AI must learn about social values ​​so we don't have problems

According to a pair of researchers, professionals working in the creation and development of AI need to “teach” their machines social and human values ​​if we don't want to have problems in the future. According to scientists - Mona Sloane and Emanuel Moss of New York University and New York City University, respectively - there is already evidence that AI can perpetuate discrimination, intensify inequalities and even cause harm in this regard. Therefore, developers should take care that their creations, rather than harming them, serve society to the best of their ability.

Social technology

Researchers argue that we are facing the emergence of a complex “technosocial” system, and in order for it to be fairer, more democratic and balanced, engineers and programmers need to include the widest range of social notions in their AIs to reduce the risk that they are harm our society.

The pair mentioned studies that showed that search engines can discriminate against black women and, as you may know, there are many fears about how autonomous cars will make decisions in the face of crash situations - where they will have to choose between running over a human or hitting a car. on a pole, for example. To avoid such dilemmas and achieve technological development that benefits humanity as a whole, machines must be taught how to be more socially intelligent as well.

Researchers know that there are already professionals concerned with integrating social ideas into their machines so that they act more in accordance with human values, although it is extraordinarily difficult to define and encode this information on a machine - not to mention that engineers and programmers are typically not. trained to understand how different racial and social categories in machine learning can impact organizations and society itself.

Thus, the pair suggests that qualitative methods of social research should be adopted in the creation and development of AIs, since these techniques usually require professionals to reflect on how their interventions impact the universe in which observations are made. In addition, qualitative research can help programmers and engineers gain a better understanding of the sociocultural and historical background behind the different categories that make up our society and how AI should fit into this context.

AI must learn about social values ​​so we don't have problems via TecMundo