Developing technological solutions without bias requires a previous exercise in which the objectives are set and a constant evaluation of the process. It is crucial to ask ourselves about the impact that development will have on people.
Asking questions, answering them and evaluating the answers: these could be the first three steps in a guide to developing an inclusive artificial intelligence (AI) that minimizes gender, culture, race, and other biases. As technology evolves and is used in new developments and applications, it will be necessary to add questions, reformulate them, discard them for new onse and review the approach. Developing an inclusive AI is a constant exercise that forces us to always think of diversity as an infinite aspiration.
News frequently warn about biases in recruiting systems’ job profiles, security, granting documents, among many other examples. It is the reason why technology companies must pay more attention than ever to the way they approach their product development. Globant published, in February 2019, its Artificial Intelligence Manifesto in which it established the principles it believes and promotes. And that it is in permanent evolution, precisely because life changes.
In times of an accelerating digital transformation, the speed of change sometimes overlooks issues that cannot be ignored.
“Technological solutions come hand in hand with policies that seek better management of resources. There are questions that need to be asked that can be uncomfortable, such as what type of technologies do we develop and for what purpose: is this type of technology in favor of companies, of people? Does it have an impact on people’s chances of developing? This is what, in the long run, impacts the discussion on diversity”, said Vladimir Garay, director of Digital Rights Incidence, a Latin American NGO that defends and promotes human rights in the digital environment.
Decision-making will depend on the answers to these questions. There are three aspects that must be clear when answering them: trust, security and dignity. They are the three pillars on which The Humanized Internet – a global association that calls for the use of technologies to defend the rights of vulnerable people – stands. And it does so from a specific place: there are 1 billion people in the world without an identity, coming from natural disaster areas, expelled by wars or by repressive regimes; this organization proposes that technologies such as AI and blockchain help solve the impossibility of exercising freedom and basic human rights and open up new possibilities.
Is it that simple? Inclusion implies diversity of genders, identities, ethnicities, cultures, beliefs. “For us, when it comes to putting together development teams, the concept is broader, because we are absolutely certain that diversity of criteria and thought is needed,” says Juan José López Murphy, Technical Director & Data Science Practice Lead at Globant. “Diversity gives you the reassurance that you are considering prospects. If a team is not exposed to certain experiences, it cannot be determined whether an issue is being addressed well. What allows for a diverse team is, at least, asking the question,” he added.
“What is my testing group?”, “How is data collection done?”, “How is the response to the algorithm handled?” These are just some of the questions with which Globant faces new developments. However, this “is not enough,” warns Agustín Huerta, Studios SVP, who points out that even when multiple variables are taken into account when considering inclusion: “Not all cultures hear voices in the same way. There are cultures that are more top-down and whose referents, being in a certain role, can lead to taking a certain position ”, he considered. The AI Manifesto is the result of an extensive discussion about what is considered inclusion and diversity.
Both professionals point out the need to figure out how to define diversity in the face of the probability of results, and how to make AI consider all the cases that it seeks to include, so that it is representative of processes that expand possibilities while preventing biases.
“The goal at Globant is for everyone to have a voice and participate. The most important thing is to achieve that openness so that all people feel as peers and feel valued,” says López Murphy. Huerta underlines his position: proposing development from that place “is reinsurance, rather than a guarantee”, because, ultimately, change is constant.
Garay summarizes a perspective, a position on the matter: “If there is indeed a commitment to diversity as a value, the questions I ask myself do not have to be formulated according to my interests but rather seeking the interests of the people on whom the consequences of the technology’s development will appear ”.