1_9788411975094

REPURPOSING LANGUAGE RIGHTS

Guiding the Uses of Artificial Intelligence

VICENTA TASA-FUSTER

ESTHER MONZÓ-NEBOT

RAFAEL CASTELLÓ-COGOLLOS

REPURPOSING LANGUAGE RIGHTS

Guiding the Uses of Artificial Intelligence

COMITÉ CIENTÍFICO DE LA EDITORIAL TIRANT LO BLANCH

María José Añón Roig

Catedrática de Filosofía del Derecho de la Universidad de Valencia

Ana Cañizares Laso

Catedrática de Derecho Civil de la Universidad de Málaga

Jorge A. Cerdio Herrán

Catedrático de Teoría y Filosofía de Derecho.

Instituto Tecnológico Autónomo de México

José Ramón Cossío Díaz

Ministro en retiro de la Suprema

Corte de Justicia de la Nación y miembro de El Colegio Nacional

María Luisa Cuerda Arnau

Catedrática de Derecho Penal de la Universidad Jaume I de Castellón

Manuel Díaz Martínez

Catedrático de Derecho Procesal de la UNED

Carmen Domínguez Hidalgo

Catedrática de Derecho Civil de la Pontificia Universidad Católica de Chile

Eduardo Ferrer Mac-Gregor Poisot

Juez de la Corte Interamericana de Derechos Humanos

Investigador del Instituto de Investigaciones Jurídicas de la UNAM

Owen Fiss

Catedrático emérito de Teoría del Derecho de la Universidad de Yale (EEUU)

José Antonio García-Cruces González

Catedrático de Derecho Mercantil de la UNED

José Luis González Cussac

Catedrático de Derecho Penal de la Universidad de Valencia

Luis López Guerra

Catedrático de Derecho Constitucional de la Universidad Carlos III de Madrid

Ángel M. López y López

Catedrático de Derecho Civil de la Universidad de Sevilla

Marta Lorente Sariñena

Catedrática de Historia del Derecho de la Universidad Autónoma de Madrid

Javier de Lucas Martín

Catedrático de Filosofía del Derecho y Filosofía Política de la Universidad de Valencia

Víctor Moreno Catena

Catedrático de Derecho Procesal de la Universidad Carlos III de Madrid

Francisco Muñoz Conde

Catedrático de Derecho Penal de la Universidad Pablo de Olavide de Sevilla

Angelika Nussberger

Catedrática de Derecho Constitucional e Internacional en la Universidad de Colonia (Alemania). Miembro de la Comisión de Venecia

Héctor Olasolo Alonso

Catedrático de Derecho Internacional de la Universidad del Rosario (Colombia) y Presidente del Instituto Ibero-Americano de La Haya (Holanda)

Luciano Parejo Alfonso

Catedrático de Derecho Administrativo de la Universidad Carlos III de Madrid

Consuelo Ramón Chornet

Catedrática de Derecho Internacional Público y Relaciones Internacionales de la Universidad de Valencia

Tomás Sala Franco

Catedrático de Derecho del Trabajo y de la Seguridad Social de la Universidad de Valencia

Ignacio Sancho Gargallo

Magistrado de la Sala Primera (Civil) del Tribunal Supremo de España

Elisa Speckmann Guerra

Directora del Instituto de Investigaciones

Históricas de la UNAM

Ruth Zimmerling

Catedrática de Ciencia Política de la Universidad de Mainz (Alemania)

Fueron miembros de este Comité:

Emilio Beltrán Sánchez, Rosario Valpuesta Fernández y Tomás S. Vives Antón

Procedimiento de selección de originales, ver página web:

www.tirant.net/index.php/editorial/procedimiento-de-seleccion-de-originales

REPURPOSING

LANGUAGE RIGHTS

Guiding the Uses of Artificial Intelligence

VICENTA TASA-FUSTER Universitat de València

ESTHER MONZÓ-NEBOT

Universitat Jaume I

RAFAEL CASTELLÓ-COGOLLOS Universitat de València

tirant lo blanch

Valencia, 2024

Copyright ® 2024

Todos los derechos reservados. Ni la totalidad ni parte de este libro puede reproducirse o transmitirse por ningún procedimiento electrónico o mecánico, incluyendo fotocopia, grabación magnética, o cualquier almacenamiento de información y sistema de recuperación sin permiso escrito de los autores y del editor.

