Artificial intelligence begins detecting depression

Dr. Hamdi Dibeklioğlu, a faculty member at the Bilkent University Department of Computer Engineering, explained that he has focused his academic work on artificial intelligence throughout his career.

Stating that he works on the automatic analysis of human behavior using artificial intelligence, Dibeklioğlu said, "This is known globally as affective computing. We derive results through machine learning based on the content of speech, the intensity and tone of the voice, facial expressions, and posture."

Dibeklioğlu emphasized that while he has been conducting affective computing research with AI for years, the subject has become very popular recently with the widespread use of AI models like ChatGPT.

"WE ARE TRYING TO DETECT DEPRESSION LEVELS WITH ARTIFICIAL INTELLIGENCE"

Stating that they have been developing algorithms to detect depression levels using AI in their recent work, Dibeklioğlu continued:

"Normally, clinical psychologists and psychiatrists make diagnoses based on observations. What we are doing is actually something similar. We are trying to detect depression levels using various data with AI, ranging from people's facial expressions to their tone of voice, way of speaking, and how or in which direction they turn their heads. While the expert is conducting the interview, the AI-powered system reaches a certain decision and shares it with the expert."

Explaining that he focused on these topics during his research in the Netherlands and continued after returning to Turkey, Dibeklioğlu said, "At every stage of these studies, we obtain ethical approvals from both the patient and the hospital to train the models. You can use this system if the person gives consent. We already refer to such data as confidential or sensitive data. These are very serious processes."

"WE ARE TRYING TO SOLVE THE RELATIONSHIP BETWEEN BEHAVIOR AND DEPRESSION LEVELS"

Stating that they use clinical data for depression detection and work together with psychiatrists, Dibeklioğlu provided the following information:

"We are trying to solve the relationship between behavior and depression levels. Our findings align with theory. For example, there are different thoughts, such as 'my child is crying a lot, are they depressed?' However, the general expectation in depression is dullness. That is, the person does not want to communicate with anyone and disconnects from their social circle. The patterns we capture are like that. In other words, when you examine the literature, you encounter the behavior of avoiding social interaction in depression. Actually, those are what emerge. We define depression level analysis in this way: while the expert is conducting the interview, the system should also examine the patient and support the doctor in some way. That is, the model should also make its own diagnosis. In this way, artificial intelligence can catch something that might be overlooked."

"THE USE OF AI IN LIE DETECTION..."

Dibeklioğlu reported that in another project, they are trying to find the degree of lying based on data such as sentences, tone of voice, and gaze, and that they have also obtained ethical approvals for these studies.

Stating that they examine various videos to determine the ratio of truth to lies in the conversations and verify them through various sources, Dibeklioğlu noted that they evaluate the content of the speech with "natural language processing" models and the tone of voice with "frequency analysis."

Dibeklioğlu expressed, "However, it is not the case that we have solved this with AI and are making 100 percent accurate predictions. It is, of course, not possible to make 100 percent successful predictions without seeing all the data in the world. But we can now reach significant success rates."

LIE

Upon a question regarding lie detectors, Dibeklioğlu said:

"This needs to be understood correctly: you cannot take this lie detection system and use it in court. You cannot tie a person's life to this. These have error rates. But you can use this in many different places. The use of AI-based lie detection could be relevant for scenarios such as student interviews or job interviews. Studies in the literature look at whether the entire conversation is true or false; we are trying to find at what level the lie is being told. This is actually a difference, because not everything in a conversation is true, nor is everything a lie."

PERSONALITY DETECTION IS ALSO PERFORMED WITH AI

Hamdi Dibeklioğlu stated that they are also conducting personality detection studies using artificial intelligence.

Explaining that they evaluate personality in different dimensions, Dibeklioğlu said, "Such as how open you are to the outside world, how open you are to innovation. Again, we collect personality data using visual and auditory elements. Actually, while you are communicating with a person, you are also interpreting them in the background; we are trying to teach this to the machine as well. Of course, the machine has its own different features; it can see much more detail and perform much more serious operations, but the essential thing is that we train the algorithm correctly."

Pointing out that personality analysis with AI is used around the world, Dibeklioğlu said, "I am not very sure how much it is used in companies right now, but I did research in the Netherlands for years. Even in 2013-2014, they were extracting personality statistics in job interviews using AI at large companies there. While we were saying 'it's still a prototype, let's not see the results as very definitive' for these studies 10 years ago, our accuracy rates are high now."

Hamdi Dibeklioğlu pointed out that one must be very careful in the field of human behavior, saying, "Because if you label this as 100 percent accurate and integrate it directly into daily life, and if a responsibility is imposed on the person due to errors there, serious problems can occur. That is why ethical approvals in behavior analysis must be scrutinized very well. Our goal here is for artificial intelligence to assist us. There is no such thing as leaving all decisions to artificial intelligence and stepping back ourselves."

IT ALSO DETECTS PAIN LEVELS

Dibeklioğlu noted that they also perform "pain level" detection with a similar system, which is important in adjusting medication dosages.

Expressing that they expect this detection to be useful especially in treatments to be applied to children and infants, Dibeklioğlu said, "You do not have the chance to ask children and infants about their pain level. In such cases, we can again derive the pain level from facial expressions and behavior."