Glossary

Key terms used across mishmash.no, each explained at three levels. Use the switcher above to choose: Simple is one plain sentence, Standard is the everyday explanation, and Advanced adds precision and the settled definitions from standards, policy, the research literature and Norwegian sources.

The same texts power the inline explanations: dotted-underlined terms like that one, which you can click to unfold, so this list and the in-text explanations always match, at whichever level you read.

Switch to Advanced to see where each definition comes from. Each term then also shows how standards, laws, research and Norwegian sources define it.

Where a term has a settled definition elsewhere, it is given underneath, with its source, in this order: international standards (ISO/IEC 22989 and related standards, NIST), policy and law (the EU AI Act, OECD, UNESCO, national strategies), the research literature, and the Norwegian definition. Norwegian definitions are quoted from Teknologirådet’s glossary of artificial intelligence, which Språkrådet has reviewed, and from Norwegian regulations and strategies; the Norwegian terms in this glossary follow Teknologirådet’s list. Quotation marks mean the wording is the source’s own; without them it is a close paraphrase. Draft standards are marked as drafts.

AI

computer programs that can learn and solve problems, a bit like how you learn from practiceAI is short for artificial intelligence: computer systems that do things we usually think need human thinking, such as recognising pictures, writing text or composing music, mostly by learning from large numbers of examplesan umbrella term for computational systems that perform tasks associated with intelligence, from hand-written rule-based systems to today’s learning-based models. The Norwegian national AI strategy defines AI systems by what they do: act, physically or digitally, on interpreted data in order to reach a goal, with some systems adapting to how earlier actions affected their surroundings In standards: “Engineered system that generates outputs such as content, forecasts, recommendations or decisions for a given set of human-defined objectives.” ISO/IEC 22989:2022, 3.1.4 (AI system) In policy: “A machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.” EU AI Act 2024, Art. 3(1) (AI system) In policy: “An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.” OECD AI Principles, as revised 2023 In the literature: “The science and engineering of making intelligent machines.” McCarthy 2007 Norwegian definition: «Kunstig intelligente systemer utfører handlinger, fysisk eller digitalt, basert på tolkning og behandling av strukturerte eller ustrukturerte data, i den hensikt å oppnå et gitt mål.» Nasjonal strategi for kunstig intelligens 2020 Norwegian definition: «Et maskinbasert system som er konstruert for å operere med varierende grad av autonomi, som kan vise tilpasningsdyktighet etter idriftsetting, og som, for eksplisitte eller implisitte mål, ut fra inndataene det mottar, utleder hvordan man genererer utdata som prediksjoner, innhold og anbefalinger, eller beslutninger som kan påvirke fysiske eller virtuelle miljøer.» KI-forordningen art. 3 nr. 1, uoffisiell oversettelse (DFD 2025) Norwegian definition: «Kunstig intelligens er informasjonsteknologi som justerer sin egen aktivitet og derfor tilsynelatende framstår som intelligent.» Store norske leksikon

AI agent

an AI that can do a job on its own, step by stepan AI system designed to carry out a task, or a set of tasks, on its own, including deciding what to do next; MishMash builds musical agents that improvise with performersa system that perceives its environment, chooses actions towards a goal and acts with some autonomy, possibly over many steps and with tools. In interactive music research a musical agent is such a system that listens and plays in real time; in current industry usage the word covers language-model systems that plan and carry out tasks In standards: “Automated entity that senses and responds to its environment and takes actions to achieve its goals.” ISO/IEC 22989:2022, 3.1.1 In standards: “Software programs that can interact with their environment, receive information, and undertake self-directed actions in service of a larger, externally-specified goal.” NIST AI 100-2e2025, glossary (agent) In the literature: “An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through effectors.” Russell and Norvig 1995, § 2.1 In the literature: “Musical agents are artificial agents that tackle musical creative tasks, partially or completely.” Tatar and Pasquier 2019 Norwegian definition: «KI-system utformet for å utføre en bestemt oppgave eller et bestemt sett av oppgaver på egen hånd.» Teknologirådet 2026

