{"id":3977,"date":"2025-10-30T11:21:50","date_gmt":"2025-10-30T10:21:50","guid":{"rendered":"https:\/\/wp.unil.ch\/iaunil\/a-four-dimensional-framework-of-humans-ai-competencies-an-integrated-approach\/"},"modified":"2026-04-20T12:44:18","modified_gmt":"2026-04-20T10:44:18","slug":"a-four-dimensional-framework-of-humans-ai-competencies-an-integrated-approach","status":"publish","type":"post","link":"https:\/\/wp.unil.ch\/iaunil\/en\/a-four-dimensional-framework-of-humans-ai-competencies-an-integrated-approach\/","title":{"rendered":"A four-dimensional framework of Humans-AI competencies: an integrated approach"},"content":{"rendered":"\n

Jean-Fran\u00e7ois (Jeff) Van de Po\u00ebl – Reflection project in progress, April 2025 status update<\/em><\/strong>
This work is licensed under CC BY-NC-SA 4.0<\/a><\/p>\n\n

Introduction<\/h2>\n\n

The rapid progress of artificial intelligence highlights a new potential form of distribution of cognitive tasks: humans and machines now act as partners with complementary competencies rather than as competing substitutes.<\/p>\n\n

Within this landscape, several research strands converge in describing cognition as shared between two distinct agents. Siemens, Dawson and Ga\u0161evi\u0107 speak of a \u00ab coordination space<\/em> \u00bb in which human and artificial intelligence fit together according to the nature of the problem encountered and the degree of uncertainty it involves. At the organisational level, Jarrahi defines this configuration as a \u00ab decisional symbiosis<\/em> \u00bb: AI processes large volumes of information quickly, while the human mobilises intuition, moral sense and contextual understanding to orient or correct the results.<\/p>\n\n

Understanding competence through the lens of this co-agency<\/em> (always led by the human!) leads to questioning how training programmes can prepare learners to orchestrate the dialogue with AI systems rather than to imitate them. In higher education, work on human-AI collaboration insists on the fact that effective learning requires the establishment of \u00ab joint regulation \u00bb mechanisms: AI provides immediate assistance, but the student remains responsible for framing, validation and integration of the knowledge produced. Competence is then constructed as a dynamic process in which each party, human and AI, brings specific cognitive resources.<\/p>\n\n

To make this process describable and actionable, the four-dimensional framework of hybrid human-AI competencies<\/strong> proposes to distinguish four domains:<\/p>\n\n

    \n
  1. AI’s specific capabilities,<\/li>\n\n\n\n
  2. essential skills and knowledge for critical verification,<\/li>\n\n\n\n
  3. AI literacy understood as the ability to interact with the tool,<\/li>\n\n\n\n
  4. the distinctive human capacities that remain beyond the reach of current algorithms.<\/li>\n<\/ol>\n\n

    Adopting these dimensions first amounts to recognizing that the automatic production of content does not exhaust the notion of competence. The machine can generate an argued text or produce a fluent translation, but it only assigns meaning to these productions according to an objective defined by the human.<\/p>\n\n

      \n
    • The first dimension, essential competencies<\/strong>, recalls that any production must be confronted with a disciplinary, methodological and normative bedrock ensuring its reliability. Without this validation step, the relationship becomes mere delegation without control. Without mastering this dimension, mobilising AI locks us into a substitutive use that has nothing to do with learning and into which we must absolutely not place ourselves.<\/li>\n\n\n\n
    • The second dimension, AI capabilities<\/strong>, serves as a benchmark to know what can be entrusted to an AI, or the tasks it is in a position to perform.<\/li>\n\n\n\n
    • The third dimension, AI literacy<\/strong>, corresponds to the professional gestures that enable one to configure and query the tool. Students learn to formulate requests, to interpret a confidence score, and to make explicit the limits of an algorithmic answer.<\/li>\n\n\n\n
    • Finally, the fourth dimension, distinctive human capacities<\/strong>, gathers abilities such as critical sense, conceptual creativity, empathy or moral judgement, indispensable to link the use of AI to social and ethical purposes.<\/li>\n<\/ul>\n\n
      \"the
      The four-dimensional framework of hybrid human-AI competencies<\/figcaption><\/figure>\n\n