En caso de erratas y actualizaciones, la Editorial Tirant lo Blanch publicará la pertinente corrección en la página web www.tirant.com.

© TIRANT LO BLANCH

EDITA: TIRANT LO BLANCH

Esther Monzó-Nebot

Rafael Castelló-Cogollos

C/ Artes Gráficas, 14 - 46010 - Valencia

TELFS.: 96/361 00 48 - 50

FAX: 96/369 41 51

Email: tlb@tirant.com

www.tirant.com

Librería virtual: www.tirant.es

ISBN: 978-84-1197-509-4

Si tiene alguna queja o sugerencia, envíenos un mail a: atencioncliente@tirant.com. En caso de no ser atendida su sugerencia, por favor, lea en www.tirant.net/index.php/empresa/politicas-de-empresa nuestro procedimiento de quejas.

Responsabilidad Social Corporativa: http://www.tirant.net/Docs/RSCTirant.pdf

© Vicenta Tasa-Fuster

Language Rights and Artificial Intelligence

Human-Human Interaction

ESTHER MONZÓ-NEBOT

Universitat Jaume I

VICENTA TASA-FUSTER

Universitat de València

RAFAEL CASTELLÓ-COGOLLOS

Universitat de València

This chapter offers an overview of the ideation of artificial intelligence, the hopes and fears it has awakened as an object in our imaginary and as a pervasive component of our present lives. It stresses the potential to both enhance the experience and threaten the livelihoods of human beings, and it highlights lessons to be learned from past industrial and technological revolutions. A central idea of this chapter is that the human vs. machine discussion needs to be reframed to highlight that any machine —also any humanoid machine— is developed by human beings whose ideas respond to their positions in societies. The chapter further identifies the sensitive situation of linguistic minorities in both past and present technological revolutions and stresses how the development of artificial intelligence needs to be matched by the development of a framework that can ensure that we build stronger economies with equal opportunities for all while also building a sustainable world for generations to come.

Keywords: artificial intelligence, language rights, language minorities, job automation, human rights, social stability

1. INTRODUCTION

“What sort of creature man’s next successor in the supremacy of the earth is likely to be?” (Butler 1863, 183). This question served Samuel Butler (signing as Cellarius) to entertain the idea of machines capable of intelligence and the potential consequences of their development. Butler discussed the possibility of creating intelligent machines that are capable of learning and adapting on their own, without the need for human guidance. He suggested that these

Esther Monzó-Nebot / Vicenta Tasa-Fuster / Rafael Castelló-Cogollos

intelligent machines may eventually surpass human intelligence and capabilities (what is now known as the singularity), leading to a situation in which machines would be the dominant form of life on Earth. His article further raised the possibility that these intelligent machines may evolve and adapt in ways that are beyond human understanding, and that they may ultimately pose a threat to the survival of the human race. Finally, it discussed the need for humans to understand and anticipate the potential consequences of creating intelligent machines, and to take steps to ensure that the development of these technologies proceeds in a responsible and ethical manner.

The quoted sentence in such a forward-looking article embeds the conclusion that our supremacy over the Earth as humans will eventually be replaced. To some, this is good news. The hope that Earth will survive the Anthropocene —the era marked by human-induced changes such as climate change, deforestation, loss of biodiversity, and pollution— is far from self-evident (see, e.g., Wallace-Wells 2019). In a less optimistic scenario, the idea may be a source of anxiety, depicting a future that is and has been very present in our collective imaginary when representing artificial intelligence (AI) for as long as AI has existed. The first reference to a robot in the 1920 playwright “R.U.R.” (Rossum’s Universal Robots) (Čapek 1920) portrayed artificial, humanoid beings created to perform the labor of humans (as those now being developed in agricultural robotics, for example; see Martin et al. 2022). These eventually rebelled against and became a threat to humanity. Similar hazards were represented in Williamson’s short story The Mechanical Man (1976) and Wilson’s novel Robopocalypse (2011), along with The Terminator (Cameron 1984) and The Matrix (Wachowski & Wachowski 1999) franchises.