AI literacy

knowing enough about AI to use it wisely and ask good questions about itknowing enough about AI to use it, question it, and understand what it does — for everyone, not just programmersa set of competencies for evaluating, communicating with and using AI critically (Long and Magerko 2020). It is a theme in MishMash’s education work (WP4) and includes understanding what AI-generated content is, how it is made, and what it costs In policy: “Skills, knowledge and understanding that allow providers, deployers and affected persons, taking into account their respective rights and obligations in the context of this Regulation, to make an informed deployment of AI systems, as well as to gain awareness about the opportunities and risks of AI and possible harm it can cause.” EU AI Act 2024, Art. 3(56) In the literature: “A set of competencies that enables individuals to critically evaluate AI technologies; communicate and collaborate effectively with AI; and use AI as a tool online, at home, and in the workplace.” Long and Magerko 2020 Norwegian definition: «Ferdigheter, kunnskaper og forståelse som gjør det mulig for leverandører, idriftsettere og berørte personer, idet det tas hensyn til deres respektive rettigheter og forpliktelser i forbindelse med denne forordningen, å treffe en informert beslutning om idriftsetting av KI-systemer samt å få bevissthet om mulighetene og risikoene ved KI og hvilken mulig skade det kan forårsake.» KI-forordningen art. 3 nr. 56, uoffisiell oversettelse (DFD 2025)

algorithm

a list of steps a computer follows to do somethinga precise sequence of steps for solving a problem; in everyday speech also the recommendation and ranking systems that decide what we see onlinea finite, well-defined procedure for computing a result from an input. Machine-learning algorithms are the procedures that fit models to data, distinct from the resulting models; the looser public sense, algorithms as curating and ranking systems, is what much of the social critique of AI addresses In standards: “Set of rules for transforming the logical representation of data.” ISO/IEC TR 24028:2020, 3.3 In the literature: “Informally, an algorithm is any well-defined computational procedure that takes some value, or set of values, as input and produces some value, or set of values, as output.” Cormen et al. 2009, § 1.1 Norwegian definition: «Algoritme er i matematikk og databehandling en fullstendig og nøyaktig beskrivelse av fremgangsmåten for løsning av en beregningsoppgave eller annen oppgave.» Store norske leksikon

artistic research

doing research by making artresearch carried out through making and performing art, where the artistic process itself produces new knowledgeresearch in which artistic practice is both method and result: knowledge is produced through making, performing and reflecting, and documented in works as well as text. In Norway it is a recognised research track alongside scientific research, and MishMash uses it as an entry point for critical work on AI In policy: “Research through means of high level artistic practice and reflection; it is an epistemic inquiry, directed towards increasing knowledge, insight, understanding and skills.” Vienna Declaration on Artistic Research 2020 In the literature: “Art practice qualifies as research if its purpose is to expand our knowledge and understanding by conducting an original investigation in and through art objects and creative processes.” Borgdorff 2006

bias

when an AI treats some people or things unfairly because of what it learned fromsystematic skew in what an AI system produces, often inherited from its training data, so that some groups, styles or perspectives are favoured over otherssystematic error in a model’s outputs that disadvantages particular groups or perspectives. It can enter through data collection, model design, training objectives or use; in creative AI it shows up as, for example, an over-representation of Anglo-American styles. Distinct from the statistical bias of the bias–variance trade-off In standards: “Systematic difference in treatment of certain objects, people or groups in comparison to others.” ISO/IEC 22989:2022, 3.5.4 In the literature: “We use the term bias to refer to computer systems that systematically and unfairly discriminate against certain individuals or groups of individuals in favor of others.” Friedman and Nissenbaum 1996 Norwegian definition: «Systematiske skjevheter i et KI-system som kan påvirke resultatene modellen produserer.» Teknologirådet 2026

Board

the group that decides the big things about MishMashthe centre’s governing body, with representatives of the partner institutions; it makes the strategic decisions about the centre’s directionthe formal governance body of the centre, responsible for oversight, strategic decisions and the quality and impact of the research. Not to be confused with the Council, which advises, or the Scientific Advisory Board, which evaluates

budget partner

a partner that gets money from MishMash and helps run itan institution that has signed the consortium agreement and receives funds directly from MishMash. Each budget partner has a seat on the Councila partner institution bound by the consortium agreement, with its own share of the budget, its own recruitment of fellows and postdocs, and a representative on the Council. Other partners take part in the centre’s work without receiving funds directly

co-creation

making something togethercreating together, between people or between people and machines; in MishMash the word describes creative processes where the contributions are genuinely sharedjoint creative production in which initiative, agency and authorship are distributed among the participants. Used across MishMash for human–human, human–machine and mixed settings; CoCreative AI names the systems built for it In the literature: Any act of collective creativity: creativity that is shared by two or more people. after Sanders and Stappers 2008