      What is the point of structuring reflection according to these four entries?<\/strong><\/h2>\n\n

      Firstly, it enables responsibilities to be clarified: the machine handles the operations for which it is efficient, while the human retains the final decision along with the moral and social burden that accompanies it. Secondly, this breakdown can potentially support the design of training programmes by identifying, for each domain, pedagogical objectives and suitable assessment modalities. For example, learning AI literacy can translate into workshops on prompt<\/em> formulation or exercises in traceability documentation, whereas the development of distinctive human capacities will rather go through ethical debates or interdisciplinary projects. Thirdly, the framework opens up a space for empirical research: each dimension can be the object of specific measurements (quality of critical reasoning, robustness of validations, efficiency of meta-cognitive orchestration) and correlated with the overall performance of the human-AI collaboration.<\/p>\n\n

      Rethinking competencies in the light of this grid does not imply ranking human and machine, but situating them in an explicit relationship of interdependence. The notion of distributed cognition<\/em> recalls that knowledge results less from the internal capacities of a single agent than from the interactions between several actors, artefacts and contexts. In this framework, the role of education would no longer be to transmit a stock of knowledge that a model can reproduce; it would be to teach how to mobilise, combine and evaluate dispersed resources, whether they are human, social or algorithmic. The four dimensions of the framework offer a map for organising this mobilisation: determining what falls to automation, what requires expert validation, what depends on informed technical interaction, and what requires an ethical and creative stance.<\/p>\n\n

      By placing the human-AI relationship at the heart of the analysis, the framework finally calls for continuous vigilance: any evolution of models, norms or social expectations can redistribute the respective share of the dimensions. It is therefore up to training institutions to maintain active observation of these evolutions in order to adjust curricula and guarantee the lasting relevance of the competencies developed. Structured in this way, the approach makes it possible to reconcile the contributions of emerging technologies with the need to preserve the intellectual autonomy and social responsibility that remain, for the moment, at the heart of human activity.<\/p>\n\n

      Dimension 1: Essential conceptual competencies<\/h1>\n\n

      In learning contexts where generative AI systems now deliver drafts, translations or summaries in a few seconds, pedagogical utility shifts towards what students still have to produce without<\/strong> algorithmic recourse: a reasoned evaluation of the reliability, relevance and limits of the information obtained. The \u00ab essential competencies of critical verification \u00bb here designate the set of knowledge and dispositions that make it possible to examine, contextualise and, if needed, contest any claim, whether it comes from a traditional source or from a language model. This approach extends international AI literacy frameworks: EDUCAUSE emphasises that the capacity for \u00ab critical evaluation \u00bb conditions the responsible use of tools, while UNESCO lists \u00ab critical thinking \u00bb and \u00ab responsibility \u00bb among the fundamental competencies to acquire. In research in the cognitive sciences, Sperber and his colleagues describe this vigilance as a mechanism of \u00ab epistemic monitoring \u00bb that enables the calibration of the confidence granted to each piece of information.<\/p>\n\n

      These competencies are structured around three purposes. First, validating internal soundness<\/strong>: logical coherence, methodological accuracy, alignment with the theoretical frameworks of the discipline. Next, verifying provenance<\/strong>: identification of authors, analysis of biases, appreciation of the quality of underlying data. Finally, situating the statement in its socio-ethical and regulatory context<\/strong>: compliance with legal rules (e.g. AI Act for high-risk tools), as well as relevance with respect to the values and purposes of the academic community. Preparing students for these three operations amounts to preserving their intellectual autonomy; it is also a professional requirement, as verification expertise is becoming a marker of credibility in most sectors facing information flows amplified by AI.<\/p>\n\n