More optimistic representations of the impact of robotics and AI on humanity have also been influential. The now classic science fiction novel I, Robot by Isaac Asimov (1950) presents a vision of a future in which robots and humans coexist peacefully and work together for the benefit of both. Asimov’s novel explored the ethical implications of creating intelligent robots and laid out the famous three laws of robotics: (1) a robot may not injure a human being or, through inaction, allow a human being to come to harm; (2) a robot must obey the orders given to it by human beings, except where such orders would conflict with the First Law; and (3) a robot must protect

its own existence as long as such protection does not conflict with the First or Second Law. Beyond having had a lasting influence on the way AI is depicted in science fiction, these laws have been widely discussed and debated by scientists, philosophers, and ethicists (see, e.g., Veruggio 2002; Sullins 2004, 2006), and they have influenced the development of ethical guidelines for the use of robots in various applications (e.g., European Commission’s High-Level Expert Group on Roboethics 2015; see also Langman et al. 2021).

Similar cooperative depictions are included in Douglas Adam’s The Hitchhiker’s Guide to the Galaxy (Adams 1979). This science fiction comedy features an AI character called Deep Thought, a superintelligent computer designed to find the answer to the ultimate question of life, the universe, and everything. Deep Thought is depicted as a benevolent and wise being, and its intelligence and knowledge are seen as a great asset to all. Another classic, Do Androids Dream of Electric Sheep? (Dick 1968) —a novel by Philip K. Dick on which Ridley Scott’s Blade Runner (Scott 1982) is based— offers a complex image of the relationship between humans and intelligent humanoids. In Dick’s universe, advanced robots called replicants have been created to serve as slave labor on other planets. While the replicants are initially portrayed as dangerous and threatening, a more nuanced view is ultimately presented, suggesting that they may have more complex emotions and desires than humans are willing to acknowledge. In the era of emotional AI, one has to be reminded that emotions can be replicated, but not felt by AI systems (Monteith et al. 2022). The fictionalized complexity of AI humanoids is also faced by Mitchell Hundred, a young programmer, who is invited to test whether an intelligent robot is capable of self-awareness in the comic book series Ex Machina (Vaughan 2014-2016). As Hundred gets to know the robot, he begins to see it as a thinking, feeling being and becomes increasingly convinced of its humanity. How far is that from happening and, most importantly, will we be prepared?

Fictional universes reflect humanity’s fear of technology just as myths such as Icarus did for our ancestors. They hypothesize solutions to a major question haunting all civilizations —will we be killed by our ambitions and inventions? Or will we finally complete a tower that can allow us to reach the knowledge of anything and everything? Every CEO in an AI company at the moment is taking steps to ad-

Monzó-Nebot / Vicenta Tasa-Fuster / Rafael Castelló-Cogollos

vance Asimov’s basic hypothesis that AI and its extensions, such as robotics, will lead to a cooperative, profitable future, and that the massive gains in productivity will positively impact everyone’s quality of life (e.g., Khedkar 2022). To prevent any social unrest which may result in undesirable limitations on the industry, the public needs to be persuaded that intelligent machines will not replace but rather enhance humanity. However, societies, including governments and political and social institutions, must do far more than persuade. They must ensure that progress is paralleled by a conscious planning that, beyond protecting individuals’ lives against machines, can protect the well-being of all groups and individuals against the dominance of some over the rest (see Vallor 2016 and Tasa-Fuster in this volume). The risk of massive quantities of people and entire communities being left behind is as real now as it was in all previous industrial revolutions (see Castelló-Cogollos and Federici in this volume). We must ensure that we all advance together, which requires complex structures of bottom-up and top-down coordination.

