K-12 AI Curricula Mapping PDF

Summary

This document from UNESCO maps government-endorsed AI curricula for K-12 education. It explores the development, integration, and content of these curricula, offering insights into how AI is being incorporated into primary and secondary education worldwide.

Full Transcript

K-12 AI curricula A mapping of government-endorsed AI curricula UNESCO Education Sector The Global Education 2030 Agenda Education is UNESCO’s top priority because it is UNESCO, as the United Nations’ specialized a b...

K-12 AI curricula A mapping of government-endorsed AI curricula UNESCO Education Sector The Global Education 2030 Agenda Education is UNESCO’s top priority because it is UNESCO, as the United Nations’ specialized a basic human right and the foundation on which agency for education, is entrusted to lead and to build peace and drive sustainable development. coordinate the Education 2030 Agenda, which is UNESCO is the United Nations’ specialized agency part of a global movement to eradicate poverty for education and the Education Sector provides through 17 Sustainable Development Goals by global and regional leadership in education, 2030. Education, essential to achieve all of these strengthens national education systems and goals, has its own dedicated Goal 4, which aims to responds to contemporary global challenges “ensure inclusive and equitable quality education through education with a special focus on and promote lifelong learning opportunities for all.” gender equality and Africa. The Education 2030 Framework for Action provides guidance for the implementation of this ambitious goal and commitments. Published in 2022 by the United Nations Educational, Scientific and Cultural Organization, 7, place de Fontenoy, 75352 Paris 07 SP, France © UNESCO 2022 This document is available in Open Access under the Attribution-ShareAlike 3.0 IGO (CC-BY-SA 3.0 IGO) licence (http://creativecommons.org/ licenses/by-sa/3.0/igo). By using the content of this document, the users accept to be bound by the terms of use of the UNESCO Open Access Repository (http://www.unesco.org/open-access/terms-use-ccbysa-en). The designations employed and the presentation of material throughout this document do not imply the expression of any opinion whatsoever on the part of UNESCO concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. The ideas and opinions expressed in this document are those of the authors; they are not necessarily those of UNESCO and do not commit the Organization. Cover design: Marie Moncet Cover credit: Ryzhi/Ryzhi/Shutterstock.com Inside icon (pp. 51-53): Marie Moncet Coordinator: Fengchun Miao ED-2022/FLI-ICT/K-12 CLD 1223_23 Printed by UNESCO Printed in France K-12 AI curricula — A mapping of government-endorsed AI curricula K-12 AI curricula A mapping of government-endorsed AI curricula 1 K-12 AI curricula — A mapping of government-endorsed AI curricula Acknowledgements The report has been produced by UNESCO’s Unit for Technology and Artificial Intelligence in Education, which sits within the Futures of Learning and Innovation Team. Fengchun Miao, Chief of this Unit, conceptualized and executed the methodology for the data collection, designed and managed the surveys, and led the authoring of the report. Kelly Shiohira of JET Education Services supported the data collection, analysed the survey data, carried out the curriculum mapping, and drafted the report. Appreciation is due especially to Juan David Plaza Osses and Iaroslava Kharkova, members of this Unit who organized the administration of the surveys and interviews with focal experts nominated by Member States, and to fellow colleagues Glen Hertelendy and Samuel Grimonprez for coordinating the production of the report. UNESCO acknowledges with gratitude the following governmental representatives for their contributions and time during the interviews to provide more detailed information about the AI curricula of their respective countries: Noha Alomari, ICT Education Specialist from the Department of Curriculum and Learning Resources at the Qatar Ministry of Education and Higher Education, Peter Bauer, Head of Department of Informatics and Media Technology at HTBLA Leonding in Austria, Marie-Thérèse Delhoune, Inspector of Secondary Education from the General Inspection Service at the Fédération Wallonie-Bruxelles in Belgium, Helder Pais, Head of the Curriculum Development Department at the Directorate-General for Education from the Ministry of Education of Portugal, and Zhang Xiong, Professor from the School of Computer Science and Engineering at Beihang University in China. The report has also benefitted from information collected from interviews with the following focal persons: Shalini Kapoor, Bettina Culter, Anne Forbes Joyeeta Das and Lucy Qu from IBM, Anshul Sonak and Shweta Khurana from Intel, Ki-Sang Song from the Korea National University of Education in the Republic of Korea, Alexa Joyce and Simran Jha from Microsoft, Irene Lee and Cynthia Breazeal from MIT; Muna Al Ansari from Kuwait; Laila Mohammend Al Atawy from Jordan, Mohammed Jumah F. Al-Enazi from Saudi Arabia; Stefan Badza from Serbia, Kyungsuk Chang from the Republic of Korea, Saffin Mathew from India, Marília Neres from Portugal, Ashutosh Raina from India, Ralitsa Voynova from the Republic of Bulgaria, Isabelle Sieh from Germany, Paula Thompson from Canada, Artashes Torosyan from Armenia, Ralitsa Voynova from the Republic of Bulgaria, and Stephan Waba from Austria. Thanks are given to Patrick Molokwane of JET Education Services for desktop research support. Gratitude is also extended to Jenny Webster for copy-editing and proofreading the text, and to Marie Moncet for designing the layout. Finally, UNESCO would like to thank the TAL Education Group for providing financial support to launch the project on AI and the Futures of Learning, through which this report was also made possible. 2 K-12 AI curricula — A mapping of government-endorsed AI curricula Table of Contents Acknowledgements............................................................................................................................ 2 Objective and scope of the report................................................................................................... 6 Scope of the mapping.................................................................................................................................................... 6 Introduction........................................................................................................................................ 7 A primer on AI terms and technologies................................................................................................................... 