Artificial Intelligence
Glossary

AI: Artificial Intelligence, a field of computer science that aims to create systems capable of simulating human intelligence, including the ability to learn, reason and solve problems.

AI ETHICS: A set of principles, standards and guidelines to ensure that the development and use of AI is consistent with ethical values such as transparency, accountability, confidentiality and security.

AI INTERPRETABILITY: The ability to understand and explain the functioning and decisions of AI models, which is essential to ensure the trust, transparency and social acceptability of AI systems.

ALGORITHMIC BIAS: Inadvertent bias or discrimination introduced by AI models due to biased training data or specific model characteristics, which may result in unfair or unrepresentative results.

ARTIFICIAL NEURAL NETWORK: A computer model inspired by the functioning of the human brain, composed of several layers of interconnected neurons, used in deep learning to perform tasks such as image recognition or machine translation.

AUTOMATIC LANGUAGE PROCESSING (ALT): An area that aims to develop methods and tools to enable computers to understand, analyze and produce natural language, such as that used by humans. The aim is to facilitate communication between humans and machines, in particular through applications such as machine translation and text generation.

COGNITIVE NEURAL NETWORK (CNN): A specialized type of artificial neural network designed to analyze spatially structured data, such as images, using convolution operations to extract important features.

COGNITIVE SERVICES: ‘cognitive services’ refer to technologies that use artificial intelligence. They represent a class of technologies that enable computer systems to process data more intelligently and simulate certain human capabilities. These services are designed to analyze, interpret and understand data in a more advanced way by simulating cognitive abilities such as perception, natural language understanding, speech recognition, etc.

CLOUD COMPUTING: A model for delivering IT services over the Internet, enabling easy, on-demand, and often scalable access to computing and storage resources, making data easier to process and analyze.

DEEP LEARNING: A machine-learning method based on multilayered artificial neural networks that extracts complex features from unstructured data, such as images, text, or sound.

ETHICAL AI: An approach to AI that integrates ethical considerations into the design and development of AI systems, with an emphasis on social responsibility, sustainability and respect for human rights.

EVOLUTIONARY AI: A field of research aimed at developing AI systems capable of self-improvement and autonomous adaptation to new environments and tasks, based on the principles of biology and natural evolution.

EXPLICIT AI: AI systems designed to be transparent and explainable in their operation, allowing users to easily understand the decisions and recommendations generated by the system.

GENERAL AI: AI concept with versatile and adaptable intelligence, capable of learning and solving effectively a wide range of tasks and problems, similar to human intelligence.

IoT (Internet of Things): The concept of physical objects connected to the Internet that can collect and exchange data, opening up new possibilities for analysis and decision-making.

LARGE LANGUAGE MODELS (LLM): artificial intelligence models specialized in understanding and generating text, exploiting massive neural networks for various applications such as machine translation and natural language analysis.

MACHINE LEARNING: A subdomain of AI that focuses on developing models and algorithms that enable machines to learn from data and improve their performance over time without being explicitly programmed.

NATURAL LANGUAGE PROCESSING (NLP): A branch of artificial intelligence that aims to enable computers to automatically understand, analyze and generate human language. It encompasses tasks such as speech recognition, language comprehension, machine translation, among others.

RECURRING NEURAL NETWORK (RNN): A type of artificial neural network designed to process sequential data, such as text or speech, using recurring connections that store previous information for contextual decisions.

REINFORCEMENT LEARNING: A machine-learning technique where an officer learns to make decisions by interacting with an environment, receiving rewards or punishment based on his or her actions, allowing the officer to optimize strategies over time.

RETRIEVAL-AUGMENTED GENERATION (RAG): A text generation model that integrates search capabilities to improve the quality and relevance of the information generated. It combines information retrieval mechanisms with text generation techniques to produce more accurate and informative results.

SEMI-SUPERVISED LEARNING: A hybrid approach combining supervised and non-supervised learning techniques, used when only certain parts of the data are labeled, making it possible to make effective use of all available information.

STRONG AI: AI concept with general intelligence equal to or better than human, capable of solving a wide variety of problems autonomously and creatively.

SUPERVISED LEARNING: A machine-learning technique where the model is trained on a labeled data set, allowing it to learn how to predict correct outcomes from new data.

The Artificial Intelligence Act: an EU regulation that aims to establish a regulatory framework for the placing on the market of artificial intelligence, taking into account aspects of security, health and fundamental rights. The Regulation classifies artificial intelligence systems according to the level of risk, ranging from ‘minimal’ to ‘unacceptable’. It prohibits certain uses contrary to European values, such as ‘social credit systems’ or mass video surveillance. “High-risk” AI systems must comply with the strictest regulatory regime for transparency, risk management and data governance.

UNSUPERVISED LEARNING: A machine-learning technique where the model is trained on a set of unlabeled data, allowing it to identify intrinsic patterns or structures in the data without human supervision.

WEAK AI: AI systems designed to perform specific or restricted tasks, without self-awareness or the ability to generalize to other areas.