This volume presents an overview of the issues and discussions that may lead us to reconstruct and, this time, maybe finalize the tower of Babel assisted by AI. It offers a blueprint of how the different language communities can be guided towards cooperation, by highlighting the challenges that risk frustrating, once again, our concerted efforts to reach further by working together. As an introduction, this chapter offers a brief overview of the key notions involved in AI, arguing that all those involved in coordinating societies, not only those involved in policing their uses of AI, should have a working knowledge of any advances that may be disruptive. A brief overview of the challenges that AI may bring about for societies follows, and the lessons from the past are stressed as a tool to help us embrace the future together. Against this backdrop, minoritized language communities are singled out. Finally, an overview of this volume’s contribution to identifying productive ways to ensure the cooperation of all our communities stresses the relevance of minoritized language communities in our future uses of AI.

2. THE BASICS OF CURRENT AI TECHNOLOGIES

As a long-standing part of our collective imaginary even when it was still only a fantasy, an AI society still seems far away in the future. Ultimately, the dystopian societies as portrayed in 1984 did not come to pass in 1984 and flying hoverboards did not exist in 2015. However, the gap between our dreams and nightmares and current AI technology is closing at a faster pace than ever before. AI can be used nowadays to predict our moods and influence our behavior. Cambridge Analytica was a political consulting firm that used data mining and psychological profiling to influence voter behavior on behalf of its clients. The company was involved in several controversial campaigns, including the 2016 U.S. presidential election and the Brexit referendum in the United Kingdom (Marwick & Lewis 2017; Tufekci 2018). AI was used to manipulate human political will and voting, but AI is also being used to help us save the planet and to cut out our energy expenditure.

AI research is grounded in philosophical notions on knowledge, representation, perception, and action. It has been developed in highly technical areas focusing upon specific applications and tasks. Offering a thorough knowledge of its bases and uses is beyond the scope of this chapter. This section offers a summary introduction to the core concepts that have shaped our perspectives and applications in the field of AI (see also Russell & Norvig [2010] 2021).

2.1 What Is Artificial Intelligence?

The most basic and contested notion grounding the field of AI is intelligence. Intelligence is a concept that has been the subject of academic discussion for centuries, and yet there is no single, universally accepted definition. The ability to learn and adapt to new situations has been a common basis, involving the ability to acquire and apply knowledge and skills with a clear goal, to solve problems and make decisions. Western philosophers have emphasized the role of reason and logical thinking in intelligence since Aristotle. According to this view, intelligence involves the ability to analyze and understand complex concepts and ideas, and to apply logical reasoning to solve problems. More recent scholarship has viewed how intelligence

Esther Monzó-Nebot / Vicenta Tasa-Fuster / Rafael Castelló-Cogollos

varies across cultures (Laboratory of Comparative Human Cognition 1982; Sternberg & Grigorenko 2004) and how it can change over time to prioritize evolving needs within social contexts (Greenfield 1997). Influenced by embodied cognition (Varela, Rosch & Thompson 1991; Gallagher 2005; Gibbs 2006), also known as 4E cognition (embodied, embedded, enacted, and extended), views on intelligence have started to understand its situatedness and relational nature. At the end of the day, humans may only perceive what offers an opportunity for action (Gibson 1966), and that is culturally situated but also influenced by social hierarchies. Within that framework, intelligence is not simply a matter of processing and manipulating symbolic representations through logical rules, but rather it involves the ability to use and adapt to the body’s sensorimotor abilities to interact with the environment and its priorities and possibilities for action in a flexible and adaptive way. Intelligence thus involves the ability to perceive, act, and think about the surrounding world (the world that is relevant to the being) in a way that is grounded in the body’s sensory and motor experiences. In other words, the body and the physical environment collaborate in shaping and constraining the mind, and intelligence is deeply rooted in the physical and social context in which it is developed and used.

