How far is the hottest AI emotion recognition from

  • Detail

How far is AI emotion recognition from us

according to a recent survey, people around the world may not have the same three outlooks, but they can basically recognize "fake laughter". No matter which country you are in, people can basically better distinguish whether laughter is true or false

this is because the muscles and sounds of a natural smile are quite different from those of a forced smile, such as tightness and fluency. For example, you can easily see that the "fake smile boy" is "fake smile"

so, since people can recognize "fake laughter", can machines do it? In other words, at the current level of technological development, can artificial intelligence recognize people's emotions through recognition technology and then make corresponding responses

we know that one of the most mature applications of artificial intelligence at present is image and speech recognition, which has been widely used in various fields such as photographing and object recognition, image enhancement, human-computer interaction and so on. However, it is obvious that at present, it is still at the level of identification and classification to a large extent. It will take time to carry out deep emotional recognition and understanding through surface recognition, and then optimize the experience of human-computer interaction

but this does not mean that people are out of touch with AI emotion recognition. As a deep extension of people's surface recognition, the changes brought by emotion recognition will be very positive. So, what is the development of artificial intelligence emotion recognition? How far is it from entering our real life

emotions are mostly the same, with different techniques

we know that people's emotions are reflected in many ways. Expression, language, action, etc., can be used as the carrier of human emotions. Different emotions will be reflected in different forms. For example, if a person is happy, he will laugh, both sides of his mouth will tilt up, and the corners of his eyes will also slightly tilt up; If you are particularly happy, you will "ha ha" laugh. Then, many researchers grasp the expressions or actions corresponding to various emotions to train and learn machines

it can be said that all the places that can reflect emotions have been searched by researchers

eye rotation analysis personality. Minke has long summarized that eye rotation reflects people's psychological activities, such as two eyes shine when excited, two eyes are dull when depressed, pupils are absent when sad, angry eyes are wide open when angry, and so on. Others believe that looking at the top left of the eye is lying, and looking at the top right is thinking. Whether scientific or not, it always proves that the eyeball plays a certain role in judging people's personality and emotions

for example, recently, a team composed of researchers from the University of Stuttgart in Germany, the University of Flinders in Australia and the University of South Australia developed a machine learning algorithm. Through a lot of training on the system, it investigated the eye movements of 42 subjects in daily life, and then evaluated their personality characteristics. For example, the algorithm can show the communication ability and curiosity of individuals, and can identify four of the big five personalities. Then, being able to judge a person's personality as a whole narrows the range of emotions that need to be identified to some extent

microexpression analysis. In many cases, people's emotions are not ups and downs, so emotions are more reflected in micro expressions, such as moving the corners of the mouth and blinking the eyes. For example, rolling your eyes may show disdain. Therefore, as a part directly related to emotion, the research on microexpression has become a competitive project for many companies

recently, MIT has used machine learning to capture subtle changes in facial expressions, so as to measure a person's psychological feelings in order to deal with this contradiction. By decomposing 18 videos into frame by frame pictures, the model can learn the emotions behind the corresponding expressions. The most important thing is that it is different from the traditional one size fits all expression recognition. It can be retrained as needed, and has a high degree of individual applicability

language performance analysis. In addition to observing color, another intuitive way to judge people's emotions is to "observe words". For example, the level of speaking voice, the speed of typing, and so on

the camera equipped with the emotional robot pepper of Softbank Corporation of Japan enables it to have the function of expression recognition. At the same time, it can realize the tone recognition of speaking to people based on cloud speech recognition, so as to obtain people's speaking emotions and realize its advertised function of "emotional robot people". At the same time, the emotion aware customer service system developed by IBM can also recognize people's emotions hidden in grammar and typing speed through learning, such as conversational emotion recognition artificial intelligence and Microsoft's Xiaobing

in addition, combining wearable devices to obtain human pulse frequency and other signs will also help to obtain emotion. However, due to the paralysis of users and the lax management of the construction site, accidents of extruded polystyrene board are also gradually increasing. With the support of face recognition, speech recognition, sensors and various data algorithms, artificial intelligence emotion recognition seems to be in a thriving trend

researchers' "hilltop", or the reason why emotion recognition is "stupid"

however, we can see obvious research characteristics from the above research status

first, technological research presents a "small hill". That is, many researchers are trying to explain and experiment through their technical fields, such as some are good at image recognition, some are good at speech recognition, and some are good at sensor data analysis. These researchers or research teams often have their own technical advantages, but there are also some shortcomings

