An interesting and untreated aspect concerning the advent of robots is the phenomenon of racism, understood in a generic sense as unmotivated favors or disfavors to a specific ethnic, religious, sexual group, etc.
The increasingly pervasive use of machine learning algorithms, once generally called expert systems, although rarely is it intended to refer to algorithms that are actually able to self-reprogram, but rather to adaptive algorithm behaviors based on the experience of the past data and successes or failures.
This type of algorithm is increasingly used in highly risky situations.
An example, but it is only an example, may be the use that is made of these methodologies in the pre-selection of personnel.
The problem is not or, at least, is not only linked to an absence of ethics, but is intrinsic to the process of self-learning itself.
Let's take the predominant use of these tools that occurs in the pre-selection phase of a short list of candidates.
If the skimming takes place with evaluation criteria, by human beings, with intentions, even unconsciously, discriminators, the program will refine its selection capacity orienting itself towards that type of discrimination, going, most likely, for its very intrinsic efficiency, to beyond the intentions and practices of human beings who evaluate their performance.
Exemplifying, if an HR manager receives 10,000 hp and of these the robot selects 20 and has a positive feedback on 10 of these, all of blue-eyed candidates, will tend, as time goes on, to select people with eyes of that color, with the consequences, from the ethnic point of view, that this choice implies.
This also in the desirable hypothesis that both the algorithm and the HR manager are in good faith and this happens due to an unconscious discrimination in the human subject and not programmed in the algorithm.
These intrinsic risk characteristics in the method can be, even subtly, exacerbated by a specific programming of the algorithm and / or requests and feedback biased by the human subject.
More subtly, as an example, from a seemingly neutral datum such as the favorite sport, in the USA, where the use of these algorithms is already widespread, a certain ancestral nationality could be favored, without ever asking for sensitive data such as the origins of the candidate.
Generalizing: if an algorithm is able to establish, even surreptitiously, a sensitive, potentially discriminatory data, for example gender, race, religion, etc. analyzing other data in its possession, even if that sensitive datum is not present, potentially those non-sensitive data can be used in a discriminatory way.
It is not an irrelevant ethical problem, because this first skimming, for example in the case of example of the curriculum vitae, can lead to paradoxical and perhaps unwanted results.
In a system like the present one, where the job offer is constantly low and the cvs sent are numerous for each position, some subgroups, maybe even unthinkable, could be completely excluded from the world of work without even knowing why and by whom. .
My example concerning personnel selection is just one of many possible, let's think about insurance, both on life and damages, on loans, etc.
Any field can be, sooner or later, included in the use of decision-making algoritms with self-learning criteria, leaving free rein to no ethic and unregulated algorithms can potentially produce incalculable damages.
