Now artificial intelligence can save women: “There is a way to predict feminicides”

So far, artificial intelligence has been used to strip women’s bodies, bend them to extreme sexual fantasiesmake porn videos. However, there is also another side of artificial intelligence, that of women can save them. “These tools must be converted to protect victims and could help us to predict possible risk situations“. Talking is Valerio de Gioiacounselor of the Court of Appeal of Rome and consultant of the Commission of Inquiry into Femicide.

As de Gioia explains to Fanpage.it, artificial intelligence can become a tool for calculating data risk percentages: “The results must always be checked and evaluated by the person who makes the final decision, but the indications given by the AI ​​could be useful for the public prosecutor and for the judge who will decide on the request to apply a measure. We could act promptly and not underestimate risky situations.” In the first six months of 2024, 51 women were victims of femicide, according to data from the Interior Ministry, “and even if artificial intelligence were to save just one woman per year it will be worth it.”

How can artificial intelligence be used to protect women?

Let’s start with a big problem. We underestimate. When we are faced with a complaint we often fail to notice the warning signs. An artificial intelligence program, on the other hand, could be very useful for giving indications on percentage of risk and then adapt the precautionary measures based on the indications given in the complaint.

How would this data be collected and analyzed?

The idea is to do it fill out a questionnaire with a plurality of questions which would also give us information that may not seem significant and may indeed be underestimated.

For example?

For example, if the person in question has lost their job, if there is a separation in progress, if they have children, if these children are minors. If filled out correctly, this questionnaire can reveal some indicators common with other femicides and then calculate the risk percentage.

Some predictive systems are already used for example by the police in the United States. Do they work?

The systems work, but as a support, it is essential that the results are always checked and evaluated by the person who makes the final decision. The indications given by the AI ​​could be useful for the Public Prosecutor and for the judge who will then find himself decide on the request for application of a measure.

Don’t they risk discriminating? Algorithms have biases based on training data.

Certain. I’ll give you a trivial example, programs must be trained, that is, a person gives instructions to follow, if the programmer finds it normal to scream in the evening, or smash plates, it’s clear that he will tell the algorithm that those they are not indicators of danger.

So how can we avoid these biases?

We must be careful not to introduce biases because whoever introduces the data must know what the real danger signals are, and we can draw them from experience and crimes that have occurred previously. For example, man’s obsessive control of money it is not a neutral fact, we have seen that economic violence is one of the ways linked to mistreatment in the family. There is also another danger.

Which?

We have to assist the woman when completing the questionnaire. Often those who file a complaint are ashamed, and not only that, there will be many questions, some of which are complex, and the risk is that it will be filled out incompletely or inaccurately, and dangerousness could be ruled out.

Still speaking of discrimination, might the algorithm not protect all women equally?

Absolutely, this is correct and has happened before. In other systems, in the absence of children the programs gave a lower risk and instead there were feminicides. For this reason it must be programmed appropriately and correct errors during constructionalways gave a low risk and instead feminicides have occurred in the presence of complaints, planning adequately, and correcting errors during construction, understanding what further modification is necessary.

Can AI be a tool for immediate and personalized assistance for victims? But also for example used to collect evidentiary material safely, without the attacker realizing it?

Certain. An AI system could be used to see the movements of the electronic bracelet, for example. A man could follow a woman even keeping the minimum distance of 500 meters. The algorithm could notice anomalous behavior, for example if it goes out when the woman goes out or comes home at the same time. So even before the measure is violated it could set off the alarm.

Has such a program ever been used in Italy?

In Italy no, they have not yet been used, there is an analysis underway to understand what the parameters with which to train the algorithm could be.

Abroad instead?

In Spain they have already been used but there have been problems, the program works but there has not been sufficient assistance when completing the questionnaire.

I imagine that it will also be necessary to create a regulatory environment that favors the responsible adoption of these programs.

Absolutely, everything that involves the use of AI must be regulated. Also because sensitive data is acquired they can be a problem for people’s privacy. So not just a law, there must also be a regulation that regulates in detail the collection of these elements and the conservation of this data.

Speaking of privacy. Could they be a problem for the victims?

Well no, all information is shared with the consent of the victim.

But could the collected data be hacked?

No, because databases do not have remote connection, that is, the most precious data in many data collection systems do not have connection. It may be more complicated to consult them because you have to ask for the specific data, have it extracted and then put it on a terminal. But you can consult them from time to timeso there is no risk.

AI has currently been used more against women than for women. I’m thinking for example of deep nudes.

Of course, with software it is possible create false images that look real and undress girls creating horrible images and videos. We have a code that severely penalizes the illicit dissemination of sexually explicit real images or videos. It does not apply if the images are fakebecause the law does not provide them. It is necessary urgent intervention by the legislator that can modify that provision, extending the protection. Not only artificial intelligence must be converted to protect women.