      List of essential competencies<\/strong><\/h4>\n\n
        \n
      • Disciplinary culture and methodology<\/strong>\n
          \n
        • Master the theoretical frameworks, methods of proof and writing standards of the discipline in order to situate any AI result within a robust scholarly reference framework.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
        • Source research and evaluation<\/strong>\n
            \n
          • Identify the origin of a document, examine the author’s authority, date and cross-check information to reduce dependence on a single source.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
          • Triangulation and corroboration<\/strong>\n
              \n
            • Confront several independent corpora, data sets or points of view before endorsing a conclusion; integrate discrepancies as elements of analysis.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
            • Argumentative and logical reasoning<\/strong>\n
                \n
              • Build an explicit chain of arguments; spot sophisms, contradictions or unjustified inferences.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
              • Epistemic vigilance<\/strong>\n
                  \n
                • Regulate the level of confidence granted to information according to reliability indicators and potential costs of error.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                • Ethical and regulatory frameworks<\/strong>\n
                    \n
                  • Apply the requirements relating to data protection, fairness and transparency; know the risk categories defined by the AI Act to situate responsibility issues.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                  • Metacognition and self-regulation<\/strong>\n
                      \n
                    • Plan control tasks, self-assess the soundness of one’s reasoning, adjust strategies when new information arises.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                    • Documentation and traceability<\/strong>\n
                        \n
                      • Record verification steps, archive versions of texts or data, provide a structured bibliography and accessible methodological notes.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                      • Critical and responsible communication<\/strong>\n
                          \n
                        • Present clearly the strengths, limits and uncertainties of a work; argue methodological decisions before an academic or professional audience.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n
                          \"the
                          Dimension 2 of the model related to the essential critical competencies<\/figcaption><\/figure>\n\n

                          Dimension 2: AI capabilities and task execution<\/h2>\n\n

                          Within the four-dimensional framework, the first dimension serves as a barometer to measure what an artificial intelligence system accomplishes autonomously or under minimal human supervision. Delimiting these capabilities is not merely descriptive; it is a prerequisite for deciding which portions of a curriculum should be reconfigured to favor analysis, interpretation and judgement rather than mere mechanical production. The initial list (content generation, information synthesis, pattern recognition, translation and predictive modelling) provides the reference baseline.<\/p>\n\n

                          Since 2024, several technical evolutions require amending this map without disrupting its spirit. The so-called \u00ab omnimodal \u00bb models process and generate text, image and audio within a single flow, reducing conversational latency to a few hundred milliseconds, as attested by the GPT-4o System Card.<\/em> At the same time, the extension of context windows (200 000 tokens for Claude 3) makes possible the synthesis of documentary sets previously out of reach. Finally, the diffusion of high-performing open-source models such as Llama 3 extends these functionalities to local infrastructures, favouring disciplinary experimentation in controlled environments.<\/p>\n\n

                          Preserving the original structure while enriching it therefore amounts to specifying, for each task, the new forms it may take. This dynamic update helps teachers to identify the activity segments that become relevant to delegate, those that require double human verification, and those that remain non-delegable due to ethical, contextual or creative constraints.<\/p>\n\n

                          List of main AI capabilities<\/p>\n\n

                            \n
                          • Content generation<\/strong>\n
                              \n
                            • Text, code, visual media and audio elements produced in a quasi-synchronous multimodal flow.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                            • Information synthesis and summarisation<\/strong>\n
                                \n
                              • Processing of extended corpora (\u2265 200k tokens) with argumentative or tabular formatting adapted to context.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                              • Pattern recognition and data analysis<\/strong>\n
                                  \n
                                • Detection of cross-modal patterns (image\u2013text, image\u2013audio) and extraction of statistical anomalies.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                • Translation and language processing<\/strong>\n
                                    \n
                                  • Multilingual conversion including prosody and tone, supporting low-resource languages.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                  • Simulation and predictive modelling<\/strong>\n
                                      \n
                                    • What-if scenarios orchestrated via external tool calls (Python, matrix computation) integrated in the dialogue.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                    • Omnichannel conversational interaction<\/strong>\n
                                        \n
                                      • Voice-text-image dialogue with latency under 500 ms, with adaptive tutoring capability.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                      • Action chaining via function calling<\/em><\/strong>\n
                                          \n
                                        • Orchestration of APIs (search, database, messaging) within a single generation session.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                        • Autonomous agency under supervision<\/strong>\n
                                            \n
                                          • Execution of planning-action-revision loops on defined tasks, with end control by the user.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                          • Knowledge extraction and structuring<\/strong>\n
                                              \n
                                            • Conversion of raw notes into queryable databases or concept graphs.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n
                                              \"the
                                              Dimension of the model’s AI capabilities<\/figcaption><\/figure>\n\n

