Artificial intelligence in selection processes
Artificial intelligence is applied daily in recruitment and selection processes in Brazil and around the world.
The first question that arises in connection with the use of AI is whether employers should inform candidates that their resumés and data will be analyzed by an AI system. Although Brazil does not have legislation that requires employers to do so, informing candidates appears to be part of good corporate data protection and privacy practice, could be included in the information on personal data processing that companies are required to give to data subjects under the General Data Protection Law – LGPD.
A second question raised by the growing use of AI in selection processes is whether AI systems – which are still very controversial, including from an ethical standpoint – have reduced or aggravated the incidence of discrimination. In other words, does AI also suffer from discriminatory biases?
These two questions point to more fundamental issues: who is responsible for the data used by the system's machine learning algorithm, and what is the employer's real intention?
When provided with data having a given profile (even if that profile reflects the reality of the hiring company's existing employees, or the resumés received by the company), the AI system will tend to select candidates with similar attributes, potentially replicating pre-existing patterns. Often, this results in a statistically significant exclusion of women and Black individuals in comparison with white men.
Thus, those who develop AI systems – and those who use them – are responsible for any discriminatory bias an AI-based candidate selection tool may exhibit, since the bias is inherent in the system, and went undetected because of insufficient testing prior to use. As for the employer's intentions, AI recruitment and selection systems allow the insertion of filters to identify candidates with valued characteristics and qualifications. Although illegal, it is possible to create a discriminatory bias in the selection of job candidates.
Still, the use of AI systems in recruitment and hiring processes has become an inescapable reality. The pressing question now is to what extent programs can be corrected to eliminate what has become known as algorithmic discrimination. Are there ways to monitor and revise systems so that they are not fed with data reflecting only part of the population, thus reproducing current discriminatory biases? Are there ways to oversee the use of IA systems, and hold employers accountable if the system shows discriminatory biases in its tracking and selection of candidates?
Although there are still no definitive answers to these questions, it is certain that if an employer uses an AI selection tool that has a discriminatory bias, the employer assumes the risk of claims by excluded candidates and actions by the Public Labor Prosecutors for violation of the principle of non-discrimination under article 7(XXX) of Brazil's Federal Constitution and ILO Convention 111: Candidates are entitled to equal opportunities when they have the same skills, and any distortion that constitutes discrimination, whether committed by a human being or a machine, cannot be justified.
Last but not least, we must remember that while diversity-related data can be captured by AI systems, measurement of real inclusion – i.e. the extent to which marginalized groups are effectively included in a given work environment, regardless of the numbers and other objective data – continues to be a complex task, and one that requires human attention.