CoCreative AI

AI that makes things together with people, more like a bandmate than a toolAI systems designed for genuine partnership between humans and machines, where both contribute to the creative resultAI systems designed as partners in a creative process, where initiative, control and contribution are shared between human and machine (Anscomb 2024). Pioneering such systems is MishMash’s stated objective, with human agency, environmental sustainability and democratic access to the technology as design requirements In the literature: AI treated as a collaborator rather than a tool: a partnership in which both the human and the system contribute to the creative outcome. after Anscomb 2024

code of conduct

the rules for how people in MishMash treat each otherthe centre’s guidelines for how everyone taking part treats one another, and where to turn if something is not right. They apply in every MishMash setting, in person and onlinea written statement of the values the centre works by and the behaviour expected of everyone taking part, with the route for raising a concern. MishMash is not a legal entity and carries no employer responsibility, so anything that calls for a formal process follows the rules of the institution where it happened

computational creativity

research on whether computers can be creativethe research field studying whether and how computer systems can behave in ways we would call creativea research field at the meeting point of AI, cognitive science and the arts that builds and evaluates systems taking on creative responsibilities (Colton and Wiggins 2012). It asks not only whether the output is good but whether the process would count as creative to an unbiased observer In the literature: “The philosophy, science and engineering of computational systems which, by taking on particular responsibilities, exhibit behaviours that unbiased observers would deem to be creative.” Colton and Wiggins 2012

consortium

a group of organisations that team upa group of organisations teaming up on a shared goal — MishMash gathers universities, research institutes, and cultural and industry partners from all over Norwaythe formal grouping of institutions bound by the consortium agreement behind the centre: universities, research institutes, and cultural and industry partners. Institutions join as partners, by signing the consortium or an associate-partner agreement; individuals join as members In policy: “In EU grants, the consortium is normally composed of the key project participants, i.e. typically the coordinator and the other beneficiaries, affiliated entities and associated partners.” EU Annotated Grant Agreement 2025

Council

a group with one person from each partner that gives MishMash advicethe body where the partner institutions are represented; it gives strategic guidance and keeps the centre and its partners coordinatedan advisory body with one representative per partner institution, responsible for strategic direction, policy guidance and alignment with the consortium’s objectives. Its meeting notes are published on the organisation pages

Creative AI

computer programs that can make new pictures, music or stories on their ownmachine systems that can produce results that are both novel and meaningful — not just random, and not just copiesmachine systems whose output is novel and meaningful and can stand on its own without a human finishing it (de Vries 2020). The term describes what the systems produce, not a claim that they are creative in Boden’s sense; whether they are, and how their output differs from human creativity, is one of MishMash’s research questions In the literature: Machine systems able to produce novel and meaningful results that stand independently. after de Vries 2020

creativity

coming up with something new that is also good, like a song no one has heard beforethe ability to form ideas or works that are both new and worth having — the human trait MishMash studies AI throughMishMash follows Boden in treating creativity as the production of ideas or artefacts that are new, surprising and valuable, and distinguishes what is new to the person from what is new to everyone. The centre studies AI through this trait rather than through intelligence in general In the literature: “The ability to come up with ideas or artefacts that are new, surprising and valuable.” Boden 2004

MishMash cube

a picture of a box that shows how MishMash is put togetherthe diagram used to explain the centre’s structure: the seven work packages on one face, the three approaches (create, explore, reflect) on another, and the perspectives of machines, humans and society on the thirda visual device proposed during the application workshops to show that the centre’s themes, approaches and perspectives are independent dimensions that cross in every project. The centre’s “mishmash” of theories and methods is organised into this structured “mesh”, which is also where the name MeshUp comes from

cultural heritage

old things worth keeping, like songs, pictures, buildings and storiesthe objects, recordings, documents, places and practices a society inherits and chooses to preserve; in MishMash the theme of WP6, on archives, libraries and museumstangible and intangible heritage, from artefacts and archives to living traditions such as folk music and Sámi joik. WP6 studies how AI can help transcribe, organise, access and represent it, with particular attention to minority cultural expressions and to the ethical and legal limits that apply In policy: “The practices, representations, expressions, knowledge, skills – as well as the instruments, objects, artefacts and cultural spaces associated therewith – that communities, groups and, in some cases, individuals recognize as part of their cultural heritage.” UNESCO 2003, Art. 2(1) (intangible cultural heritage) In policy: Monuments, groups of buildings and sites of outstanding universal value from the point of view of history, art or science. after UNESCO 1972, Art. 1