                                              <\/p>\n\n

                                              Dimension 3: AI literacy<\/h1>\n\n

                                              AI literacy refers to the set of knowledge, skills and dispositions that enable one to interact in an informed, critical and responsible way with artificial intelligence systems. In higher education, the EDUCAUSE framework AI Literacy in Teaching & Learning<\/em> describes this competence as the capacity to (1) understand the basics of how AI works, (2) critically evaluate its uses, (3) apply the tool in authentic situations, and (4) guard against its misuses. UNESCO, for its part, distributes twelve student competencies into four similar groups, insisting on safety, responsibility and civic engagement.<\/p>\n\n

                                              In curricula, this dimension is distinguished from the \u00ab essential competencies \u00bb mentioned previously: the latter aim at validating content, whereas AI literacy concerns the capacity to dialogue with the tool, parameterise its responses and regulate its effects. The two registers however remain interdependent: without literacy, the student cannot mobilise disciplinary knowledge to formulate relevant queries; without a disciplinary bedrock, technical mastery of the model risks being limited to superficial manipulation. The categories below take up the four axes of your model and update them in light of the 2024-2025 literature, notably recent work on prompt engineering<\/em> in education and international analyses of the state of AI literacy.<\/p>\n\n

                                              Literacy categories<\/strong><\/h4>\n\n
                                                \n
                                              • Technical understanding<\/strong>\n
                                                  \n
                                                • Basic principles of models (training data, probabilistic inference, context windowing).<\/li>\n\n\n\n
                                                • Reading provider documents: system cards<\/em>, release notes, performance metrics.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                • Critical evaluation<\/strong>\n
                                                    \n
                                                  • Identification of potential biases, analysis of the relevance of sources internal or external to the model.<\/li>\n\n\n\n
                                                  • Assessment of the level of confidence to grant to outputs according to the intended use.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                  • Practical application<\/strong>\n
                                                      \n
                                                    • Elaboration of prompts<\/em> and query chains (RAG, function calling<\/em>) to solve a defined problem.<\/li>\n\n\n\n
                                                    • Integration of AI into disciplinary workflows while documenting the process.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                    • Ethical and sociotechnical considerations<\/strong>\n
                                                        \n
                                                      • Knowledge of regulatory obligations (AI Act, GDPR) and principles of fairness, transparency and sustainability.<\/li>\n\n\n\n
                                                      • Reflection on the social, environmental and professional impact of decisions made with AI.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n
                                                        \"the
                                                        Dimension 3 of the model, AI competencies<\/figcaption><\/figure>\n\n

                                                        Dimension 4 \u2014 The distinctive human capacities<\/h2>\n\n

                                                        The rise of language models and autonomous agents prompts a substantial shift of analytical work towards automated processes; however, it also highlights the persistent limits of computational intelligence when it comes to interpreting an open situation, managing social relationships or deciding under uncertainty. The literature devoted to human-AI collaboration uses the term co-agency<\/em> to describe the articulation of two cognitive systems with different properties: the machine excels at large-scale computation, the human retains superiority in contextual integration, normative evaluation and sense-making. Work from organisational psychology shows, for example, that the value of a hybrid result depends less on the raw capacity of the model than on the ability of human actors to orient the task, recognise the relevance of an algorithmic suggestion and decide to abandon or transform it when conditions evolve.<\/p>\n\n