8 Artificial intelligence.................................................................................................................................................................9 AI techniques...............................................................................................................................................................................9 AI technologies........................................................................................................................................................................10 Ethical AI.....................................................................................................................................................................................10 AI literacy...................................................................................................................................................................................11 Pedagogical concepts and terminologies.............................................................................................................11 Existing frameworks of reference on AI curricula...............................................................................................12 AI Literacy: Competencies and Design Considerations............................................................................................13 AI4K12: Five Big Ideas and K–12 AI Curriculum Guidelines.....................................................................................14 The Machine Learning Education Framework..............................................................................................................16 Methodology..................................................................................................................................... 18 Data collection................................................................................................................................................................18 Criteria for selecting government‑endorsed AI curricula................................................................................18 List of government-endorsed AI curricula...........................................................................................................19 Limitations to the survey analysis............................................................................................................................20 Key findings of the analysis of government-endorsed AI curricula....................................... 21 Curriculum development and endorsement.......................................................................................................21 AI curriculum development and endorsement mechanisms.................................................................................21 Vision and motivations for developing AI curricula...................................................................................................22 Pilot testing and evaluation of AI curricula..................................................................................................................22 Example: Qatar curriculum development foundations and principles..............................................................23 Curriculum integration and management............................................................................................................25 Allocation of curriculum hours..........................................................................................................................................26 Essential conditions for supporting AI curricula..........................................................................................................27 Example: The introduction of AI by the CBSE in India...............................................................................................28 AI curriculum content...................................................................................................................................................30 Main categories of AI curriculum content.....................................................................................................................30 Time allocations for AI curriculum categories..............................................................................................................30 Coverage of AI curriculum categories.............................................................................................................................31 Example: AI curriculum content in Austria...................................................................................................................36 Learning outcomes of AI curricula...........................................................................................................................38 Methodology for analysing learning outcomes..........................................................................................................38 Framework for the categorization of learning outcomes........................................................................................38 Mapping of learning outcomes by AI categories........................................................................................................39 Example: Progression of AI learning outcomes in the Republic of Korea..........................................................45 3 K-12 AI curricula — A mapping of government-endorsed AI curricula Curriculum implementation.......................................................................................................................................46 Teacher training and support.............................................................................................................................................46 Learning tools and environments..................................................................................................................................... 46 Suggested pedagogies.........................................................................................................................................................48 Example: Implementation of the Information Science and Technology Curriculum for Senior High Schools, China.........................................................................................................................................................................49 Key findings and recommendations............................................................................................. 51 Curriculum development and endorsement.......................................................................................................51 Curriculum integration and management............................................................................................................52 Curriculum content and learning outcomes........................................................................................................52 Curriculum implementation......................................................................................................................................53 Concluding comment....................................................................................................................... 54 References.......................................................................................................................................... 55 Appendix............................................................................................................................................. 