embodied AI

AI with a body, like a robot that can move and senseAI with a body — systems that sense and act in the physical world, like robots or musical machines on a stageAI that learns and acts through a body situated in an environment, so that perception, action and learning are coupled (Duan et al. 2022). In MishMash it ranges from robots and musical machines on stage to interfaces that respond to a performer’s movement In the literature: AI that learns through interaction with an environment from its own point of view, rather than from static datasets. after Duan et al. 2022

explainability

being able to explain why an AI did what it didhow far people can understand why an AI system produced a given result; deep-learning models are often hard to explainthe degree to which a model’s behaviour can be accounted for in human terms, either by design (interpretable models) or after the fact (explainable AI, XAI). For creative practice it matters as much for control as for accountability: an artist needs to know which knobs do what In standards: “Property of an AI system to express important factors influencing the AI system results in a way that humans can understand.” ISO/IEC 22989:2022, 3.5.7 In standards: “Explainability refers to a representation of the mechanisms underlying AI systems’ operation, whereas interpretability refers to the meaning of AI systems’ output in the context of their designed functional purposes.” NIST AI 100-1 (AI RMF 1.0), 3.5 In the literature: “Given an audience, an explainable Artificial Intelligence is one that produces details or reasons to make its functioning clear or easy to understand.” Arrieta et al. 2020

FAIR principles

four rules that make research data easy to find and reusefour principles for research data: findable, accessible, interoperable and reusable. Data that follow them have a persistent identifier, are described so that people and machines can find them, can be fetched by standard means, work together with other data, and carry a licence that says how they may be useda set of guiding principles for the management and stewardship of research data, published in 2016 by a group of researchers, publishers and funders. Findable: a globally unique persistent identifier, rich metadata that include the identifier, and registration in a searchable resource. Accessible: retrievable by the identifier over a standard, open protocol, with metadata that remain available even when the data no longer are. Interoperable: formal, shared languages for knowledge representation, FAIR vocabularies, and qualified references to other data. Reusable: rich attributes, a clear usage licence, provenance, and community standards. The principles say nothing about openness in itself; closed data can be FAIR when the conditions for access are stated In the literature: The FAIR Guiding Principles for scientific data management and stewardship: data and metadata should be Findable, Accessible, Interoperable and Reusable, for machines as well as for people. Wilkinson et al. 2016

generative AI

AI that makes new things, like a picture from a descriptionAI that can generate new content such as text, images, sound, video or program code, from patterns learned in large collections of examplesmachine-learning models, today mostly large neural networks trained on very large datasets, that produce new samples resembling their training data in response to a prompt. Commercial examples are Dall-E for images, ChatGPT for text and Suno for music; MishMash studies such systems as material, instrument and object of critique In standards: AI system based on techniques and models that aim to generate new content, such as text, audio, code, video or images. ISO/IEC 22989:2022/DAmd 1 (draft, 2025), 3.1.38 In policy: “The class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content. This can include images, videos, audio, text, and other digital content.” US Executive Order 14110 (2023), as cited in NIST AI 600-1 In the literature: “The term generative AI refers to computational techniques that are capable of generating seemingly new, meaningful content such as text, images, or audio from training data.” Feuerriegel et al. 2024 Norwegian definition: «KI som kan generere nytt innhold som tekst, bilder, lyd, video og programkode.» Teknologirådet 2026 Norwegian definition: «Generativ kunstig intelligens er teknikker innen maskinlæring der målet ikke bare er å analysere data, men også å lage nye data, for eksempel i form av tekst, bilder, lyd, video eller programmeringskode.» Store norske leksikon

