2. From #wordofthemonth we had the opportunity to open a path of AI Glossary, send us requests for new entries.
3. If you have suggestions of interesting definition, write us and help us build a common language.
Activity - As an activity we understand a frequently reocurring assortment of actions that seek to achieve a particular goal, e.g. cleaning a room or preparing a meal can be done in many different ways, what matters is that the goal - an orderly room or a specific meal - has been achieved.
Anaphora - Anaphora resolution refers to the process of identifying and connecting pronouns or other referring expressions in a text to their corresponding antecedents (previous nouns or phrases). It supports language processes in building effective AI tools.
Benchmark - A benchmark is a standardised test that is used to compare the performance of different methods or thecniques. Benchmarks are instrumental in revealing the strenghts and weaknesses of scientific advances, thereby fostering collaboration between scientific communities.
Common-sense - Common-sense knowledge is what everybody knows, but nobody tells you. Behaving in everyday situations, which object to choose for a task, how to operate in routines. This is acquired by living in the world, simple for a child, but one of the hardest problems in AI history. Computers don't 'live in the everyday world'.
Constructions - Are language building blocks, conventionalised mapping between linguistic forms and meanings, which exist on the phonological, lexical, syntactical and even pragmatic levels of language. They are a key element in building artificial agents, capable of human-like language use.
Ethics - Ethics are moral principles that govern a persons's behavior. For AI, ethics has to do with wheter an AI system has been properly constructed, for example with respect for the privacy of the data sources, and wheter the system is used properly, for example that it is not used for cheating, criminal actions or the hidden manipulation of others.
Explanation - The ability to explain one's actions and decisions can be considered a hallmark of true understanding. Artificial intelligence, therefore, needs to be able to provide reasons for its behavior so that human users can understand and trust it.
Flexibility - As every situation is unique, intelligent systems need to be able to cope with new and unforeseen circumstances - this requires flexibility. MUHAI implements this flexibility via a true understanding of the given context, for example, to know that if no knives are available some scissors or other sharp objects can also be used for cutting.
Framing - Sometimes, when entities in the background (frame) change they affect the meaning of some entity in the foreground (picture). For artificial intelligence this "framing problem" is very hard, as there is a potentially infinite amount of entities in the background that could exert an influence on the meaning of that situation.
Grammar - A grammar captures all linguistic knowledge that is needed to support language-based communication. Grammars are acquired, personal and dynamic. They emerge and evolve ad a result of intentional communication.
Graph - A knowledge graph stores information as a network of facts, connecting different pieces of knowledge together like a web. It helps computers understand and process information more intelligently, making it easier to find answers and make decisions.
Grounding - Symbols can be abstract, referring to imaginary concepts, or they can be grounded. In that case, the symbols refer to categorisations of real world objects.For instance, the symbol RED that we attribute to objects, by sensing the light reflection and identifying if the sensed data falls within a defined sensory subspace.
Knowledge - For human-centric AI copious amounts of formalized knowledge are needed. This machine understandable representation of encyclopedic and common sense knowledge is central for building systems that truly understand the state of affairs at hand in an explainable manner to assist their uman users felicitously.
Hallucination - Humans may experience sensorial misleading; seeing, hearing, feeling, or smelling something that is not real. For generative AI based on large language models, this refers to the production of false information, for example inventing plausible but non-existing book references.
Inference - Is a method of obtainging conclusions from previous knowledge. For a method to count as inference, we often require it to be at least reproducible and invariant to factors considered irrelevant to some purpose. We would also require the method to be reliable: get correct conclusions based on available information.
Intelligence - Intelligence refers to the capacity of a human or artificial agent to solve problems by creatively applying previously acquired knowledge and skills.
Introspection - Humans can think about their own thinking. No other animal can do this, but surprisingly, some AI can. These programs monitor their progress when solving a problem, and they learn from such introspective observations to improve their behaviour.
Language - Language enable us to communicate flexibly using dynamic systems of shared conventionalized forms and patterns. A goal of artificial intelligence is to understand written and spoken language as proficiently as humans do it to mine textual data and provide intuitive interfaces.
Learning - Learning is the process of constructing new knowledge, skills and behaviours based on prior experiences, reasoning and information processing. Learning leads to insightful understanding of the world and prepares us for facing new challenges and making sense of new situations.
Meaning - The meaning of words is neither fixed in some lexicon nor definable by some logic. Meaning arises dinamically in a given situation and is constructed by the context in which a word is used. Therefore, the study of meaning is the study of a process by which meaning is assigned to things.
Memory - From a human-centric perspective memeories are personal collections of the traces of experiences together with the individual semantic conceptualizaton of these experiences. With new memories that are formed continously over time and the corresponding conceptualization that evolve accordingly, human-centric personal dynamic memories are created.
Model - A model, say a model of a house, leaves out many details in favor of key aspects. In AI logic, models define certain aspects of entities able to encompass everything within their range. A configuration of entities that conform to it is called an interpretation of the model.
Narrative - Humans create narratives to make sense of their lives by observing the world, selecting and relating specific items and events. Narratives assign meanings to our experiences and can be spun into larger networks for understanding and valuing the world from different perspectives.
Neurosymbolic - AI merges human-like reasoning with machine learning, integrating symbolic knowledge representation with neural networks. It combines the structured understanding of symbolic AI with the pattern recognition capabilities of neural networks, enabling AI systems to reason more like humans.
Ontology - Ontology is used in fields ranging from philosophy to computer science. In the latter, it is a schema of the data in a particular domain, capturing the meaning of each term with that domain. It is technically an abstraction hierarchy of concepts, with typed relations between them.
Perception
refers to a brain process during which sensorial stimulation is translated into an experience. In AI it refers to the way machines interpret and use data collected in their surroundings by sensors (cameras, microphones, etc.).
Pragmatics - Pragmatics is a linguistic branch that studies language in context. It delves into the ways speakers use language to achieve their communicative goals, examining the intricacies of language beyond its literal interpretations.
Robustness - AI systems are said to be robust if they are able to survive and recover from substantial perturbations. Robustness is an inherent property of evolutionary systems, which are based of variation, selection and self-organisation.
Sentiment - At first sight, it is a concept far from machines. However, AI is extensively employed to perform sentiment analysis by assessing the tone of a given text and extracting information about the writer’s feelings.
Simulation - A simulation is the process in which humans and machines explore "what if" scenarios in a safe space, which allows them to better predict, anticipate and understand the challenges of the world.
Symbol - The emergence of symbols was important for human cognition, as symbols are arbitrary cnventionalized entities that represent something without being that something. Humans are skilled symbol users especially when using language that consists almost entirely of symbolic entities, such as words.
Terms - Using logic, formal ontologies seek to define terms, such as "Knife" or "Fork", to capture human knowledge about such entities. For example, that the object referred to by these terms are types of cutlery. The ensuing logical terminology can help computers in understanding everyday entities.
Trust - The belief that an AI agent will act in a manner that helps the trustor to fulfil their goal in a situation where they are dependent on the other party. Trust becomes very pertinent in vulnerable and risky situations.
Understanding - Our understanding has deeply changed. it is not an abstract and logical reasoning process, but understanding emerges through an embodied mental simulation. When we see someone putting pasta in boiling water we mirror these actions in our mind and understand them.