We are living in an accelerated digital race in which our capacity for analysis is blocked by our willingness to be surprised. Let’s admit it: We are part of the value chain of new and more attractive applications, increasingly personalized, that slip into our hands often inadvertently.
Behind the festival of applications that we use in our day to day lives, with artificial intelligence as the common denominator, the doubt arises as to whether all of this leads us to the best of options or whether along the way we lose the ability to decide or our ethical sense.
Artificial intelligence brings together a wide range of computer techniques and methodologies intended to complement, if not replicate and improve, human cognitive abilities. Its main fields of development are called Machine and Deep Learning and these focus on the development of algorithms and models that enable computers to learn, make predictions and make decisions based on a given set of data and without explicit programming.
Sydney J. Harris, an Anglo-American essayist and journalist, summed it up very well in his works from the end of the last century: “The real danger is not that computers think like people, but that humans start to think like computers.”
Would we admit that artificial intelligence is capable of reproducing human biases such as sexism or racism? Is it possible for a machine or an application to be capable of having preferences?
For some time now, work has been underway to highlight the importance of transparency in computer algorithms on which all these applications are based, including, for example, route planners, data analyzers and validators, mobility service platforms, etc.
For example, an application could prioritize mobility services that benefit certain demographic groups, such as those in certain neighborhoods of the city versus underserved areas. Or it could happen that in the design of an application they unintentionally overlooked realities that affect access to services for specific groups in society, such as the elderly or groups with special sensitivities, if transport routes are discontinuous or are modified based solely on cost-efficiency metrics. It is also a reality that those algorithms that optimize their results based on user responses or demand patterns are prone to inadvertently amplifying pre-existing biases in the data they use.
It is necessary to understand the machine learning process to understand how algorithmic discrimination occurs, know that there are tools to combat it and analyze and apply the applicable regulations and legislation regarding it. Thus enhancing algorithmic justice.
What is meant by an algorithm? fair, impartial and fair? Can algorithms guarantee greater neutrality and efficiency? How can we make our humanity prevail in a calculated world?
The recent exhibition Code and algorithms, (1) which could be seen until February 18 in Brussels, within the framework of the Spanish presidency of the EU, has sought to make this phenomenon understandable by generating questions but above all answers to bring knowledge and reflection closer.
The choice of algorithm thus such as the quality of the basic data set used to learn and generate patterns, -which will be the reference for making decisions- consequently, it has societal and ethical considerations, not to mention the environmental impact.
And in the end, an algorithm is a set of computing commands. We know that the complexity of the calculations as well as their nature, usually proprietary, usually prevents their transparency. But we tend to overlook it. We trust technology so much, in general, in society, which does that we increasingly delegate decisions of greater importance to algorithms.
Philosopher Yuval Noah Harari warns us of this danger in the book “21 Lessons for the 21st Century”, a consistent essay on the consequences of a world in which the main decisions are made by complex computer calculations that are not known or understood.
There have been remarkable cases that make visible the importance of ethics and transparency of algorithms. One that caused an outcry in the press was that of Timnit Gebru, co-director of ethics for AI at Google, who was fired(2) while working on the risks of large AI language models, trained on huge amounts of data. Gebru is known for her strategic work to reveal AI discrimination, develop methods to document and audit AI models, and advocate for greater diversity in research.
Just as we wouldn’t buy a car without a seat belt, let’s normalize algorithm transparency as the new norm. In the interview ‘Make Algorithmic Audits As Ubiquitous As Seatbelts’ (3), Gemma Galdón-Clavell, PhD in Security and Technology Policies, explains the consequences that bad practice can bring and yet the importance of deploying methodologies that lead us to be aware that an algorithm is neutral, responds to the required functionality and has no preferences.
Due to their complexity as well as their nature, usually proprietary, which usually prevents transparency, understanding how they work, in short, that they are FRAND Fair, Reasonable and Non-Discriminatory, it is an objective that especially from the public sector takes on special relevance and leads us to promote the auditability of the algorithms in our applications.
We have a recent framework that promises to drive this need: On March 13, 2024, the European Parliament approved the AI Law.(4) which is expected to enter into force in May.
The AI Law will introduce comprehensive rules to govern the use of AI in the EU, becoming the first major economic bloc to regulate this technology and we will therefore have in it, the legal reference base to promote auditability policies that will give us the guarantees that we need as a society.
REFERENCES (1) https://www.spainculture.be/es/region/bruselas/codigos-y-algoritmos-sentido-en-un-mundo-calculado/ (2) https://www.technologyreview.es/s/12986/fue-terrible-y-despiadado-los-ultimos-dias-de-timnit-gebru-en-google (3) https://www.forbes.com/sites/ashoka/2021/04/26/make-algorithmic-audits-as-ubiquitous-as-seatbeltswhy-tech-needs-outside-help-to-serve-humanity/?sh=19ada67d49a1 (4) https://www.euronews.com/my-europe/2024/03/13/lawmakers-approve-ai-act-with-overwhelming-majority
