Artificial intelligence diagnosing diseases and robots with empathy function, i.e. what good can self-learning algorithms bring
An unknown buyer paid over $ 435,000 for "Portrait of Edmond Bellamy" displayed at Christie's, a prestigious New York auction house in October 2018. The price was 45 times higher than the starting price. It would not be surprising if not for the signature in the corner of the work proving that the painting was painted by a learning algorithm. It was the first Artificial Intelligence canvas to be put up for sale and declared a work of art.
In 2017, the AI algorithm wrote a new chapter of the adventures of Harry Potter. The text analysis program got acquainted with the seven parts of the series written by JK Rowling and based on them wrote a three-page story entitled "The Handsome One". And although the content was naive (e.g. "Magic: Something Harry Potter thought was very good"), full of absurdities ("Ron saw Harry and immediately started eating Hermione's family") and hard to accept by fans of the wizarding saga, the experiment has been hailed as a success. Behind the project was Botnik Studios, bringing together writers and artists who excel in combining technological possibilities with human creativity.ADVERTISEMENT
This is not the end. In 2018, the American singer Taryn Southern released an album "I Am AI" composed and produced with the help of artificial intelligence algorithms. HuffPost commented on the premiere with the words: "It's only a matter of time before the AI program creates a song or movie that will be nominated for a Grammy or an Oscar."Harry Potter chapter created by Botnik Studios (photo: botnik.org/content/harry-potter.html)
There are more examples of the increasing use of artificial intelligence. The popular Chinese website Alibaba recently launched a machine learning-based program to help customers choose their garments. This "digital stylist" remembers what type of clothes we prefer, suggests the right accessories, appropriate colors, and even finds the best price offers. Five million users have already used the services of the mod algorithm.
Are we entering a time when even human creativity will have to give way to AI systems? Can everything that proves the uniqueness of a man be captured by an algorithm?
Nick Bostrom, Swedish philosopher and head of the Future of Humanity Institute, part of the University of Oxford, says that creating an AI system that matches human capabilities will be the last invention our species will create. "When it arises, it will simply become better than us. She will be able to deal with tasks that overwhelm us. In addition, she will do it faster and with a lower probability of error" - the philosopher argues. He adds that we appeared on this planet recently. What we achieved as a species had to do with brain development, with the ability to think, analyze data, and draw conclusions from it. AI is a natural, next step in the development of our civilization, and it seems that we have no way to avoid it.
Panic is not advisable
Artificial Intelligence ignites the imagination and seems to be the answer to every problem. Fast medical diagnosis? Let's use the SI algorithm. Finding problematic provisions in legal contracts? Another program is being developed.
Apparently, the internet giant Amazon has implemented a system that uses artificial intelligence, which not only monitors the giant's employees, but even has the ability to fire them when they do not show adequate efficiency. After the disclosure of this information, Jeff Bezos's company issued a statement in which he argued that although employees are actually tracked by automated programs assessing their work, and supervisors receive reports generated in this way, the company never gave the system permission to fire employees.photo: Shutterstock
It does not change the fact that self-learning algorithms cause serious fears in people about the future. But from the beginning. What is Artificial Intelligence at the moment?
- First of all, it is a poorly defined concept - says Dr. Marcin Kurdziel from the Department of Computer Science at the AGH University of Science and Technology in Krakow. "What we call Artificial Intelligence is just surprisingly well-functioning pattern recognition algorithms and nothing else." It does not think on its own, does not translate knowledge gained in one area into another area, and the impression of intuitive operation is caused simply by the possibility of detecting complex patterns in the data - he adds.