large language model

AI that has read huge amounts of text and can write text backan AI model trained on very large amounts of text so that it can process and produce natural language; ChatGPT is built on onea neural network, typically a transformer with billions of parameters, trained to predict the next token in text and then tuned with human feedback. It learns statistical patterns in its training data, which lets it answer, summarise, translate and generate, and which also explains its characteristic failures, such as inventing facts In standards: Machine learning model that encodes natural language with many parameters to perform natural-language-processing tasks. ISO/IEC 22989:2022/DAmd 1 (draft, 2025), 3.3.20 In policy: “An AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications.” EU AI Act 2024, Art. 3(63) (general-purpose AI model) In the literature: “Large language models (LLMs) refer to Transformer language models that contain hundreds of billions (or more) of parameters, which are trained on massive text data.” Zhao et al. 2023, § 2.1 Norwegian definition: «KI-modeller som er trent på svært store mengder tekst for å behandle og generere naturlig språk.» Teknologirådet 2026 Norwegian definition: «En språkmodell er en statistisk modell av et språk som brukes innen språkteknologi. Modellen gir en sannsynlighetsfordeling over sekvenser av ord og kan derfor brukes til å analysere og generere tekst basert på naturlige språk, slik som norsk.» Store norske leksikon (språkmodell)

LCA

a way to measure how much something harms the environment over its whole lifelife-cycle assessment — a method for measuring the total environmental footprint of a product or system, from creation to disposallife-cycle assessment: a standardised method (ISO 14040) for compiling the inputs, outputs and environmental impacts of a product system from raw materials to disposal. MishMash applies it to the environmental cost of training and running AI models In standards: “Compilation and evaluation of the inputs, outputs and the potential environmental impacts of a product system throughout its life cycle.” ISO 14040:2006, 3.2

machine learning

when a computer gets better at something by practising on lots of examplescomputer methods that get better at a task by learning from examples, instead of following rules written by handthe subfield of AI in which a model’s parameters are fitted to data so that performance on a task improves with experience (Mitchell 1997). It spans supervised, unsupervised and reinforcement learning and underlies nearly all current creative AI, in contrast to the rule-based systems of early computer art and music In standards: “Process of optimizing model parameters through computational techniques, such that the model’s behaviour reflects the data or experience.” ISO/IEC 22989:2022, 3.3.5 In the literature: “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks T, as measured by P, improves with experience E.” Mitchell 1997 Norwegian definition: «Maskinlæring er en spesialisering innen kunstig intelligens hvor man bruker statistiske metoder for å la datamaskiner finne mønstre i store datamengder. Man sier at maskinen «lærer» i stedet for å bli programmert.» Store norske leksikon

Management Group

the small group that runs MishMash from day to daythe group that handles the day-to-day coordination and administration of the centre. It meets weeklythe centre’s executive group: the director and the two deputy directors, who carry one of the centre’s three perspectives each, machines, humans and society, together with the administrative coordinator and the research advisor. It meets weekly, and prepares what the Board decides

member

a person who has joined MishMashan individual who has signed up for one or more work packages; most are connected to a partner institution, but MishMash is open to researchers without onethe individual level of affiliation. Members take part in work package activities and MeshUps, are listed in the directory, and can apply for seed funding together with other members; membership carries no funding in itself

MeshUp

MishMash’s weekly online meeting, every Thursdaythe centre’s weekly half-hour online meetup on Thursdays at noon, with a short presentation and discussion; open to all members and numbered from #01the centre’s main meeting format, which keeps a distributed consortium of more than 200 people in contact: a weekly Zoom session with a few announcements, a fifteen-minute presentation and moderated questions. Each MeshUp is listed as an event on the site

multimodal

AI that can handle more than one kind of thing, like both pictures and wordsan AI model that can process several types of data, such as text, images, sound and video, and combine themmodels trained on more than one modality, so that representations are shared across, for example, text and images or audio and movement. Multimodality matters for the arts, where meaning is rarely carried by one channel alone; a model that handles only one type of data is unimodal In standards: “A model that processes and relates information from multiple sensory modalities that each represent primary human channels of communication and sensation, such as vision and touch.” NIST AI 100-2e2025, glossary (multimodal models) In the literature: “Multimodal machine learning aims to build models that can process and relate information from multiple modalities.” Baltrušaitis, Ahuja and Morency 2019 Norwegian definition: «En KI-modell som kan behandle flere typer data, som tekst, bilder, lyd og video.» Teknologirådet 2026

