Artificial Intelligence can become an ally within business processes, as they allow faster real-time decision-making in different sectors; however, it still cannot be completely trusted, as it may include biases that generate discrimination.
“AI is programmed by prejudiced people.” That is one of the most common phrases in the world of technology and talking about this topic is important, specialists mention. In the future, automated technology will guide important decisions such as who can access a job, a bank loan or life insurance.
The issue has already generated controversy in the past. In 2014, an Amazon team made a software to review resumes, but after a year of use they realized that their algorithm discriminated against women for technical positions and preferred men for the same tasks.
Noemí Pereira, data engineer at NTT Data, highlighted that in order to mitigate gender bias in AI, developers must first be made aware of the problem so that it happens to a lesser extent until it disappears, while at the technical level there are different tools to identify and manage the data on it.
Companies using AI, said Jesús Mantas, global managing partner at IBM Consulting, are facing more internal and external demands to design their algorithms to be fair, secure and trustworthy.
For the expert, one way to reduce bias is to form multidisciplinary and collaborative teams, in which many areas and skills of the company are integrated, with the purpose of feeding quality data to the systems.
According to a , 68% of organizations surveyed recognize that having a diverse and inclusive workplace is important to mitigating bias in AI.
However, these figures also show where to pay attention, as AI teams remain substantially less diverse than their organizations’ workforce: 5.5 times less inclusive of women, as well as four times less inclusive of people. of the LGBT+ community.
Germán García, country manager of Herman Pro —a startup of applied AI business tools—, contributed that elements such as gender, race or age are not quality data for an AI, since these are subjective matters that do not convey the capacity of the person and, rather, can generate discrimination.
Experts who participated in the discussion of the , have also highlighted that in the evolution of AI an intersectional vision is necessary where groups of people with disabilities and Afro-descendants are taken into account.
Andrea Cuernavaca, an expert in Artificial Intelligence strategy at NTT Data, highlighted that since 2018 in the European Union efforts have been made to promote ethics in the development of AI at the regulatory and normative level, thanks to a bill published by the European Commission, where one of the central points is to pay attention to the representativeness of the data to deal with AI bias.
“One of the lines is that the proposal mentions is the generation of a voluntary code of conduct for companies, so that they subscribe to generate more balanced development teams, where more women are included,” he concludes.
