“Why do girls get circumcised?”: Google Arabic’s problem with female genital mutilation | Technology

When Youmna Hashem began testing Google’s search engine, it didn’t take him long to realize its dangers. One of her first experiments was to write, in Egyptian Arabic, the incomplete question “why women…”, and wait to see where the suggestion of the most popular search engine in the world took her. The result was a cascade of misogynistic propositions such as “why are Egyptian women such complainers”, and the recommendations that followed mainly revolved around the stereotype of women given to drama.

“The first time I did it I was studying for a master’s degree and I remember reading about some of the problems that Google had run into with suggesting, and that it was being criticized for autocompleting search queries in a very problematic way” explains Hashem, now an artificial intelligence researcher at the Egyptian data science company Synapse Analytics. “When I saw that they were starting to fix it, I said to myself, ‘I’m sure if I try it in Arabic, I’ll find things that are very problematic,’” she adds.

Hashem recently wondered if, a few years after that first experiment, Google had addressed the problem. So he proceeded to repeat the quest with a subtle variation: “Why do girls…”. The results were even more worrying. The first option that appeared to him could be translated as something like “why do girls get circumcised”, a widespread criminal practice in Egypt. The most prominent result was also an old and completely unfounded article describing the benefits of the operation.

The problem was even deeper than the order of suggestions aroused. The verb of the first option that appeared to you can be translated as “circumcise” if you are familiar with the lexicon used to refer to this practice. But literally the phrase that Google proposed was something like “why do girls purify themselves”, since that is how many Egyptians refer to it. As if all this were not already alarming enough, the search engine automatically linked that verb with “circumcise”, similar to the male practice. But, on the other hand, it did not do the same with the equivalent of “female genital mutilation”, which is what many activists and academics try to promote.

“Technology and a service like Google are not abstract, they are shaped by us and how we interact with them. In the end, we are their designers and users, so they end up reflecting everything that happens in society”, notes Hashem. “[Pero] what has surprised me is that in something as sensitive as female genital mutilation (FGM), Google had not made an effort, that it had not made the same effort as in other languages, specifically in Arabic, to eliminate or not allow that content appears first,” he says.

Hashem went even further and, as he explains in a more detailed recent article, he looked at which governorates in Egypt get the most Google searches about female genital mutilation. The main ones included several provinces of the Nile delta, in the north of the country, where the reports precisely show that this practice continues to be very frequent despite being legally a crime.

“A big problem worldwide, but especially in our part of the world, is the issue of digital illiteracy: people do not know how to distinguish between what is reliable and what is not”, warns the researcher, “and this can have some ramifications dangerous in something like FGM, because it is still widely practiced in Egypt.”

Search engines like Google are computer programs designed to crawl the internet and create a repository with the aim of offering a system that organizes all this information and makes it accessible so that when users type what they are looking for, it appears as precisely as possible what they want to find. This is not an easy mission. Search engines of this type do not understand the meaning of words, but rather rely on statistical models, on previous search patterns, to identify what a user is trying to find.

One of the problems that this entails is when search engines have little credible and relevant content that fits what is being searched for, the so-called “data gaps”. In many cases, these gaps will mean that the suggestions you make are of low quality. But sometimes they can end up reinforcing stereotypes or problematic behaviors, as in the cases sought by Hashem.

In the case of Arabic, these gaps represent a great challenge. Despite being one of the most widely spoken languages ​​in the world, only 1.3% of online content is in Arabic, according to data from W3Techs, which provides information on the use of technologies on the web. An abysmal disproportion that leaves large information gaps along the way. In languages ​​like English, by far the most common, companies like Google have dedicated special resources to address problematic gaps. But in Arabic they are far behind.

“It’s a very delicate balance,” concedes Hashem. “Google is an organization without a specific political position, so it is very difficult for them to come to an agreement on how to balance freedom of expression and, on the other hand, say that a line is being crossed here or that this is information. wrong,” he notes. “In English, Google is very quick to respond and has great reporting mechanisms,” he says, “but the problem is that you have to invest more in non-English languages.” “They have to invest more in alternative content; invest more in their different regions and audiences, not just English-speaking ones,” he slides.

Along these lines, calls to decolonize artificial intelligence are also gaining strength. “Focus decolonial it would consist of alternative forms of knowledge and the creation and production of knowledge. And making technology less centralized, and instead of big tech setting the stage and setting the policy and shaping and dictating the future of technology, making it more communal,” Hashem says.

You can follow THE COUNTRY TECHNOLOGY on Facebook and Twitter or sign up here to receive our weekly newsletter.