second, the laboratory limitations of technical research. Although three-year-old children can "see people's faces", it is not easy for machines. Therefore, the current emotion recognition is still in its infancy, and many technologies still exist in laboratories or papers. It will take at least three to five years for this to come true

even if some companies develop finished applications, they have been criticized and appear stupid. For example, pepper is often scolded by roast, and he can't hear what people say clearly, and Xiaobing's roast is even more unspeakable. The user stickiness of products launched before the technology is mature is naturally not worth mentioning

so why is it so difficult for pepper and Xiaobing to recognize emotions

normally, it is very imaginative and attractive to use machines to recognize human expressions for emotional judgment. Because even people sometimes find it difficult to detect some flash expressions, which makes it impossible to judge each other's psychological emotions at this time. However, with mature image recognition and sound capture capabilities, artificial intelligence, it seems that dealing with these things should be like chopping melons and vegetables

then, there is a reason why AI is "stupid"

for example, a reason we analyzed above. Research teams or companies often use one or two recognition technologies to make emotional judgments, or grasp a micro expression or a voice, which is obviously imperfect and one-sided. For example, "anger" and "motivation" are two kinds of emotions. Crazy motivation often shows strange anger, which is also full of angry eyes and roaring. How can AI recognize these two emotions

at this time, simple expression recognition and voice recognition reflect limitations, and it is also necessary to add action recognition. For example, whether the hand is a tight fist like refueling, or a provocative gesture with the index finger pointing at the other party. If we add other factors such as language content recognition to form a comprehensive factor judgment, it will be more conducive to the accurate identification of emotions

that is to say, emotion recognition is not only a matter of "observing words and expressions", but also a matter of "observing their actions", which is a comprehensive and three-dimensional analysis of people

in addition, emotion recognition may be more difficult because of false appearances. For example, the fake smile mentioned at the beginning of the article. It is difficult to identify people's most intuitive emotions alone, and how to crack the emotions hidden behind the camouflage is even more difficult

(speak with Xiaobing very low and slowly and get a ruthless answer)

but the difficulty of emotion recognition is obviously not just the points we analyzed above. The reason is that emotion is a kind of psychological behavior, and it will also show different forms of expression due to individual differences. For example, the same gesture represents different meanings in different cultures, just like veneering is a courtesy behavior in some countries and a rogue behavior in some countries

therefore, the complexity of emotion and psychology does not necessarily lead to its inevitable correlation with human actions and expressions. But this does not mean that there is no significance to study it. Instead, it allows us to make it clear that the study of emotion recognition does not necessarily make each emotion have a corresponding external performance, but can maximize the use of artificial intelligence to help us understand emotions, so as to explore more possibilities

can not only chat, but also break cases

so, what convenience can the improvement of artificial intelligence emotion recognition ability provide us

first of all, the embodiment of human-computer interaction is more natural and smooth, which will also directly reverse people's daily evaluation of "artificial intelligence" as "artificial intellectual disability". This will play a positive role in the care of empty nesters and children. By capturing people's emotions, it can provide psychological comfort for the elderly and children. On this basis, the use of artificial intelligence for emotion recognition can also better help solve mental illness problems and share the energy of psychologists. Especially in the area of dialogue, mature AI will take care of the patient's emotions in the process of dialogue, so as to slowly alleviate the disease

in addition, in the care of the elderly, they can also judge whether the elderly have Alzheimer's disease according to the recognition of micro expressions, so as to make timely reminders. For example, the emotion recognition algorithm developed by MIT mentioned above not only recognizes emotions, but also must be able to skillfully and accurately analyze micro expressions

secondly, emotion recognition can help improve the efficiency of criminal interrogation. A scene that often appears in film and television dramas is that the suspect sits in the interrogation room as if nothing had happened and remains silent regardless of how the police interrogate him. We know that when the police use the known evidence to "cheat" the suspect, they also want to know how far the suspect's psychological bottom line is

then, install a camera and sensor in the interrogation room, and the technology room on the other side can analyze and detect the suspect's pulse, body temperature, expression, laryngeal wriggle and other details in real time, which is more conducive to mastering the suspect's psychological changes, so as to know the interrogation process like the palm of your hand. For example, Yixing procuratorate and China University of political science and law jointly established a micro reaction laboratory, and through the successful capture of the transient expression of the suspect, adjusted the direction of the trial, and successfully guided the suspect to account for the facts

it is foreseeable that with the maturity of emotion recognition, this technology will soon be deployed to the police system

finally, microexpression will remind people in real time of situations involving safe operation. Modern social work presents a high-intensity situation, and people are almost exhausted in order to survive

Copyright © 2011 JIN SHI