With the ongoing global pandemic, caused by the coronavirus disease (COVID-19), remote learning has become a norm for most students and teachers. However, many students have started to lose their motivation to attend their classes due to the significant lack of human interaction and even more loaded schoolwork. As a result, teachers have become burdened with the double responsibility of adopting a new teaching method that suits online learning and keeping students attentive, which leaves them with no spare time to look into each student and understand them. Emotion reading artificial intelligence (AI) is an emerging solution that can narrow the gap between teachers and students in an online setting. Still, these types of facial recognition may require further technological developments and other considerations before being used in the field.
The unprecedented COVID-19 pandemic has challenged education sectors worldwide as schools have been forced to shut down. In response, more than 90 percent of schools have adopted remote learning policies, according to the United Nations International Children’s Emergency Fund (UNICEF). While online learning can be more effective for students that tend to learn slower than their classmates, younger students who are easily distracted and students without stable access to the internet are more likely to struggle. In the age of technology, AI is a leading solution to the drawbacks of remote learning.
Facial Recognition AI in Online Classrooms
4 Little Trees (4LT) is an educational software developed by the Hong Kong startup Find Solution AI, which aims to provide personalized content and detect users’ emotions. According to the 4LT website, founder Viola Lam initially created the software to aid her autistic son’s learning experience, as children with autism often struggle with verbally expressing the problems they face while studying. Launched in 2017, 4LT originally featured motivational study plans, and it has now incorporated emotion reading technology that functions as a helping hand for teachers in online classrooms. Over the past year, the number of schools using 4LT has increased from 34 to 83.
4LT is essentially devised for one-on-one observation, as it has access to the front camera of a user’s device and tracks muscle points on their faces to assess their emotions. Specifically, it observes the movement of facial muscles, such as the corners of the mouth, and matches it with its database to distinguish emotions. The AI also monitors students’ learning habits by analyzing their strengths and weaknesses by recording how fast they answer questions. By doing so, it creates a personalized study plan for each student that distributes different classwork and tests compatible with each student.
Many developers of facial recognition AI like 4LT claim that the technology provides teachers with a clearer picture of their students’ personalities. However, as Professor Christian Wallraven (Department of Artificial Intelligence, Department of Brain and Cognitive Engineering) has pointed out, meaningful implementation of facial recognition in education remains limited. For one, current technology can only detect six emotions — happiness, sadness, disgust, fear, surprise, and anger — all of which may not be the most relevant to measuring students’ level of attention in class. While he acknowledges that other more basic parts of 4LT may enhance learning experiences, the effectiveness of its facial recognition AI requires authentic shreds of evidence.
The Hidden Downsides of Facial Recognition
Breach of privacy is one of the recurring concerns regarding the application of facial recognition AI. Constantly monitoring the public generates an enormous database that the general populace do not have access to, which can plant a sense of mistrust and fear in their minds. Professor Wallraven highlights the importance for potential users of these facial recognition AI to “demonstrate the need for it, and [prove] that the need is served by technology”. In other words, these AI must match the needs of its users and deliver what it promises in order to be effectively implemented in society.
In addition, the accuracy of facial recognition is heavily correlated to race and gender. A Massachusetts Institute of Technology (MIT) study in 2018 tested three facial analysis programs, only to discover that error rates were below 0.8 percent for light-skinned men, but up to 34 percent for dark-skinned women. The significant gap highlights the inherent inequality rooted in the development process, as databases are predominantly based on white men. This bias explains why using facial recognition on school campuses is controversial. Earlier in 2020, the University of California, Los Angeles (UCLA) canceled their plan of implementing facial recognition after strong student disapproval. This was because a test run that used photos from a mugshot database revealed that people of color were misidentified significantly more than other ethnic groups. Facial recognition requires more research before being enforced to the public, especially as the application of this flawed technology may only heighten preexisting social problems.
Besides the fact that introducing facial recognition AI to school campuses stirs up underlying problems in the technology, the plan itself may not be quite cost-efficient. Professor Wallraven provided an interesting point of view that schools are trying to implement facial recognition to prevent possible school shooters or sex offenders from entering the school, but this simply addresses the symptoms, not the disease. He explains that school administrators should transfer their focus and financial resources to building mentally healthier school communities. Using facial recognition for such reasons will only put the majority of students in constant uneasiness, and those who are truly determined to commit the crimes will eventually work their way around the system.
As remote learning has started to completely replace traditional schools in many countries, educators now have to devise new ways to engage students through their computer screens. 4 Little Trees presents a creative example as it incorporates facial recognition AI into online classrooms by analyzing students’ studying behaviors and identifying their emotions. While their efforts of enhancing students’ learning experiences in this unprecedented setting are meaningful, the true efficiency of its emotion reading ability remains questionable. With facial recognition AI being relatively recent, pushing the technology into society may be a hasty move. Instead, people should focus more on resolving its underlying errors and preparing society for it through proper education and policies.