Intel has developed aArtificial Intelligence Solution which has aroused quite a bit of controversy: it is a system that, according to Protocol, is able to detect the emotional states of students who participate in distance formations based on their faces and their body language.
The company has developed the solution in conjunction with Classroom Technologiesa company that sells Class, a virtual classroom software that works on Zoom and with which, thanks to Artificial Intelligence, they want to help teachers in the use of educational techniques allowing them to understand the mental state of their students and know if they are bored, distracted or confused. Its ultimate goal is therefore to improve the distance learning experience.
This technology, which Intel wants to bring to markets other than the US where it has been launched, uses video streams from students, which it supplies to an Artificial Intelligence engine along with contextual information in real time. With all this, the software is able to classify the students’ understanding of the subject or topic that the teacher is explaining. Based on what the system indicates, teachers can repeat the topic, reinforce what has been explained with examples to improve comprehension or advance in the explanation if the system detects that they have understood what has been explained.
There is no doubt that both Intel and Classroom Technologies have good intentions regarding this system. But it has raised blisters in various fields, and has opened a debate focused on privacy, ethics, science and the uses of Artificial Intelligence.
In particular, whether the basic scientific premise behind the solution, the fact that body language and other cues can be used to deduce a person’s state of mind, has a use that is within the bounds of the ethics. But the debate also points out that the scientific community has not yet been able to reach a definitive conclusion about what external signals mean about the state of people, so transferring it to Artificial Intelligence systems can lead to errors. of some consideration.
Another problem that experts see in this system is that the expression of emotions is different depending on the culture of each region or country. There are countries where smiles are reserved only for the most intimate circles and the family. And like this reaction, many others, which can confuse the system. Also, we still don’t fully understand the ways in which people express their status. There are dozens, and even hundreds, of small expressions that are still not correctly interpreted in all cases, such as the dilation of the pupils in response to a stimulus, perspiration or an increase in heart rate.
However, according to Nese Alyuz Civitci, a machine learning researcher at Intelthe model developed by the company was created in collaboration with a team of expert psychologists, who were responsible for providing information in this regard, and analyzed the raw data captured in classes in real time using laptops and 3D cameras. After obtaining the videos, the psychologists examined them and labeled the emotions they detected in them. But for these labels to be considered valid, at least two of three psychologists who had examined the same video had to agree on the labels.
Despite this, Civitci himself acknowledges that it was extremely difficult for him to identify the small physical differences between possible labels. Y Sinem Aslan, an Intel researcher who helped develop the technology, says Intel’s emotion analysis AI didn’t have an assessment focused on whether it accurately reflected students’ emotions, but was more concerned with whether its results were reliable or relevant to students. the teachers.
The system can therefore be beneficial and lead teachers to ask the right questions when it is most appropriate, to a student who is struggling. But it can also negatively impact student performance and well-being, depending on its accuracy and how teachers use it.
