news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Text | Brain Pole

Public safety is like the "immunity" of our society. We may rarely pay attention to it in normal times, but once an accident happens, it means the danger has arrived, which can be as minor as a "cold" or as serious as "fatal".

Security checks can detect tangible harmful items, but it is difficult to detect intangible harmful emotions. These harmful emotions, such as extreme hatred, conflicts, revengeful social psychology, etc., are like time bombs that will cause innocent casualties once they explode.

Therefore, public safety cannot be limited to inspecting objects.

AI emotion detection is an invisible barrier that enhances public safety "immunity".

According to technology media WIRED, cameras at British train stations are currently using Amazon's emotion detection system to monitor situations inside train stations. Once someone with an abnormal emotion is found, staff will be notified for further inspection.

The fact that emotion detection can be implemented in European countries, which have strict protection of personal privacy data, may be enough to explain something.

This article discusses what AI emotion detection does right and why it is necessary to be accepted by the conservative UK.

It is reported that smart cameras equipped with emotion recognition algorithms will be used to detect the emotions of train passengers, predict potential emotions through facial scanning, and thus detect abnormal behaviors such as theft and intruders.

The most difficult part of this matter is not the technology, but how to accept facial detection on the public in European countries, which have the most stringent regulations on facial recognition?

Europe's use of facial recognition technology in public places has undergone a significant change:

Total ban.Since the explosion of deep learning technology in 2017, face recognition in the CV field has become one of the mainstream applications. However, Europe has strictly regulated this, and issued a regulation in December 2019.Artificial Intelligence White Paper (Draft)It stated that the use of facial recognition technology in public places will be banned for 3 to 5 years.

Rock back and forth.After the promulgation of the most stringent AI and data regulations in history, many analysts and technology professionals said that these restrictions have greatly hindered the advancement and application of artificial intelligence technology in Europe, leading to a stagnation in the development of the AI ​​industry. However, Europe's attitude towards this has been wavering. In February 2021,EU Data StrategyThe document softened its tone, saying that there would be strict restrictions but not a total ban. In October of that year, another resolution was passed prohibiting the police from using facial recognition technology in public places.

Temporary relief.Finally, on March 13, 2024, Europe passed the landmarkArtificial Intelligence Act, has softened its attitude towards facial recognition technology, prohibiting "the application of purposeless large-scale surveillance based on sensitive biometric features (such as facial recognition)" and also giving a pass to the reasonable use of the technology.

Why is AI emotion detection given “special treatment”?

This brings us to the "impossible triangle" of public safety: cost, benefit, and experience.

Public safety is "prevention is better than cure". Once a high-risk incident occurs, the cost of "fixing the fold after the sheep have been lost" is often human life, so stricter pre-emptive supervision should be adopted. Generally speaking, it is security inspection.

However, there is an "impossible triangle" in security checks, which requires a complex trade-off between input costs, security benefits, and public experience.

If the security check process is not sensitive enough and "missed inspections" occur, harmful "viruses" will have an opportunity to take advantage. Dangerous groups will develop "drug resistance" to the security check system, just like viruses become more resistant to drugs, and find loopholes in the security check process to circumvent them.

But we are also very clear that indiscriminately adopting airport-level security standards in any place and conducting body searches for contraband will not only affect traffic efficiency and people's travel experience, but also require a larger number of security personnel, bring high costs, and the marginal benefits are very low.

In particular, subways, high-speed railway stations and scenic spots in Europe generally did not have security checks before, which was equivalent to "opening the door" to dangerous goods and terrorist threats. It is almost impossible to widely accept the implementation of regular security checks from scratch, which would require the deployment of security personnel, equipment and other costs from scratch, and educate the public to change their long-standing travel habits.

In this context, the application of AI emotion detection is expected to find a new solution to the "impossible triangle" of traditional security checks.

AI emotion detection is not a new technology, so why has it become a life-saving straw for European public safety at this time?

This starts with a few changes:

The first change is the advancement of algorithms, which has greatly increased security benefits.

The emotion detection system used by the UK high-speed rail station is a technology that Amazon has been exploring for many years in e-commerce, medical care, public safety, marketing and other fields.The Rekognition system can recognize a variety of emotions including happiness, sadness, anger, surprise, disgust, calmness, and confusion. After the 2023 iteration, it can also recognize "fear."

For example, in e-commerce scenarios, store cameras are used to judge consumer emotions in order to optimize product displays; in personal entertainment scenarios, smart devices such as Amazon's Alexa can perceive user emotions in a timely manner and provide comfort or advice when users are in negative emotions such as anger or sadness.

It can be said that the algorithms for sentiment analysis and emotion recognition are now quite mature, and their accuracy and detection precision can meet the complex requirements of balancing efficiency, safety and experience in public places.