                                                        The notion of distinctive human capacity<\/em> thus refers to competencies that AI does not perform autonomously: making sense of weak signals, conceiving novel interpretive frameworks, arbitrating between incompatible values, building trust within a collective. In research on sense-making<\/em>, Weick underlines that interpreting an ambiguous environment requires a continuous back-and-forth between lived experience, tacit knowledge and shared narratives. Now, the majority of generative models operate outside this embodied cycle; they recombine regularities drawn from past corpora without direct interaction with the immediate context. Similarly, conceptual creativity, that is, the capacity to establish unusual connections between distant fields or to modify the rules of the problem, remains difficult to formalise in machine learning, except in the form of statistically plausible variations.<\/p>\n\n

                                                        Moral responsibility offers another example of functional asymmetry. When clinical or financial recommendation systems suggest a choice, research in applied ethics recalls that final legitimacy rests with a human agent capable of assuming the consequences of the decision. This burden implies the evaluation of dimensions that are not entirely captured by the data: the vulnerability of the persons concerned, the fairness of the procedure or the compatibility with local legal norms. In professional environments, the coexistence of mixed teams, made up of people and algorithmic agents, reveals a growing need for relational competencies: empathy, negotiation, distributed leadership. These qualities support trust, which is essential to exploit the capabilities of AI efficiently without generating dependence or resistance.<\/p>\n\n

                                                        Finally, several studies emphasise the importance of metacognitive orchestration, defined as the capacity to plan, monitor and adjust the respective contribution of human and machine throughout a process. This competence includes recognising situations in which algorithmic assistance brings benefit (speed, exhaustiveness) as well as those in which it risks introducing biases or masking important contextual signals. It implies the construction of handover rules<\/em>: explicit moments at which decisional authority shifts from AI to the human operator, and vice versa.<\/p>\n\n

                                                        Placing these distinctive human capacities at the heart of training therefore responds to a twofold objective. On the one hand, it is a matter of ensuring that automation does not dilute intellectual autonomy nor social responsibility; on the other hand, it is necessary to equip learners with cognitive resources able to complement, rather than replicate, the strengths of artificial intelligence. The present dimension encapsulates these requirements by identifying the areas where pedagogical investment remains essential to maintain a balanced and reliable collaboration between humans and AI systems.<\/p>\n\n

                                                        List of distinctive human capacities<\/strong><\/h4>\n\n
                                                          \n
                                                        • Holistic sense-making<\/strong>\n
                                                            \n
                                                          • Detect and link weak or contradictory signals to build situated interpretations.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                          • Conceptual creativity<\/strong>\n
                                                              \n
                                                            • Develop frameworks, metaphors or novel hypotheses in order to orient or go beyond AI’s proposals.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                            • Moral judgement and responsibility<\/strong>\n
                                                                \n
                                                              • Appreciate the ethical, legal and social implications of an AI-assisted decision and assume the associated responsibility.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                              • Empathy and relational leadership<\/strong>\n
                                                                  \n
                                                                • Maintain trust, motivation and alignment of stakeholders in teams in which algorithmic agents also operate.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                                • Interdisciplinary synthesis<\/strong>\n
                                                                    \n
                                                                  • Combine knowledge from heterogeneous domains to create solutions or analyses that AI would not have inferred.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                                  • Metacognitive orchestration<\/strong>\n
                                                                      \n
                                                                    • Plan, monitor and adjust the scope of AI’s intervention throughout a work process.<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                                    • Embodied and contextual knowledge<\/strong>\n
                                                                        \n
                                                                      • Mobilize field experience, local norms or material constraints to validate or correct algorithmic recommendations.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n
                                                                        \"the
                                                                        Dimension 4, the distinctive human capacities<\/figcaption><\/figure>\n\n

                                                                        <\/p>\n\n

                                                                        Conclusion<\/h1>\n\n

                                                                        The four-dimensional mapping presented \u2014 AI capabilities, essential critical verification competencies, AI literacy and distinctive human capacities \u2014 proposes an integrated framework to analyze and develop competence in an environment where cognition and action are distributed between human agents and algorithmic systems. Each dimension fulfills a specific function:<\/p>\n\n

