attach much importance to Elon Musk's cautionary words.
In a sensational PR coup, Saudi Arabia recently granted citizenship to the robot. At the same time, among others, Tesla founder Elon Musk warns that superintelligent computers will take over the world.
Perhaps, many who hear the term "artificial intelligence" first of all come up with some fantastic movie like "The Terminator" or "The Matrix". A thinking, feeling, and often evil computer creature that is about to destroy us.
All this is very far from how things are with real AI, much more mundane and primitive. In a way, the term AI is misleading. The computer cannot feel and is not aware of its own existence.
The computer can learn to recognize animals in pictures. But while a person sees a cute cat, the computer analyzes how the pixels in the picture are added in various patterns and can therefore distinguish the pattern depicting a cat from the one depicting a dog. But he doesn't understand or care about what the pixels are actually showing.
“The fact is that we are not able to create an intellect that could compare with the human. The fact remains that, at best, we can achieve intelligence on the level of, say, serpentine,”says Emmanul Mozhene, head of research at Google in Zurich.
The Encyclopedia Britannica defines AI as "the ability to perform tasks that are usually associated with human intelligence." It could be playing chess or getting the correct diagnosis with an X-ray.
So by artificial intelligence we don't mean that the computer is really smart, but that it can solve problems in a way that makes us believe that it is smart.
Promotional video:
In the beginning, AI was based on programming a computer according to clear rules within the area where it was supposed to perform analysis. As it turns out, this works well in contexts that can be described in terms of logic, such as chess. Back in 1997, the IBM Deep Blue computer, thanks to clear rules and enormous power, beat the then world champion Garry Kasparov.
But the world is not always logical. Photo analysis is a good example. If you need to teach a computer to recognize a picture of a cat, you give it instructions, for example, that a cat has 1) a mustache, 2) two pointed ears, 3) two eyes, 4) a tail.
If you forget to tell the computer that not all photos are taken from the front, then it will fail if you show it a photo taken in profile with only one ear and one eye visible.
Just like a person, a computer learns from its mistakes. Machine learning - machine learning is the method that is now used by most to create advanced AI. It implies that the computer is given many examples of what it will try to do, such as creating a certain type of art, analyzing X-rays, or driving a car. Then he must try to do it himself. Thanks to the fact that a person tells what is right and where is a mistake, the computer performs the task better and better. Finally, he begins to do it as well or even better than the best people in the field.
We have already stated that a computer (in our case, not yet) cannot have feelings. The computer cannot be angry, it cannot be happy or sad. But this does not prevent him from being able to recognize human feelings. A computer can, for example, see that a person is stressed, afraid or happy.
“He can even better than a human at recognize feelings. A computer that plays poker can not only determine, just like a living player, whether an opponent is showing signs of anxiety or joy. It can see your heart rate by analyzing your complexion,”says Mikael Haglund, CTO at IBM Sweden.
AI is inherently an incredible bore. He can be called a computerized scientist.
A computer may be as good as the best doctors to analyze X-rays for unusual forms of cancer, but the same computer cannot calculate that a person may have stomach ache from a kilogram of candy.
This does not mean that the computer is completely hopeless in terms of transferring its knowledge from one area to another. A computer that has learned to recognize a cat will learn to recognize a dog faster. And a computer that has learned to diagnose breast cancer from an X-ray will quickly learn to find other diseases. The correct use of machine learning can be of great benefit. But it must be borne in mind that a computer, unlike a person, cannot question the masses of information that are given to him at the very beginning. Poor raw data can render AI useless or even harmful. There is an example when controversial scientist Michael Kosinski from Stanford University in the USA wanted to show that his AI can determine the homosexuality and heterosexuality of people only from photographs. The whole point was that Michael Kosinski and his colleagues uploaded photos from various dating sites where people did everything to appear as attractive as possible to their target group. According to widespread criticism of the experiment, the computer was excellent at recognizing the style of clothing, but if the same experiment were carried out with completely neutral photographs, it would have failed. Even if we find it difficult to believe that computers today can experience feelings or consciousness, there are many examples of robots being treated like humans. There are many reports from the American army about how soldiers developed strong feelings not only for living human comrades, but also for robots used in combat. Recently, the state of Saudi Arabia took another step forward and granted citizenship to a robot named Sofia. Sofia can communicate tolerably well with people and even look surprised or happy depending on the context. But it doesn't take too long to realize that her intelligence is far from human. As for Elon Musk's warnings, they do not bother Emmanuel Mozhenet too much. "I really respect Ilona, but here he is wrong."
Mikael Törnwall