The second change is the continuous advancement of the digital society, which makes the cost of AI emotion recognition controllable.

Smart cities are constantly improving in many countries and regions around the world. After years of iteration, the technology and products of smart cameras used for urban security are now quite mature. Among them, "software-defined" smart cameras can load different algorithms online to realize vertical smart applications. AI emotion detection can be updated and upgraded without replacing hardware, and it will not bring excessive cost pressure.

The third change is that in recent years, people’s concerns about public safety and privacy have decreased, so they are more receptive to emotion detection.

With the turbulence of the world's political situation and economic environment, the public in many countries has increased their concerns about public security. Especially in Europe, where the number of illegal immigrants has increased, transnational crimes and conflicts have posed many threats to public security.

In this context, AI emotion recognition can not only detect potential signs of tension and conflict in advance and avoid public safety incidents caused by emotional out-of-control, but also does not make people feel uncomfortable like exposing facial privacy.

This is because the data collected by emotion detection is not "strongly identifiable".

As stated in The Right to Privacy, privacy is “the right to be left alone.” Compared with “strongly recognizable” face recognition,AI emotion detection in public places will not disturb the public or involve personal privacy information before the crisis is exposedThat is to say, emotions cannot be used to uniquely identify and authenticate natural persons, and your name will not be known. This greatly reduces the public's sense of being monitored.

From this perspective, AI emotion detection has achieved a relatively good balance in terms of cost, benefit, and experience, and should become an invisible barrier to public safety in addition to traditional security checks.

AI emotion detection can prevent malicious and extreme security incidents, which will definitely bring more benefits than disadvantages to the whole society. So, when can we start using it?

Frankly speaking, the smart security systems in large and medium-sized cities in China are already doing very well. It is not difficult to launch emotion detection algorithms online. The difficulty lies in:The road to localization of algorithms will take some time.

The biggest obstacle is that the data sets are not large enough or good enough.

I visited a teacher at a university in Shandong, who used deep learning technology to develop a micro-expression recognition algorithm. Micro-expressions are characterized by very short duration, small changes in movement amplitude, and difficulty in concealing and suppressing, so they are very suitable for emotional detection of potentially dangerous people.

Micro-expressions need to be collected through psychological experiments and then analyzed and processed by computers. The subjects need to be shown some stimuli that have been proven by psychologists to induce micro-expressions. A high-speed camera is pointed at the face of the subject and the computer stores the images frame by frame. After the data is collected, it is necessary to annotate it, label it with emotional tags, and spatial and temporal characteristics, that is, the start time, end time, and climax time of the expression.

The teacher mentioned that before his team established the MMEW database, there was a lack of particularly large public databases for research on micro-expressions. The largest database only had 247 samples, and the image resolution was not high.

It requires both psychological experiments and computer engineering. Since micro-expression datasets are so difficult, why should we build them from scratch?

We know that emotional expression is affected by culture, social life and other backgrounds. Most of the emotional images in the world are based on data collected from foreign faces, and they may not be very accurate when identifying the emotional state, intentions and behaviors of Chinese faces. Therefore, the real application of AI emotion detection in the field of public security still needs to start with the solid construction of domestic high-quality data sets.

Another question is, now that we have the algorithm, who will sell it?

A more accurate statement is who will provide computing power and a series of supporting services. Emotion recognition is a complex process that requires powerful computing power to quickly analyze massive amounts of data and provide real-time feedback on emotion recognition results, which places extremely high demands on computing power. In addition, AI is a continuously evolving technology, and emotion detection models need to be continuously learned and optimized to improve recognition accuracy and generalization capabilities. Taking the Rekognition system that has been deployed at the UK high-speed rail station as an example, it is one of the cloud services launched by Amazon AWS, which conducts learning, analysis, and functional improvements in the cloud.

In China, since it involves public information data, it must be privately deployed locally, and the ultimate core competitiveness is still algorithm technology and ToB service capabilities. Therefore, domestic CV algorithm companies will be more competitive than cloud vendors in this market, but how to solve the cost of computing power and the problem of service efficiency is a long-standing problem for algorithm companies, which requires more sophisticated and reasonable business design.

A nine-story tower starts with a pile of earth. Although the localization of AI emotion detection still needs to be consolidated little by little from the aspects of data, algorithms, and business, using technology to enhance public safety "immunity", avoid serious safety accidents, and achieve a balance between cost, benefit, and experience should gradually become the consensus and goal of the entire society.

People are the purpose and technology is the means.

Technology that defends people's right to life should be given the right to develop, even in Europe, which is extremely strict on new technologies. This may be the biggest revelation that the implementation of AI emotion detection brings us.