                                                                          \n
                                                                        • AI capabilities<\/strong> describe an automation perimeter in constant evolution. Characterising them makes it possible to determine what can be delegated safely and efficiently, and to redirect pedagogical effort towards the formulation of objectives, supervision and interpretation.<\/li>\n\n\n\n
                                                                        • Essential competencies<\/strong> provide the disciplinary, methodological and normative bedrock that makes reasoned auditing of generative productions possible; they guarantee scientific reliability and regulatory compliance.<\/li>\n\n\n\n
                                                                        • AI literacy<\/strong> constitutes the operational link: it equips the learner with the technical gestures and knowledge needed to configure the tool, evaluate its outputs and regulate its uses in authentic situations.<\/li>\n\n\n\n
                                                                        • Distinctive human capacities<\/strong> recall the irreducible share of responsibility, conceptual creativity and empathy; they ensure the social and ethical anchoring of decisions made with the support of AI.<\/li>\n\n\n\n
                                                                        • <\/li>\n<\/ul>\n\n

                                                                          By articulating these four components, the framework sheds light on two major challenges. On the one hand, it enables explicit distribution of responsibilities<\/strong>: AI ensures the speed and scale of information processing, while the human retains the final decision, moral judgement and contextualisation. On the other hand, it offers an operational basis for curriculum design<\/strong>: each dimension can be the object of learning objectives, dedicated activities and suitable assessments, thus facilitating the progressive integration of AI into disciplines without sacrificing academic requirements.<\/p>\n\n

                                                                          The framework also opens up several avenues of research and follow-up:<\/p>\n\n

                                                                            \n
                                                                          1. Measuring the efficiency of the orchestration<\/strong>: developing indicators to link the quality of human-AI collaboration to overall performance and the reliability of productions.<\/li>\n\n\n\n
                                                                          2. Observing the evolution of the boundaries<\/strong>: documenting the regular shifts of the automation perimeter in order to continuously adjust the essential competencies and the literacy required.<\/li>\n\n\n\n
                                                                          3. Evaluating the ethical and societal impact<\/strong>: analysing how the decisions made in a hybrid team affect equity, transparency and responsibility within organisations and communities.<\/li>\n\n\n\n
                                                                          4. <\/li>\n<\/ol>\n\n

                                                                            Finally, the four-dimensional approach recalls that training does not aim to replace human expertise with computational power, but rather to cultivate the capacity to coordinate<\/strong> these two resources in a reflexive, responsible and sustainable manner. By providing learners with solid conceptual benchmarks, mastered interaction techniques and a keen sense of responsibility, institutions can support the integration of artificial intelligence while protecting intellectual autonomy and the social purpose of education.<\/p>\n\n

                                                                            References<\/h2>\n\n

                                                                            Anthropic. (2024). Claude 3 model family : Model card<\/em> (v 1.0). Anthropic.<\/p>\n\n

                                                                            Bano, M., Pan, Z., & Papadimitriou, A. (2025). AI literacy and trust: A multi-method study of human\u2013GAI team collaboration. International Journal of Human-Computer Studies, 199<\/em>, 102086.<\/p>\n\n

                                                                            Dawson, S., Joksimovic, S., Mills, C., Ga\u0161evi\u0107, D., & Siemens, G. (2023). Advancing theory in the age of artificial intelligence. British Journal of Educational Technology, 54<\/em>(5), 1980-1993.<\/p>\n\n

                                                                            Dellermann, D., Ebel, P., S\u00f6llner, M., Bitzer, M., & Leimeister, J. M. (2019). Hybrid intelligence. Business & Information Systems Engineering, 61<\/em>(5), 637-643.<\/p>\n\n

                                                                            EDUCAUSE. (2024). AI literacy in teaching and learning: Executive summary<\/em>. EDUCAUSE.<\/p>\n\n

                                                                            European Union. (2024). Regulation (EU) 2024\/1689 of the European Parliament and of the Council of 11 January 2024 laying down harmonised rules on artificial intelligence (AI Act)<\/em>. Official Journal of the European Union.<\/p>\n\n