Let's look at the algorithms used in medical diagnosis. Many institutions, including Google creates systems that analyze medical databases. Thanks to this, the appropriate algorithms are able to later make an accurate diagnosis in most cases. The programs in question are so good that when in 2018 a competition between respected radiologists and BioMind, an artificial intelligence system built by the AI Research Center for Neurological Disorders and Capital Medical University in Beijing, was organized, the former lost their way. BioMind was as correct as 87 percent. In addition, the system managed to study 227 cases in just 15 minutes. A group of 15 doctors needed the same half an hour, and their correctness was only 66 percent.photo: Shutterstock
How it's possible? Has the medical intuition failed? No, the SI system simply has continuous access to all implemented knowledge, to all data, millions of analogous cases, hence the high effectiveness of diagnosis.
- Machine learning algorithms, based on neural networks, work very well in such situations. Perhaps even too good, because too sensitive diagnostics may go hand in hand with an increase in the number of unnecessarily additional tests. But this does not mean that computers will take over the role of doctors, they will be rather complementary to their knowledge and skills, says Marcin Kurdziel.
The robot feels good
What if AI systems were taught empathy? Researchers from the Universitat Autonoma de Barcelona have built a robotic arm that can follow the face of the interlocutor and read his emotions. Thanks to this, it imitates the behavior of a domestic animal and becomes a digital friend. The creators were inspired in their invention by a moving lamp from the opening credits of Pixar's films.
Researchers from the AGH UST Department of Computer Science who work with the Pepper robot deal with similar issues. - We teach Pepper to follow the interlocutor's eyesight, imitate human behavior during a conversation, such as tilting the head, looking away in certain situations, various reactions that are supposed to make him perceive him as naturally as possible - says Dr. Eng. Piotr Nawrocki.
Pepper, a humanoid machine created by SoftBank Robotics, is undoubtedly adorable. Or cute, because when asked if he is a boy or a girl, he replies that he is a robot, but he likes people very much.
Pepper was created to provide information at airports and shopping centers. It is also often used in kindergartens to attract children to new technologies. The robot has a fairly simple structure with limited mobility and language skills, but it is great for research. In the future, it can help, for example, in the creation and implementation of empathy systems.
- Electronic devices, thanks to built-in cameras, microphones and other sensors, have the ability to collect information about the state of the interlocutor. It is enough to observe specific points on the face to be able to identify basic emotions, such as joy, anger, fear or sadness. Microphones record the pitch and volume of the voice, all these are data that can help the machine understand our mental state and adjust to it in the future - explains Dr. Eng. Bartłomiej Śnieżyński from AGH.
Then such a robot or computer assistant could adjust its behavior to human emotions. It would be a bit like a robotic nanny from the "Jetsons" cartoon, who recognizes whether her pupils are having a good or bad day, whether they should be comforted, or perhaps suggest a favorite dessert.Pepper Robot (photo: Shutterstock)
Electronic interactive toys are already eagerly bought by, for example, Japanese people, whose way of working means that they do not have time to create interpersonal relationships. In the future, robots equipped with an empathy system may become carers of the elderly, take care of their health, make sure that they adhere to a diet, order food or contact a doctor. All you need is a wristband that measures the pulse and body temperature, similar to sports watches that measure human activity. Such a wristband would connect wirelessly to the robot so that it could evaluate and transmit health information directly to an automated medical center.
Recently, a team of researchers from the University of Vermont proved that properly constructed algorithms can detect early symptoms of depression in children only by the way they construct sentences. Perhaps in the future, the electronic assistant will be able to monitor the mental health of the owner and predict upcoming breakdowns.
Algorithms are already influencing our reality to a large extent. They take care of everything, from providing information on what time the courier will deliver the shipment, through traffic congestion on the road, to determining the optimal method of loading container ships. AI programs also take care of our safety.
On the other hand, Marcin Kurdziel teaches AI systems to analyze text. It uses the methods of deep machine learning (neural networks) to create patterns of recognizing text fragments and nuances of verbal construction, i.e. intuitive understanding that allows us, people, to quickly distinguish symbols and analogies contained in the language.