neural network

a computer program built from many tiny connected parts that learn together, loosely inspired by the braina machine-learning model made of layers of simple units whose connections are adjusted during training; deep learning means networks with many layersa function composed of layers of weighted sums and non-linearities, whose weights are fitted by gradient descent on training data. Deep networks with many layers are behind today’s image, sound and language models; their strengths are scale and flexibility, their weaknesses opacity and a hunger for data and computing power In standards: “Network of one or more layers of neurons connected by weighted links with adjustable weights, which takes input data and produces an output.” Deep learning is the “approach to creating rich hierarchical representations through the training of neural networks with many hidden layers.” ISO/IEC 22989:2022, 3.4.8 and 3.4.4 In the literature: “Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction.” LeCun, Bengio and Hinton 2015 Norwegian definition: «Nevrale nettverk er en teknikk som brukes som byggesteiner innen maskinlæring og kunstig intelligens. De kan ses på som en veldig grov forenkling av hvordan nervecellene i hjernen fungerer.» Store norske leksikon

NVA

the Norwegian archive where research is registeredthe Norwegian National Research Archive (Nasjonalt vitenarkiv), where Norwegian research is registered — mishmash.no pulls publications, people, and project data from it every nightNasjonalt vitenarkiv, the national research archive run by Sikt, which replaced Cristin as the register of Norwegian research output and the basis for reporting to funders. mishmash.no syncs publications, people and project data from its open API nightly, and MishMash results should be registered under the centre’s NVA project

ORCID

a number that shows which researcher wrote whata free, unique ID for researchers that connects them to their publications across systems and employersOpen Researcher and Contributor ID, a persistent identifier maintained by a non-profit organisation that tells researchers with similar names apart across publishers, funders and institutions. The directory on mishmash.no uses ORCID records to keep people’s publication lists current

partner

an organisation that has joined MishMashan institution that has signed the consortium agreement or an associate-partner agreement with MishMash; people, by contrast, join as membersan institution, research-performing or not, bound to the centre by the original consortium agreement or a later associate-partner agreement. Partnership gives access to the network and its events and is meant as a door opener for the institution’s people, not as an end in itself

PhD fellow

someone who is paid to do research for a few years to earn the highest university degreea researcher employed, usually for three or four years, to complete a doctoral degree; MishMash’s partners have announced a large number of these positions, both scientific and artistica fixed-term research position leading to a PhD, requiring a completed master’s degree. Fellows are employed by a partner institution, not by MishMash, and follow either the scientific track or the artistic-research track; the directory lists them by work package Norwegian definition: «En stipendiatstilling skal føre til oppnådd doktorgrad og bidra til at den ansatte kvalifiserer seg for relevante karrierer ved høyere utdannings- og forskningsinstitusjoner og andre sektorer i arbeidslivet der forskningskompetanse kreves.» Universitets- og høyskoleforskriften § 3-19

postdoc

a researcher who has finished a PhD and is doing more research for a few yearsshort for postdoctoral fellow: a researcher with a completed PhD employed for a few years to carry out a research project, often on the way to a permanent positiona fixed-term qualifying position, typically two to four years, that requires the PhD to be submitted before applying and defended before appointment. MishMash’s postdocs are employed by the partner institutions and usually anchored in one work package Norwegian definition: «En postdoktorstilling har som formål at den ansatte skal opparbeide seg en forskerprofil og kompetanse som gjør vedkommende kvalifisert for å søke stilling som førsteamanuensis.» Universitets- og høyskoleforskriften § 3-18

prompt

what you tell an AI to do, usually by typingthe text, command or other input you give a generative AI system to steer what it does or makesthe input that conditions a generative model’s output: a question, a task, background, examples, or a combination. Designing prompts for better results is called prompt engineering. Teknologirådet’s Norwegian glossary, reviewed by Språkrådet, recommends instruks as the Norwegian term In standards: Input to a generative AI system that provides instructions on how to process the input. ISO/IEC 22989:2022/DAmd 1 (draft, 2025), 3.6.19 In the literature: In prompt-based learning the input is rewritten, using a template, into a textual prompt with unfilled slots, which the language model fills probabilistically; the answer is derived from the filled-in text. after Liu et al. 2023 Norwegian definition: «Tekst, kommando eller annen informasjon som gis til en KI-modell for å styre hva den skal gjøre eller generere.» Teknologirådet 2026

