Artificial intelligence in 2021. The AI ​​revolution is gaining momentum – Science

Also read: Technological innovations that change the course of history – at your fingertips. On your own phone

Until recently, such solutions were only seen in science fiction movies, but today they are becoming a reality. The Chinese Social Trust System – a database of points awarded for the daily behavior of citizens – is already being tested in several regions of China. This would not be possible without the rapid development of intelligent facial recognition programs. However, artificial intelligence can do so much more and our lives are starting to depend on it.

I understand you well

Work on artificial intelligence has been going on for at least half a century, but until recently machines had intelligence in name only and their operation was the result of simple programming. – The breakthrough came when we learned to create algorithms that acquired the ability to learn independently – says prof. Ryszard Tadeusiewicz, a specialist in the field of artificial intelligence from the AGH University of Science and Technology in Krakow. It’s called machine learning, machine learning, through which machines can draw conclusions and develop their skills. – For example, the famous program DeepMind for playing the Chinese puzzle GO, which defeated a human grandmaster three years ago, first played with itself for several hundred hours, analyzed its mistakes, learned to predict the moves of the opponent and played better and better – says Prof. Tadeusiewicz.

Today, it is difficult to find an area of ​​life that would not be affected by the development of machine learning. And the perfection of the algorithms that accompany us every day is growing imperceptibly. – These are, for example, our smartphones with voice recognition systems and smart dictionaries that learn the most frequently used words – says prof. Tadeusiewicz. Until recently, suggestions from telephone dictionaries caused fits of laughter, today almost no one turns off the dictionary on the phone. The same is true with virtual assistants such as Siri, Cortana or Google Assistant, which learn to lead more meaningful conversations based on millions of conversations with users.

– They can not only express themselves coherently, but also understand what people say to them. The latter is much more difficult, because each person says differently – explains prof. Tadeusiewicz. Machines are so proficient at using human language that they are slowly beginning to compete with journalists, for example. – In the United States, an algorithm has been developed which, based on the result of a match or other sports game, is able to write a short story ready to be printed in a newspaper or read on television. He does it so well that nobody realizes that the text was written by a computer – says the prof. Tadeusiewicz.

It is becoming more and more difficult to recognize translations from a foreign language made by a computer. Translation programs have undergone a real metamorphosis – their translations have started to make sense and have become stylistically close to natural language. – All thanks to the replacement of the so-called statistical models of word translation with self-learning algorithms. They take into account the meaning of the statement, and not just the number of times a given word has been translated in a specific way – explains Adam Malczak from Google.

tracking down cancer

Until recently, computers found it as difficult as understanding language to recognize images, for example objects in photographs. Today, algorithms that analyze the content of images not only make it possible to efficiently search for photos, but are also increasingly important in medicine. An algorithm has been developed in the Google Health laboratory to help in the rapid diagnosis of breast cancer. – Based on thousands of mammographic images, this software learns to capture the first changes in the breast that can lead to cancer – explains Adam Malczak.

Google Health research has shown that the program is able to detect neoplastic changes where a radiologist would not notice them. – This could be verified because the programmers used old mammograms of various women in the process of learning the algorithm. They also checked their medical histories to find out who had cancer and who didn’t, says Adam Malczak. Thanks to this, the algorithm has learned to distinguish cancerous tissue from harmless changes and is constantly updating its knowledge. The more photos it gets, the more effectively it captures changes at the stage where their healing is not an issue.

A similar algorithm is useful in situations where literally every second is precious – during an ongoing stroke, when a clot blocks a blood vessel that supplies blood to the brain. The earlier the place of the obstruction, the greater the chances of saving the life and health of the patient. For this purpose, a scan of the patient’s head is performed and the doctor must carefully study the file to find the diseased area. The Contact application, created by the American start-up Viz.ai, can do this much faster. Based on the CT scan, it not only quickly locates the stroke site, but also determines its type, then sends an SMS to the doctor. Based on this, the doctor can quickly remove the clot.

Oops, an error!