                                                                            Hutchins, E. (1995). Cognition in the wild<\/em>. MIT Press.<\/p>\n\n

                                                                            International Organization for Standardization. (2024). ISO\/IEC 42001:2024 \u2013 Artificial intelligence management system requirements<\/em>. ISO.<\/p>\n\n

                                                                            Jarrahi, M. H. (2018). Artificial intelligence and the future of work: Human\u2013AI symbiosis in organizational decision making. Business Horizons, 61<\/em>(4), 577-586.<\/p>\n\n

                                                                            Lee, D., & Palmer, E. (2025). Prompt engineering in higher education: A systematic review to inform curricula. International Journal of Educational Technology in Higher Education, 22<\/em>(1), 7.<\/p>\n\n

                                                                            Li, Y., Du, Y., Zhou, K., Huang, J., & Shi, Y. (2023). Evaluating object hallucination in large vision\u2013language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing<\/em> (pp. 250-263).<\/p>\n\n

                                                                            Lyytinen, K., Hansen, S., Henfridsson, O., & Yoo, Y. (2024). Design principles for artificial-intelligence-augmented decision making. Journal of Strategic Information Systems, 33<\/em>(1), 101743.<\/p>\n\n

                                                                            Markauskaite, L., & Goodyear, P. (2022). What capabilities do learners need for a world with AI ? Computers & Education: Artificial Intelligence, 3<\/em>, 100056.<\/p>\n\n

                                                                            Mercier, H., & Sperber, D. (2011). Why do humans reason ? Arguments for an argumentative theory. Behavioral and Brain Sciences, 34<\/em>(2), 57-74.<\/p>\n\n

                                                                            OpenAI. (2024). GPT-4o system card<\/em> (v 1.0). OpenAI.<\/p>\n\n

                                                                            Pan, Z., Moore, O. A., & Papadimitriou, A. (2025). Contextual factors influencing AI literacy and trust. Computers & Education, 210<\/em>, 104806.<\/p>\n\n

                                                                            Puranam, P. (2021). Human\u2013AI collaborative decision making as an organization-design problem. Journal of Organization Design, 10<\/em>(1), 17-27.<\/p>\n\n

                                                                            Shrestha, Y. R., Benner, M., & von Krogh, G. (2023). Human\u2013AI collaborative decision making as an organization-design problem. MIS Quarterly Executive, 22<\/em>(4), 275-289.<\/p>\n\n

                                                                            Siemens, G., Dawson, S., & Ga\u0161evi\u0107, D. (2022). Human and artificial cognition: Rethinking learning at the intersection. Computers & Education: Artificial Intelligence, 3<\/em>, 100107.<\/p>\n\n

                                                                            Topol, E. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25<\/em>, 44-56.<\/p>\n\n

                                                                            UNESCO. (2025). AI competency framework for students<\/em> (2nd ed.). UNESCO Publishing.<\/p>\n\n

                                                                            Weick, K. E. (1995). Sensemaking in organizations<\/em>. Sage.<\/p>\n\n

                                                                            <\/p>\n","protected":false},"excerpt":{"rendered":"

                                                                            Human-AI co-agency and joint regulation. Four axes for training: AI capabilities, critical verification skills, AI literacy, and the distinctive human capacities that remain essential.<\/p>\n","protected":false},"author":1002797,"featured_media":3183,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_seopress_robots_primary_cat":"","_seopress_titles_title":"","_seopress_titles_desc":"","_seopress_robots_index":"","footnotes":""},"categories":[23],"tags":[],"class_list":{"0":"post-3977","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-education-en"},"_links":{"self":[{"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/posts\/3977","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/users\/1002797"}],"replies":[{"embeddable":true,"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/comments?post=3977"}],"version-history":[{"count":1,"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/posts\/3977\/revisions"}],"predecessor-version":[{"id":3978,"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/posts\/3977\/revisions\/3978"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/media\/3183"}],"wp:attachment":[{"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/media?parent=3977"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/categories?post=3977"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wp.unil.ch\/iaunil\/en\/wp-json\/wp\/v2\/tags?post=3977"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}