Artificial Intelligence in Spanish

The cultural dimension of AI development: language as the architecture of thought

Language is far more than a communication system; it is the structure upon which we build our thinking, organise our reality, and transmit our culture. This premise, fundamental to cognitive linguistics, takes on critical relevance when we talk about Artificial Intelligence. Large language models are not simple translators or word processors: they are systems that learn to "think" in the language they are trained on, absorbing not only its grammar and vocabulary, but also the cultural patterns and conceptual structures of that language.

When a language model is trained predominantly on English-language texts, it learns to produce text in the way that language does it. This means assimilating Anglo-Saxon argumentative structures, ways of organising information that are characteristic of American or British culture, and reasoning patterns that may be alien to other linguistic cultures. The result is an inherent bias that goes beyond the purely linguistic to become a cultural and cognitive bias.

The invisible bias: when AI thinks in another language

This bias manifests in subtle but profound ways in the quality and accuracy of responses generated by AI models. A model trained predominantly in English may translate correctly into Spanish, but its responses will bear the hallmarks of Anglo-Saxon thinking: the way arguments are structured, a more direct communication style, the organisation of ideas that prioritises efficiency over context, or even cultural references assumed to be universal when they are in fact specific to a particular cultural tradition.

Consider some concrete examples: reputation analysis of a bank based on news or social media that fails to capture the use of humour or irony, customer service systems that respond in culturally inappropriate ways, or educational tools that teach reasoning structures that are foreign to our own pedagogical contexts.

Linguistic corpora: the technological foundation of AI in Spanish

Contemporary Artificial Intelligence is based on a fundamental principle: learning from data. Unlike traditional expert systems, where programmers manually encoded rules and knowledge, large language models learn by identifying patterns in enormous volumes of information. This transfer of human knowledge to machine learning models is the foundation of all current artificial intelligence. If we want AI models to be capable of solving linguistic tasks, we must first show them examples of how humans solve those tasks. We understand "solved task" as information encoded in different formats: text, image, audio, or video. In the case of natural language processing, and in order to achieve systems with a high level of linguistic competence that can communicate fluently with us, we need to transfer to these systems the greatest possible number of human-produced texts. These structured sets of textual data are what we call corpora.

A linguistic corpus is far more than a simple collection of texts. It is a systematic, structured, and representative set of real linguistic productions that captures how a language is used in its natural contexts. Corpora are to language models what lived experiences are to a human being learning their mother tongue: the primary source of knowledge about how communication works.

When we talk about the corpora or datasets used to train large language models such as GPT-4 or Claude, we are talking about an extraordinary variety of sources: books of every kind and genre, content written on websites, large repositories of world knowledge such as Wikipedia, but also less formal linguistic productions such as those we write on social media, in public reviews of products or services, and even in emails. This variety is essential: it allows these language models to process and handle text in different languages, registers, and styles, adapting to the communicative context.

If we want AI models to be capable of solving linguistic tasks, we must first show them examples of how humans solve those tasks. In the case of natural language processing, and in order to achieve systems with a high level of linguistic competence that can communicate fluently with us, we need to transfer to these systems the greatest possible number of human-produced texts that are authentic and representative.

The strategic importance of AI in Spanish: the future speaks Spanish

Developing AI in Spanish is not merely a matter of convenience or service improvement: it is a question of digital and technological sovereignty. In a world where AI is radically transforming how we work, communicate, learn, and create, depending exclusively on technologies developed in other cultural and linguistic contexts means ceding control over fundamental aspects of our society.

Technological sovereignty in AI entails the capacity for independent development, without relying exclusively on imported solutions; control over the data that feeds AI, which is fundamental information about how we think, communicate, and organise our reality; and autonomy in innovation, enabling us to innovate according to our own needs, priorities, and values.

Spanish is spoken by more than 500 million people worldwide, representing an enormous market that is largely underserved by current AI technologies. Developing solutions specifically for this market represents an extraordinary economic opportunity.

Companies that succeed in developing AI technologies that are genuinely competent in Spanish will gain privileged access to growth markets such as Latin America, a region undergoing rapid digital acceleration; specialist niches ranging from educational applications to healthcare tools; and will foster a new hub of innovation from the Global South that is not centred on Silicon Valley.

AI that works well in Spanish is fundamental to the digital inclusion of millions of people. It is not simply a matter of technology being "translated", but of it being genuinely accessible and useful for Spanish-speaking users of all educational levels, ages, and social backgrounds.

Digital inclusion through AI in Spanish means access for all, particularly older people, users with lower levels of digital literacy, and rural communities; educational tools that teach in authentic Spanish with culturally relevant examples; and more accessible public services, from healthcare to public administration.

Finally, developing AI in Spanish is a contribution to a more diverse and balanced global technology ecosystem. The concentration of AI development in Anglo-Saxon contexts risks producing technologies that, whilst technically sophisticated, reflect a culturally limited view of the world.

By decentralising the generation of digital solutions, the participation of diverse societies in the evolution of AI is encouraged, ensuring that different perspectives, values, and ways of seeing the world form part of the new technological landscape. This not only benefits Spanish speakers, but enriches the global development of AI, making it more robust, adaptable, and truly universal.

At GNOSS, we are committed to this transformation, developing AI solutions that not only understand Spanish, but think in Spanish. This means using authentic corpora, prioritising texts written originally in Spanish by native speakers; developing specific models trained to capture the particularities of the Spanish language; attending to dialectal diversity, recognising and respecting all variants of the language; and pan-Hispanic collaboration with institutions, universities, and companies from across the Spanish-speaking world.

Our goal is to create an AI that adapts to the way we live, think, and create in Spanish, thereby contributing to a more inclusive, equitable, and culturally diverse digital future.

Because the future does not only speak Spanish: it thinks in Spanish.