In the USA a law against discriminatory algorithms
Large companies that use systems based on machine learning (two examples: facial recognition or advertising targeting algorithms) may soon have to inspect the underlying algorithms to verify whether there is impartiality. This is what emerges from a new bill that is promoted in the US by Democratic senators Corey Booker and Ron Wyden, and by Representative Yvette Clarke .
Should the "Algorithmic Accountability Act" bill be approved, the Federal Trade Commission will have the responsibility to establish a set of rules for the evaluation of "highly sensitive" automated systems and the companies will have to verify if the algorithms underlying the functioning of these systems are partial or discriminatory, as if they pose risks to the privacy and safety of consumers.
This is a bill that would go to large companies with access to large amounts of information: in particular, it would target companies with a turnover of more than 50 million dollars a year, which holds information relating to at least 1 million individuals or devices, or acting primarily as a "data broker" with the purchase and sale of consumer data.
The realities that should be subject to the possible law will therefore have the obligation to evaluate all those algorithms that in one way or another are going to affect the legal rights of consumers, those who try to predict and analyze their behavior, those who they analyze large amounts of sensible data and those that are used to monitor large, publicly accessible physical spaces. It can be seen how this could cover a large part of the economic activities linked to the world of technology, putting the various companies in the situation of having to respond in a timely manner in the event that an inspection raises some kind of problem related to privacy, to the security and in general to the rights of the person.
A few months ago the Republican Alexandria Ocasio-Cortez had raised the issue, on the occasion of the celebrations in New York for the Martin Luther King Jr. Day. The algorithms are created by human beings, and these algorithms are still linked to the basic assumptions of 'human being. They are simply automated assumptions, and if any prejudice is not eliminated in them, it simply goes to make it automatic "then Ocasio-Cortez declared.
After all, machine learning systems produce outputs / results that reflect the training they received over time. If the incoming information is somehow conditioned (or the way in which the information is to be processed is conditioned) the same will also happen for the outgoing results. And these conditionings can be present in an unintended way and reflect those involuntary conditionings that revolve around racial, gender and social class issues. Not to be excluded even the possibility of malicious, if some attacker decides to train machine learning systems intentionally with flawed data. A problem that can have practical implications, but above all ethical, not indifferent.
On the new bill, Senator Wyden said: "Computers are increasingly coming into play in the most important decisions that affect Americans' lives - whether or not they can buy a house, have a job or even go to prison. But instead to eliminate injustices, all too often these algorithms depend on partial vices or assumptions or data that actually reinforce discrimination against women and people of color ".