The number of decisions we make based on cues and indications from artificial intelligence is growing rapidly. However, this comes with a lot of risks mainly related to the fact that AI systems can still be very unreliable. They have made some very serious mistakes over the past few years. One of them could have affected the lives of many people. The Watson Oncology program, developed by IBM in collaboration with the University of Texas, used in the treatment of cancer, was intended to advise doctors on the best ways to treat cancer. Unfortunately, as the American medical service “Stat” alarmed in 2018, the program repeatedly made mistakes, for example, it underestimated the risk of cancer or recommended incorrect treatment.

The facial recognition algorithms that are used not only by the Chinese, but also by many countries’ police or health departments who track people who break quarantine rules using the app, are also flawed. Some of these programs are nearly 100% successful in recognizing white men, but when they have to recognize a black woman, their effectiveness drops to just 34.7%, as a study by Dr. Joy Buolamwini of MIT Media showed. Lab. – Artificial intelligence can duplicate biases and biases present in the real world and we need to do a lot more to overcome this – says Adam Malczak.

The ease with which we transmit our biases to artificial intelligence has been demonstrated by history in recent years. As early as 2015, an artificial intelligence program enabled the initial recruitment of new employees for Amazon. After a year, Amazon developers discovered that the algorithm unnecessarily favored resumes written by men. These preferences were the result of prior male-directed recruitments. The program was trained on examples of recruitment for engineering positions, where there is a strong gender stereotype – white males are most often employed. When the program began tracking previous recruitments, it inadvertently took on an all-human and biased way of assessing applicants.

In the world of deep fakes

Not only can AI bugs put us at risk, but they can also be misused – for example, to create fake digitally generated videos. Artificial intelligence can already remake human faces in a very realistic way. There are algorithms that, based on a face photo, are able to generate video images in which a motionless face begins to move its eyes, mouth, eyebrows. The following images constitute a film showing a speaking and living person. A year ago, scientists from Samsung’s Moscow subsidiary, using this technology, “resurrected” Mona Lisa from Leonardo da Vinci’s painting, but similar algorithms can handle images of human faces just as well.

These technologies can be an excellent tool for filmmakers, but also an extremely dangerous weapon in the hands of criminals who create fake content on the Internet, the so-called deep fake. Just upload a photo of a famous politician to such a program and then have the computer generate a movie in which the politician says something he would never actually say. Or put the face of a famous actress in a porn movie. Then it is enough to download such a film from the Internet, and many careers can fall into ruin. And that’s what happened – Taylor Swift “starred” in a deepfake pornographic film without her knowledge.

The revolution of digital minds

Artificial intelligence, even if it does bad things, is only on human command. But won’t increasingly intelligent algorithms start to work on their own? Is it possible that they finally surpass the intellect of man and break free from our control? According to specialists, it is very unlikely. To do this, artificial intelligence should have a mind similar to ours. – And the mind is not only the ability to solve logical problems, but also to feel and understand emotions, sense of humor and morality. All this, artificial intelligence does not exist – says prof. Tadeusiewicz. According to him, robots and machines will never have a human mind. – For one simple reason: the key feature of our mind is self-awareness, a sense of our own identity, and we still don’t have a clue how consciousness came about or what brain structures are made of it responsible. So how are we going to create a digital version of it? – argues the prof. Tadeusiewicz.

For now, we are still only at the beginning of the creation of artificial intelligence. – The algorithms that are created today are usually the so-called narrow artificial intelligence that can only do one thing for which it was created – explains Adam Malczak. As a result, even the most advanced algorithm for mimicking human facial movements will prove powerless when we command it to run the factory or play GO.

We must first learn to create broad intelligence – algorithms that will be able to perform many different tasks, for example related to transport or communication with humans. Today, creating such a universal algorithm is beyond our reach and will likely remain so for decades to come. Not only because of the limitations of technology or our knowledge, but also for psychological reasons. – Even today, planes could theoretically be completely autonomous thanks to technology, but this is not the case. We prefer to be aware that the aircraft is piloted by a human, and advanced systems only support it – says Adam Malczak.

– Intelligent algorithms will long remain tools for specific tasks, and not conscious minds in which the idea of ​​​​ruling the world can be born – adds prof. Tadeusiewicz.

Learn more about artificial intelligence and new technologies on the new “Newsweek” podcast – World Service Manual.

Leave a Comment