Both algorithms and artificial intelligence itself already intervene and influence many aspects of our daily lives: from the music we listen to to the movies we watch, through the people we know or even access to bank financing.
Little by little we have allowed artificial intelligence to make decisions on our behalf on certain matters, as in many other areas, due to the benefits that they can offer us in terms not only of efficiency, but also of speed of decision.
While it is true that dependence on artificial intelligence continues to grow, it is also legitimate to ask the question of the risks of such dependence. And, as many specialists agree to point out, a better understanding of our perception of AI is fundamental , precisely because of the aura of infallibility and objectivity that we attribute to it, perhaps in an abusive way.
For many, the algorithms are impartial and neutral, and represent a kind of incorruptible justice of the peace because it is based on mathematics. But the reality is a bit different. And, unlike most expert systems based on rules specified a priori, AI learns from data from the past, determining the decision rules that make it possible to reproduce it .
Thus, under the assumption that the same causes produce the same effects, artificial intelligence is able to use those same rules in order to predict the future. But it is a learning system fed, trained and interpreted by a human being . And indeed, almost by nature, potentially biased, either consciously or unconsciously.
For this reason, a growing research team agrees that artificial intelligence bias may originally end up leading to biased results , especially for minority populations and women.
Facial recognition technologies, just to name a few, have come under increasing scrutiny because they have been shown to tend to detect white faces better than the faces of people with darker skin, as well as the faces of men, compared to women. In this way, the errors in these systems have been implicated in a series of false arrests due to an identity error .
Concern about bias in facial recognition technologies led to a series of bans related to their use. In fact, in June 2019, San Francisco (in the United States) became one of the first cities in the country to ban the use of facial recognition technologies not only by the police, but also by other city departments. . And California followed suit a few months later, in January 2020, imposing a three-year moratorium on the use of this technology in police body cameras.
AI bias would reflect an unfair world
Experts wonder if racial profiling has also been digitized, due to the fact that various technologies have taken facial recognition much further, analyzing and interpreting facial attributes and other data to carry out risk assessments or identify unusual or unusual behaviors. threats.
This is what experts call “anomaly detection”, and increasingly, these technologies are implemented by law enforcement agencies, security companies or the airport’s own security systems.
In a recently published study, experts showed that these types of abnormality detection algorithms were more likely to predict that African Americans or darker-skinned men were abnormal.
And since anomaly detection is often applied to people suspected of being or having unusual behavior, it is clear that ensuring fairness becomes extremely important. In other words, if one of these algorithms is used for surveillance purposes, it is more likely to identify people of color.
But this would not mean that the machine is originally biased. Simply that you have seen more white faces in the data than you were previously entered with , so you would only have learned to associate that with normality. And this bias would reflect a world that, in reality, continues to be very unfair.
Not surprisingly, the danger really comes from considering AI as a system capable of making decisions for itself. Therefore, if the information that has been provided from the outset is partial or incorrect, the decision will be equally incorrect.
