A little over a year has passed since the European Commission presented its proposal for the regulation of artificial intelligence. A proposal characterized above all by the classification by risk levels of artificial intelligence algorithms and the banning of some of them, including social scoring systems that categorize people based on their behavior or technologies capable of influencing them in a manner the behavior is subliminal.
The text, on the whole, had been positively received by the experts, above all because it confirmed – after the data protection regulation – the will of the European Union to take the global lead in the regulation of new technologies, differentiating itself from both digital authoritarianism in Chinese sauce both from the total freedom that the United States is leaving to a sector with profound social implications. At the same time, the criticisms had focused above all on the absence of a complete ban on biometric recognition systems, which despite the limitations introduced had left the nations several loopholes.
Policy
The new rules for artificial intelligence under consideration by the European Parliament
by Luca Bertuzzi
Twelve months after the presentation of this proposal by the European Commission, the first package of amendments by the two co-rapporteurs of the European Parliament has now arrived: the Italian Brando Benifei (Democratic Party) and the Romanian Dragos Tudorache of the Liberal Party. “It is a first proposal for changes shared by both: now begins a phase in which we will individually present our further amendments and during which we will intensify the changes on some points, where there was no agreement from the beginning. In general, we aim to build a regulation system for artificial intelligence that has the protection of the human being at its center ”, Benifei tells Italian Tech.
The guide
What the EU is doing to regulate cryptocurrencies
by Giacomo Barbieri
The goal is to conclude the work – and therefore officially present the European regulation on artificial intelligence – by the end of the year, thus concretely affecting the uses and effects of a technology capable of infiltrating every aspect of life. social and economic. “A joint amendment that we have tabled concerns for example the ban on the use of predictive police algorithms, because they are too much at variance with the value of the presumption of innocence. Indeed, with these tools, a presumption of guilt seems to be created “, explains Brando Benifei referring to systems based on artificial intelligence – used above all in the United States – which would be able to predict where and when crimes will occur and that they have been accused of not only being ineffective, but also at the root of dangerous discrimination.
The package of amendments, on the other hand, lacks the complete ban on biometric recognition systems (among which the highly criticized facial recognition stands out), which – just like the predictive police – have often been accused of reinforcing the discrimination already existing in our society and of being able to be the basis of mass surveillance of the population: “We did not intervene on this aspect with the first package of amendments because there were different positions ”, continues the MEP from the Democratic Party. “Personally, I am for complete prohibition and I will propose an amendment that eliminates any sense of using remote biometric recognition systems in public places for identification purposes. There is already a certain level of restriction in the Commission’s draft, but I want to propose a complete ban. Compared to the alleged advantages, the risks are too many. Even as regards the fight against crime, the data do not justify the dangers we want to avert in terms of population control ”.
In order to be able to monitor the concrete use of these systems, the establishment of a freely searchable database of all uses of high-risk algorithms by public authorities: “We need to know if and how an artificial intelligence tool is used in sensitive sectors, like almost all those in which the public authority operates”, continues Benifei. “These are, for example, those systems that are used to establish whether a citizen is entitled to a certain subsidy or if he is eligible for citizenship. We need to be able to know if this type of systems is being used. Furthermore, it must always be possible to appeal to a human being when it is believed that one’s right has been violated “.
This is a crucial element to prevent scandals from happening again like the one that hit the Dutch government and the algorithm used to detect scams in the provision of child benefits. But wouldn’t it be the case that a database of this type was also extended to some private companies operating in sensitive sectors? “In the first package there is no enlargement, but I will propose an amendment to this end: we will have to discuss to understand how much we can extend it also to private realitiesbut I believe it must be a substantial expansion ”.
Given the use of algorithms that, for example, monitor the performance of private sector workers – often without a company’s employees being aware of it – this would be an important step: ” Work it is a very hot area and on which we have not inserted any amendments because there was no agreement between us ”, explains Brando Benifei. “I think that the involvement of the social partners must be strengthened, especially in cases where artificial intelligence tools are inserted in the workplace for purposes of surveillance, productivity verification, organization of shifts and more. In some cases, such as that of emotional recognition used to evaluate hiring or firing, these algorithms should probably be completely banned: in addition to being degrading, they represent a clear imbalance of power between employer and employee “.
Among the other amendments already tabled, it is also worth highlighting the cancellation of the obligation of the total absence of errors in the datasets used to train artificial intelligences: “It was unrealistic and even the Commission admitted it”, the MEP explains. “With our amendment, however, it is required to be able to prove that you have done everything possible to avoid errors in the dataset used to train the AI ”.
From law enforcement to surveillance, from transparency on the use of algorithms to the world of work. Among the areas that they are not affected by the new regulation however, there is the one that, perhaps more than any other, has raised very heated discussions in recent years:use of autonomous weapons. “It was decided to exclude the military sphere because the Commission felt it needed its own specificity and we also believe that separate regulation is needed”, concludes Banifei. “Europe has in any case already expressed itself in favor of a ban on killer robots. On the other hand, the combination of these technologies with artificial intelligence poses a serious risk to humanity and it is therefore necessary to proceed with determination to ban their use ”.