A study done at Ben Gurion University on artificial intelligence found that when identifying a person's age, the computer makes a mistake in estimating the age of the person photographed more than humans.
That is, a person who has to estimate the age of his colleague photographed in a still photo, will know how to score closer to his real biological age than the computer that will use artificial intelligence.
The researchers who conducted the study are Prof. Zvi Ganel from the Department of Psychology and Prof. Carmel Sofer from the Department of Neuroscience and Cognition, both from Ben-Gurion University of the Negev, together with Prof. Melvin Goodale from the University of Western Ontario in Canada.
• One mistake in Google's artificial intelligence system cost it 100 billion dollars
• The Internet Association report: a 3-fold jump in complaints about TikTok vulnerabilities
• "I didn't imagine that my friends from the Negev and I would build a satellite from scratch at the age of 17"
In the study, the researchers collected artificial intelligence performance data from 21 artificial intelligence technologies used to estimate age based on a face image (such as Amazon's, Microsoft's and Everypixel's) collected in 2020-2022.
The performance of the artificial intelligence was compared to the performance of 30 male and female undergraduate students from Ben-Gurion University.
The results showed that artificial intelligence is less accurate and more biased than human observers in estimating a person's age.
Thus, for example, humans tend to rate smiling faces as older compared to the faces of the same people with a neutral expression, and are less accurate in age judgments of older faces.
According to the study, contemporary artificial intelligence technologies also exhibit these biases and inaccuracies, even to a greater extent than humans.
Professor Carmel Sofer, photo: Danny Machlis, Ben-Gurion University of the Negev
To better understand the implications and implications of the research on the field of artificial intelligence, we spoke with Prof. Zvi Ganel from the psychology department at the university: "When you look at the performance of humans and artificial intelligence," he notes, "dealing with human faces is very interesting. We would have thought that the computer would fall more into the age of the person, but it actually turns out that things are not like that."
"The first example is of a smiling face: people are more inclined to attribute maturity to a smiling person than when he is photographed with a neutral face. It turns out a bias that is shared by humans and artificial intelligence. And not only that, but the error of the artificial intelligence is twice that of the human. That is, if a person increases the age The person photographed is estimated by three years compared to his real age, the artificial intelligence will estimate him by six years more."
How did you conduct the research?
"We divided the still images into three age groups: 20-40 years old, 40-60 years old, and 60-80 years old. Both the students and the artificial intelligence estimated the age of the photographed person as older than his actual age, but the gap between the artificial intelligence's estimate and the real age was greater."
"In the case of the 40-60-year-old photos, both the artificial intelligence and the humans scored more or less the correct age, but in the older group, 60-80 years old, the artificial intelligence assigned them a much younger age than the students' estimate."
Differences between women and men
Another interesting area where both the students and the artificial intelligence were wrong, is deciphering the age of women.
It turns out that when a student looks at a picture of a woman, he estimates her to be younger than her actual age.
Here, too, the artificial intelligence was more wrong than the human.
When asked why such powerful computers misjudge age in such a blatant way, Professor Ganal assesses: "It is true that artificial intelligence has the ability to process and absorb images more than humans, but it is very possible that they also inserted fewer images of older people and fewer images of women.
Women were seen as younger.
Illustration, photo: Getty Images
"This, in my opinion, is the first explanation for the gap - that is, when there is less information about a certain age group, the greater the bias will be. It is very possible that this is also the reason for the mistakes regarding women. Of course, this mistake can be easily corrected by introducing a larger amount of pictures of older men and women ".
However, Professor Ganel does not stop at this explanation, and says that there is still a significant difference between the way artificial intelligence computers analyze the human brain: "You have to remember that the computer studies images and analyzes age based on the images alone. This is a very limited task. On the other hand, The human brain is not only engaged in such tasks, but it has a higher level of abstraction and conveys what it sees a higher level of evaluation. A person has critical thinking and other forms of understanding reality in a more abstract and deeper way, therefore people make less mistakes in guessing the age of the person standing in front of them. In this respect, there is still a significant difference between what artificial intelligence can do and what humans can do."
Can this gap be bridged? What should be done to make the computer more accurate?
"In the end, I assume that they will be able to develop algorithms that will allow the computer to correct itself, reduce the sloppiness and develop thought processes that are similar to the abstract thought processes of humans. Probably in the first stages we will witness 'human biases' in computers as well - that the computer will make mistakes like humans. In the later stages It is more likely that the computers will know how to correct these biases, but these are steps that are still far from us."
were we wrong
We will fix it!
If you found an error in the article, we would appreciate it if you shared it with us