European states should strengthen their legislation to protect fundamental rights in the face of artificial intelligence, which is now difficult to challenge when it can be a source of errors and discrimination, according to a report published on Monday.
“Much of the interest is focused on its potential to support economic growth.
How it can affect fundamental rights has received less attention, ”writes the European Agency for Fundamental Rights (FRA), based in Vienna, Austria, in the 100-page document.
Artificial intelligence (AI), a somewhat catch-all expression, refers to technologies that allow machines to imitate some form of real intelligence, to “learn” by analyzing their environment instead of executing simple dictated instructions. by a human developer.
This software, which brings together a vast field of applications (voice assistants, voice and facial recognition systems, advanced robots, autonomous cars, etc.), are now used by public authorities as well as by the medical community, the private sector. and teaching.
42% of European companies use AI
On average, 42% of European companies use AI.
The Czech Republic (61%), Bulgaria (54%) and Lithuania (54%) are the countries where it is most widespread.
Artificial intelligence is particularly popular with advertisers for targeting online consumers through algorithms and "the coronavirus epidemic has accelerated its adoption," according to the report.
FRA investigators carried out around 90 interviews with public and private organizations in Spain, Estonia, Finland, France and the Netherlands.
"One of the risks is that people blindly adopt new technologies, without evaluating their impact before using them," David Reichel, one of the authors of the text, told AFP.
Artificial intelligence can thus violate privacy, by revealing a person's homosexuality in a database for example.
It can also lead to discrimination in employment, if certain criteria exclude categories of the population on the basis of a surname or an address.
When they receive an incorrect medical diagnosis or are denied a social benefit, European citizens do not always know that the decision was taken automatically by a computer.
They are therefore not in a position to be able to contest it or lodge a complaint, even though errors can occur: artificial intelligence, created by humans, is not infallible.
In a recent example, the British Court of Appeal ruled that the facial recognition program used by the Cardiff police force may exhibit racial or gender bias.
“Technology is changing faster than the law.
We must now ensure that the future EU regulatory framework for artificial intelligence is unequivocally based on respect for human rights and fundamental rights ”, underlines FRA Director Michael O'Flaherty.