In the field of AI, intelligence is often understood as the ability to perform any tasks that are associated with human intelligence, such as learning, problem-solving, decision-making, using language, and, also, adapting to different and new situations. Accordingly, AI systems have used algorithms and mathematical functions to exhibit intelligent behavior in a variety of ways, such as by analyzing data, recognizing patterns, and making predictions. Some AI systems are designed to be general-purpose and can learn and adapt to perform a wide range of tasks, while others are designed to be specialized and excel at a specific task or set of tasks. The latter is what is known as weak AI, a system that can perform whatever it has been trained to do; regardless of the number of tasks they may be able to complete, weak AI systems can only apply the rules they have been trained to apply. Also in AI, embodied cognition has had a major impact (Anderson 2003; Cruz 2019; Mitchell 2020). It has had important effects on the goals of artificial intelligence, on how its applications are conceived and developed. The ability to interact with the environment

features increasingly higher in what AI aims to deliver and the architectures being developed. Through their own interactions, strong AI systems can learn beyond their initial programming, which makes their responses unpredictable —to a certain extent, as their similarity-based learning makes any response dependent on the available data (see Domingos 2015). Taking a step further, the very notion of intelligence has been made dependent on the ability of a being to transform the environment “in its own image” (Kurzweil 2014, 1).

As the awareness of the environment becomes more prominent in how we understand and design intelligence, and the ability to transform it in self-referential ways may become the next step, the way in which we understand this environment will shape our future applications and extensions of AI, and this will —no doubt— shape our shared future. Whereas some AI efforts are developing sensory systems for robots to interact with the environment directly, others are trying to incorporate the environment in indirect ways, such as feeding great quantities of human data, such as pictures or texts, in the systems. Either way, the notion of the environment will shape how the technologies to (directly) capture or (indirectly) analyze (un)selected input are developed. We are only now starting to see what lessons machines can learn on the basis on human-made technologies. Is it surprising that the synthetic minds reproduce synthetic social interactions that reproduce our own mistakes? Scholarship has focused on the challenges of AI to deal with diversity (Crawford & Calo 2016), particularly the ways in which AI systems can perpetuate and amplify existing biases (Savoldi et al. 2021). It is of the utmost importance that the information we enable our machines to learn from can account for an environment that sees each and every one of us (see Monzó-Nebot in this volume). This will be key in ensuring our cooperation and collective and individual wellbeing (Sotala & Yampolskiy 2017b, 2017a).

2.2 How Does AI Learn?

As important as understanding what is being asked from AI is, knowing what we have already taught it to do is key in comprehending its possibilities. Notions such as machine learning, neural networks, deep learning, and natural language processing are central in shap-

Monzó-Nebot

Vicenta Tasa-Fuster

Rafael Castelló-Cogollos

ing the new technologies that have already been released into societies, through smartphones, fitness wearables, or coffee machines. AI systems are often implemented using programming languages and so computer science and programming concepts are permanently in the picture. However, understanding how AI systems are working requires multidisciplinary approaches, and this multidisciplinary understanding is fundamental in guiding future developments.

Machine learning is a technique to achieve artificial intelligence. It involves the use of algorithms and statistical models to enable computers to learn from data, without being explicitly programmed. The machine is provided with large amounts of data and then left to extract patterns from it using a basic algorithm. After finding relationships in the data, the machine can make predictions when new data is presented. Aristotle told us that rule-based reasoning leads to good decisions, and this is the spirit of machine learning. Machine learning has been extensively used in language and speech processing, visual recognition, decision-making, and robotics. Recent years have seen significant progress in areas such as deep learning, a type of machine learning that uses neural networks (an algorithm inspired by the structure and function of human brain) with many layers (hence the “deep” in the name) that allow for data that would otherwise be unrelated to be considered together in pattern finding. This architecture simplifies the process and deep learning algorithms are able to learn and make decisions based on large and complex datasets. Nonetheless, they still require enormous amounts of computing and energy resources (Aggarwal 2019, 455).

