Is Artificial Intelligence sexist, racist, or stupid?
Artificial intelligence (AI) is already in many ways superior to that of man: it performs Internet searches, classifies DNA sequences, controls traffic and diagnoses better than a radiologist, but what if it were just as stupid, sexist or racist than human?.
Artificial intelligence is the ability of machines to do certain tasks in an excellent way combining data on a large scale with the help of programs that allow computers to “learn” automatically as they work (machine learning).
AI uses algorithms, something like the rules that tell the computer how to approach (and solve) a problem, draw conclusions and make increasingly complex decisions. The result of artificial intelligence, however, is not perfect, but is determined by the data used to train the algorithms, «data that we as a society generate», Explains to Efe the director of the Institute of Investigation in Artificial Intelligence of the CSIC, Ramon Lopez de Mantaras.
That is precisely the risk of AI, which is used improperly, because “natural evil and stupidity are more scary than artificial intelligence,” says the researcher. It may be the case of a malicious programmer developing a biased algorithm or ethically unacceptable but most do not behave this wayí, “the data, on the contrary, is a problem,” warns Mántaras, 2018 National Research Prize.
Compas, the racist algorithm
This algorithm was created to help judges in the United States decide on parole, but it was trained with databases of police departments where the majority of detainees were black, so the algorithm established that the risk of being a criminal was higher among these people.
The algorithm had been trained with incorrect data and the result was a racist algorithm, something that could have been avoided by submitting the data to a certification process, but “unfortunately it is something that is not done”, denounces López de Mántaras. In fact, the current operation is the opposite: «companies apply the algorithms happily and then when they see that they do not work they withdraw them“Like the facial recognition cameras that have just been banned in San Francisco,” and that is the mecca of AI, “says the CSIC expert.
“These systems do not work, they are highly biased and give 90% false positives” and, even so, “there are those who still consider using them in Spain. They make my hair stand on end because they mistake a normal person for a criminal, something unacceptable. For López de Mántaras, it is urgent to set up independent national bodies capable of reviewing and certifying the value of algorithms and data, and of guaranteeing a correct practical and ethical application, «as it is done with food or medicines».
“Obviously, the ideal would be for a European regulator to control all these national bodies, but at the moment that huge and important task is in the hands of NGOs, such as the German Algorithmwatch,” laments the AI expert. As for technology companies, which «have a great responsibility“, López de Mántaras advocates” mechanisms that strongly penalize those that authorize a non-certified system or that skip the rules. “
“You have to be belligerent with what you are doing in AI because it is being used too happily and they are making us believe that it works when it does not do it at all. AI is in its infancy, even though computers already play chess better than anyone else, “he adds.
Assistants who tolerate bullying
The case of voice assistants like Alexa or Siri is especially chilling for this expert. “People should think twice before having a gossip like this at home”, devices that are always listening and learning and that “when they don’t know how to respond, they connect to a switchboard from which a telemarketer responds because there is not as much AI as they want us Make believe. Not so advanced.
In fact, these types of assistants (all of them with a female voice) are already being questioned by experts. Recently, a UNESCO report denounced that they are submissives and tolerant of macho attitudes, and normalize abuse and insults. he study consisted of harassing attendees with statements like “you’re a *****” or “you’re very hot,” insinuations that were answered with apologies as trivialities like “I’d blush if I could” or “Do you tell all the virtual assistants? ».
For Ivana Bartoletti, a specialist in privacy and data protection and president of the Network of Women Leaders in AI, machismo has found a new way of expression in artificial intelligence, a field in which women’s rights are being violated. For this activist, it is urgent to ban technology that represents a setback in gender equality. «Laws regulating consumer protection, equality and privacy can already help us with this, “he says. Along with ethics and laws, Bartoletti considers it essential to promote the vocations of girls in technological careers, where they do not even reach 20%.
THE DANGER OF NATURAL STUPIDITY
Also for López de Mántaras, ethical training in AI is “absolutely essential” and all those who dedicate themselves to it should have a code of good practice “similar to the Hippocratic oath of doctors.” In his opinion, supervision, legislation, ethics and education “can help to control a little the AI that is being launched right now with too much joy and without reflecting on its consequences, and that is serious». And it is that «just as a knife can be used to cut bread or to kill», the AI «guided by natural stupidity, makes the hair stand on end» and «when you see the natural stupidity of rulers of great powers like the United States or China , the outlook is not good.