Research Council of Norway

the Norwegian government agency that pays for research, including MishMashNorges forskningsråd, the national agency that funds research and innovation; it funds MishMash as a research centre, and is often abbreviated RCNthe government agency under the Ministry of Education and Research that allocates public research funding through competitive calls. MishMash is its project number 357438; publications should acknowledge it with the wording on the acknowledgment page, and be registered under the centre’s project in NVA

responsible AI

making and using AI in ways that are fair, safe and good for peopledeveloping and using AI with attention to fairness, transparency, privacy, sustainability and accountability, not just capabilityan umbrella term for the practices and principles that keep AI development and use accountable: ethics guidelines, impact assessment, transparency, bias mitigation, environmental accounting and legal compliance such as the EU AI Act. MishMash’s REFLECT approach and WP5 take it as a central concern In the literature: Human responsibility for the development of intelligent systems along fundamental human principles and values, to ensure human flourishing and well-being in a sustainable world. after Dignum 2019

rule-based system

old-style AI that follows rules written by people, like a recipeearly AI built from hand-crafted rules — like the first computer systems for music composition and painting; also called symbolic AIsymbolic or knowledge-based AI, in which behaviour follows from explicitly encoded rules and representations rather than from learned parameters. It powered the first computer composition systems and Harold Cohen’s painting program AARON, and it is still combined with learning-based methods in hybrid systems where transparency and control matter In the literature: “A physical symbol system has the necessary and sufficient means for general intelligent action.” Newell and Simon 1976

Scientific Advisory Board

experts from other countries who check that MishMash’s research is gooda group of international experts who evaluate the centre’s research and advise on its direction; abbreviated SABtwelve international researchers and practitioners, from creative computing and music technology to law and communication, who bring an outside scientific perspective, evaluate quality and progress, and support dissemination to international audiences

seed funding

small pots of money to try out a new ideasmall grants the centre gives its members to start joint activities, test ideas or bring people together, announced in calls several times a yearinternal grants for collaborative activities among partners, awarded in numbered rounds. Applications are read by the work package leaders together with the management team, with a soft partiality check so that no one assesses a proposal they have a direct interest in

Stakeholder Board

people from outside research who tell MishMash what is usefula group of representatives from partner organisations, industry and cultural institutions that keeps the research relevant to practicethe centre’s link to users of its research: it advises on relevance and applicability, connects research with practice, and guides the centre towards practical innovation and uptake of results

stretchtext

text you can click to see more, right where you aretext the reader can stretch — unfolding more detail in place, an idea proposed by Ted Nelson in 1967 and used on this websitea hypertext form proposed by Ted Nelson in 1967 in which the text expands and contracts in place instead of jumping to another page. On mishmash.no it is combined with three reading levels, so a page can carry its own glossary at the depth the reader has chosen In the literature: Text that the reader expands and contracts in place, so detail arrives without leaving the page. after Nelson 1967

training data

the examples an AI learns fromthe collection of examples, such as texts, images or recordings, that a machine-learning model learns its patterns from; what is in it shapes what the model can dothe dataset a model’s parameters are fitted to. Its size, provenance, licensing and representativeness determine the model’s capabilities, its biases and its legal standing, which is why the data behind generative models is central to the copyright and rights questions MishMash studies In standards: “Data used to train a machine learning model.” ISO/IEC 22989:2022, 3.3.16 In policy: “Data used for training an AI system through fitting its learnable parameters.” EU AI Act 2024, Art. 3(29) In the literature: The set of examples on which a model’s error is measured and reduced during training, as distinct from the separately collected test set used to estimate how well it generalises. after Goodfellow, Bengio and Courville 2016, ch. 5 Norwegian definition: «Data som brukes til å trene et KI-system ved å tilpasse dets lærbare parametere.» KI-forordningen art. 3 nr. 29, uoffisiell oversettelse (DFD 2025)

work package

a team working on one part of a big projecta work package is research-speak for a team of people working on one part of a big projectthe unit of organisation in a research project: a defined set of tasks, people and results with its own leadership. MishMash has seven, WP1 to WP7, each built around one of the centre’s themes; the abbreviation WP is used throughout the site In the literature: The work at the lowest level of a work breakdown structure, for which cost and duration are estimated and managed. after PMBOK Guide, 6th ed.

work package leader

the person in charge of one of the seven teamsthe researcher responsible for one work package: its plan, its people and its results; abbreviated WPL, and helped by one or two “sidekicks”each of the seven work packages has a leader and one or two sidekicks who coordinate its scientific and artistic research, quality and dissemination. Together they form the Work Package Leader Group, which also assesses seed-funding applications with the management team

Missing a term? Suggest one via the website project.