Machine learning provides a machine with the capability to learn from data and experience through algorithms. Deep learning does this learning through ways inspired by the human brain, building layers of interconnected neurons to process and transmit information. Better synthesizing human brains, deep learning does a better job at perceiving data and patterns. Thus, such systems are efficient at predicting what humans may do, learning the boundaries between types of data (discriminative or descriptive models) or generating new data based on the distribution of individual classes and the underlying patterns (generative models). When generating new data, AI-based large language models (see also §2.3), for instance, work on the probability and likelihood of specific words (and parts of words)

appearing in specific sequences. Even though they are jointly probable, whether the resulting forms and messages actually exist or are coherent cannot be ensured. So-called hallucinations are these confidently inaccurate results of generative AI that introduce information that may be clearly incoherent with our knowledge of the world, or apparently truthful and yet inaccurate. Al’s potential for introducing erroneous data in mainstream media or specialized environments, shaping opinion, beliefs, and practices, poses considerable challenges for the future of human processes and societies (Braun 2019; Davis 2019).

2.3 What Can AI Do?

Ray Kurzweil, a well-known futurist, predicts that by the year 2045 we will reach the so-called point of singularity and have robots as smart as humans (Kurzweil 2005). Elon Musk predicts that the human mind and body will be enhanced by AI implants which would make us partly cyborgs (see also Ren & Yi 2002). Since the human brain is still a mystery, it is no surprise that AI has a lot of unventured domains. For now, AI is built to work with humans and make our tasks easier. Natural language processing and robotics are two of the areas where AI research has successfully developed advanced applications.

Natural language processing (NLP) is a branch of artificial intelligence and computer science that deals with the interaction between computers and humans using natural language (see Dunn 2022). The goal of NLP is to enable computers to understand, interpret, and generate human language in a way that is both accurate and useful. NLP systems can be applied in a variety of ways, including language understanding, generation, and also interaction. Through parsing, semantic role labeling, and named entity recognition, descriptive models can be used to understand (or represent) the meaning of a sentence, phrase, or a whole document by analyzing its syntax and semantics. On the other hand, generative models can be used to generate text, such as in machine translation, text summarization, and text-to-speech synthesis. These models can build the basis of conversational agents, such as chatbots or virtual assistants, that can interact by processing and responding to human input in natural language. Depending on the task, the approaches to NLP vary. Some of the

most common are rule-based systems, statistical models, and neural networks. As in other domains of AI, past theoretical divisions and battles over competing architectures are fading while a growing number of researchers are adopting hybrid systems in a rapidly growing field. It is however paramount to understand that the most advanced architectures are also the most complex, and their energy demands are also the highest.

Robotics is the branch of AI that deals with the design, construction, and operation of machines equipped with the technology to allow them to be intelligent and able to move on their own. Robots can be programmed to perform a wide range of tasks, such as assembling products, exploring other planets, or performing surgery. Robotic humanoids such as those portrayed in fiction are strong AI systems. However, most available robots are weak AI systems and cannot perform beyond their programmed tasks (Go, Kang & Suh 2020). As technology continues to develop, robots are becoming more advanced, intelligent, and versatile, and they are being used for an increasing number of purposes and industries, such as healthcare, logistics, and manufacturing. However, even though they enjoy a high profile in fiction, robots are only a small part of the AI that pervades our environment in the modern world.

3. WHAT DOES AI MEAN FOR HUMANS?

Not too long ago, while speaking at AI for the Next Era, OpenAI CEO Sam Altman advised against curtailing AI “from a place of fear and despair.” It is worth noticing that Sam Altman is a white male 36-year-old English-speaking American citizen living in the United States whose net worth at the time of writing is $250 million dollars. It is difficult to imagine how any future social development could leave him behind. However, all technological revolutions have given rise to both advantages and disadvantages which are unequally distributed following pre-existing patterns of inequality. Letting those with shared privileges imagine the futures for their imagined communities may force us to repeat the cycle. We need to ask who stands to benefit, and who may be left behind.