Let us try to explain this with the example of the word "garden". For people, this word will automatically be associated with others, such as: flowers, trees, grass, plants, greenery, leisure and many more. The machine does not know this connection, we have to teach it to work properly with the text. So just create a neural network.
Those who use Google Translator will probably remember that a decade ago, the system was mainly suitable for a simple one-to-one translation of truncated words. Naturally, it was a breakthrough anyway - there was no need to refer to dictionaries, the translation time was shortened many times. It operated thanks to the machine translation mechanism created by Franz Josef Och, the then head of the team responsible for the Google Translate program. The system was based on a huge database and was clearly groundbreaking for its time.photo: Shutterstock
In 2016, engineer Harold Gilchrist decided to improve it and created GNMT: based on Google Neural Machine Translation. The program is able to translate entire columns and even pages of text while maintaining the context of entire sentences and in a way that is as close as possible to human speech. Moreover, in the case of expressions with many distant meanings, he gave them by way of examples, thanks to which the reader can decide for himself which reference is most appropriate.
But translating is one thing and creating text is another. If we learn AI to work with text, it will be able to independently deal with our correspondence, create report summaries, search and provide us with the necessary information. We will be able to give such a system the following task: "Find me information on the conflict between Słowacki and Mickiewicz, make a summary of the highest-rated books on this issue, confirm the authors' quotes proving the conflict and define the chronology of the conflict". Even with a considerable knowledge of the subject and at the disposal of one of the largest Polish libraries, it would take a man at least a few days to complete this task. The AI system will be able to execute them immediately. What's more, on our recommendation, he will write an e-mail summarizing his work. He will be able to create various types of text, from private correspondence, through research, to the creation of company or legal documents. Not because he will think, but because his machine "intuition" will be based on huge databases and implanted schemas.
Machine learning-based systems are already capable of generating text, but as in the case of the Harry Potter stories, they sometimes produce naive and often absurd texts. Works on solving this problem include Open AI, a San Francisco-based research institution dedicated to creating free access to AI systems and promoting human development. Researchers have created, inter alia, GPT-2 text generating system. Interestingly, it is the first system in the history of this institution that has not been made available to recipients. Why? It turned out to be a great tool for generating fake news.
- We already have a problem with fake news influencing the real choices of consumers or voters. In the future, if rational legal regulations are not created, this problem may become unmanageable - says Marcin Kurdziel and adds that legal regulations should be created immediately, because neither politicians nor the public are aware of the advancement of work on the development of AI much less consider the consequences.
- I do not believe in the Skynet scenario, Artificial Intelligence will not create Terminators killing people. We don't have to be afraid of this. But we should be afraid of people who will try to make bad use of it. For example, create bots that generate thousands of false information in a second and influence the decisions of a large part of society, which, through technological exclusion, can become an easily manipulated mass. If the rulers do not notice this threat early enough, it may be too late in a few years, says the scientist.
It seems that we are still a long way from creating a true Artificial Intelligence that will be able to extrapolate the acquired knowledge in one area to another, self-improve and make integral decisions. In the coming years, AI algorithms will be refined to work in a context on the one hand, and to better cooperate with humans on the other. But it will be the friendly Jarvis rather than the bloodthirsty Ultron from the Marvel comics and movies.
First of all, the programs will improve medical diagnostics and will support researchers in creating precisely targeted drugs for genetic diseases. And who knows, maybe in some time - as Elon Musk and his Neuralink dream - we will combine our mental abilities with the abilities of AI to create a new, better version of the human. But more on that in the next episode.
Thank you for reading our article to the end. If you liked it, try our new newsletter with the most interesting and the best texts of the portal.
CLICK TO SUBSCRIBE TO THE NEWSLETTER >>>
Ewelina Zambrzycka-Kościelnicka. Journalist and editor dealing mainly with popular science topics. Associated, among others, with Życie Warszawy and Weekend.Gazeta.pl as well as with Magazyn Wirtualna Polska.