Technology and industry have gone hand in hand for most of their history. All previous technological and industrial revolutions have had a complex and multifaceted impact on human rights and values across the planet. On the positive side, they contributed to the development of modern systems of governance and the spread of democracy, which have helped to protect and promote human rights. They have also facilitated the spread of knowledge, ideas, and innovations around the world, which has contributed to the development of human rights movements and the promotion of human rights values. However, the rapid pace of industrialization led to the environmental degradation and the depletion of natural resources that is haunting our present, resulting in particularly negative impacts on the rights of indigenous communities and other groups that rely on natural resources for their livelihoods. Additionally, the industrial revolutions contributed to social and economic inequalities by letting the market decide on the distribution of the resources created, which was particularly damaging to marginalized and disadvantaged groups. To ensure that the benefits of industrialization are shared by all people across the planet and that human rights are respected and protected, these consequences and the actions that caused them need to be unequivocally addressed in any future technological developments, particularly AI.

The impact of AI on human rights is currently the subject of much debate. While AI has the potential to bring about significant benefits for humanity, it can also have negative impacts if it is not designed, developed, and implemented responsibly, as the Cambridge Analytica case regrettably showed. The use of AI raises concerns about the collection, storage, and use of personal data, which can threaten the right to privacy, as AI systems may process vast amounts of data and personal information that can be knowingly used for targeted advertising, political manipulation, or even racial and gender discrimination. Indeed, discrimination can be sustained by unawareness (of both machine trainers and consequently machines) of all our human diversities and by the lack of access of specific groups, including linguistic groups, to the resources that allow for AI development (see Forcada and García-González in this volume). As has been abundantly discussed, AI systems can perpetuate and even exacerbate biases in the data they are trained on, leading to discriminatory outcomes

Esther Monzó-Nebot / Vicenta Tasa-Fuster / Rafael Castelló-Cogollos

(see, e.g., Howard, Zhang & Horvitz 2017; Beretta et al. 2019). For example, machine translation systems have shown a tendency to exacerbate gender bias (Savoldi et al. 2021; Wang, Rubinstein & Cohn 2022), not only in gendered languages, and facial recognition technology has been shown to have higher error rates for people with darker skin tones or those who are gender nonconforming (Leavy 2018). These biases may have outstanding ethical and social consequences in areas such as the weaponization of robots in the military (Singer 2009).

Such collective biases and their transposition to AI may imply serious hazards for fundamental human rights. There is a risk that AIbased decision-making systems may not provide adequate guarantees of due process, particularly in the criminal justice system. The fundamental right to a fair trial can be consequently violated. Depending on their programming and training, AI-driven content moderation can limit the freedom of expression and censorship of certain views or opinions. Likewise, AI algorithms and systems can be designed to sort and present information in ways that are biased and discriminatory, thus limiting access to information for certain groups. Further, the increasing use of AI in the workplace has the potential to automate many jobs, leading to job displacement and unemployment, which can have a significant impact on workers’ rights and economic security, especially those already affected by other social disadvantages and removed from decision-making centers.

There is growing literature on the social, economic, and cultural effects of AI and AI research. The ethical debate has prolifically discussed the idea of building machines with ethical values. Anderson and Anderson (2011) collected a series of essays on the topic, scrutinizing the importance of developing machine ethics, the requirements of the task, challenges awaiting the project, various approaches that have been taken or proposed, and the future of the field overall. Other works have been tackling the social consequences of robotics (Lin, Abney & Bekey 2012), how AI is affecting human social interactions (Turkle 2018), and the prospects of commercializing emotional AI (Monteith et al. 2022). All these discussions imply the need to determine the values that will be privileged in designing an ever-expanding technology, those that may be inherited by future AI systems, but particularly those that we as societies are using when