58 Survey sent to representatives of Member States..............................................................................................58 UNESCO mapping of government-approved AI curricula......................................................................................58 General Information...............................................................................................................................................................58 AI curriculum 1.........................................................................................................................................................................58 List of figures Figure 1. Number of AI curricula by integration type...................................................................................25 Figure 2. Time allocation per year of AI curricula...........................................................................................26 Figure 3. Per cent of curricula engaging each grade level..........................................................................27 Figure 4. Support for implementation undertaken.......................................................................................28 Figure 5. Thematic approach to the interdisciplinary integration of AI into the curriculum..........29 Figure 6. AI implementation actors and procedures.....................................................................................29 Figure 7. Boxplot of focus areas by per cent of curriculum hours............................................................31 Figure 8. Allocation of curriculum time by topic area...................................................................................32 Figure 9. Percentage allocations for AI foundations......................................................................................33 Figure 10. Percentage allocations for ethics and social impact...................................................................34 Figure 11. Percentage allocations for understanding, using and developing AI..................................35 Figure 12. Percentage allocations by topic area................................................................................................37 Figure 13. Curriculum Standards, Republic of Korea.......................................................................................45 Figure 14. Average pedagogical engagement profile.....................................................................................49 4 K-12 AI curricula — A mapping of government-endorsed AI curricula List of tables Table 1. AI Literacy Competency Framework.................................................................................................13 Table 2. ‘Big Idea 1: Perception’ concepts and learning outcomes........................................................15 Table 3. The Machine Learning Education Framework, with learning outcomes and definitions..........................................................................................................................................17 Table 4. K–12 AI curricula, endorsed and implemented by governments..........................................19 Table 5. Governmental K–12 AI curricula in development........................................................................20 Table 6. Non-governmental AI curricula included in the study as benchmarks...............................20 Table 7. Essential conditions for supporting AI curricula...........................................................................27 Table 8. AI curriculum areas..................................................................................................................................30 Table 9. Curriculum engagement by topic area............................................................................................31 Table 10. Curriculum engagement for the AI foundations category by topic area............................33 Table 11. Curriculum engagement for the category ethics and social impact by topic area..........34 Table 12. Curriculum engagement for the category of understanding, using and developing AI, by topic area........................................................................................................................................36 Table 13. Knowledge outcome mapping...........................................................................................................39 Table 14. Skills outcome mapping........................................................................................................................42 Table 15. Values and attitudes outcome mapping.........................................................................................44 Table 16. Suggested pedagogical approaches and specifications...........................................................48 5 K-12 AI curricula — A mapping of government-endorsed AI curricula Objective and scope of the report As AI technology represents a new subject area for Scope of the mapping K–12 schools worldwide, there is a lack of historical UNESCO is investigating the current practices knowledge for governments, schools and teachers to of developing and implementing AI curricula in draw from in defining AI competencies and designing primary and secondary school education from a AI curricula. This mapping exercise analyses existing global perspective. ‘AI curricula’ in this study refers to AI curricula with a specific focus on the curriculum structured programmes of learning on AI-related topics content and learning outcomes, and delineates that: 1) are endorsed by either national or regional development and validation mechanisms, curriculum governments; and 2) target learners in general school alignment, the preparation of learning tools and education from kindergarten to grade 12. This study required environments, the suggested pedagogies, does not cover AI curricula designed for specialized and the training of teachers. Key considerations are technical and vocational education and training (TVET) drawn from the analysis to guide the future planning institutions, higher education institutions, or informal of enabling policies, the design of national curricula or learning opportunities. institutional study programmes, and implementation strategies for AI competency development. 6 K-12 AI curricula — A mapping of government-endorsed AI curricula Introduction A diverse range of AI technologies are currently in use the inclusive and equitable use of AI in education; internationally, and there is a growing recognition leveraging AI to enhance education and learning; of the importance of AI in the context of labour and fostering skills development for jobs and life with AI; in terms of its impact on everyday life. There is ‘wide and safeguarding education data so that its use is consensus’ that AI will ‘affect occupations at all levels ethical, transparent and auditable (UNESCO, 2019a). of pay and education’ (Royal Society UK, 2018, cited However, currently relatively few initiatives focus on AI in the Microsoft Computer Science Framework, 2021). in K–12 contexts, leading to a recent recommendation A 2018 analysis by McKinsey concluded that by 2030, that policy-makers should ‘provide an enabling policy 70 per cent of global firms are expected to adopt at environment and curricular spaces for exploring AI’ least one type of AI technology. However, AI adoption (Miao et al., 2021, p. 34). will widen existing gaps between countries (Bughin et al., 2018a). Currently, in the United States, machines As a leading part of the international community and perform as many as 30 per cent of workforce tasks conversation on technology in education, UNESCO has (Kelly, 2020). Additionally, increasing mismatches led a number of important developments in the AI in/ between skills being taught in schools and TVET for Education space. institutions and skills needed by the job market In 2015, the Qingdao Declaration (UNESCO, 2015) are anticipated in correlation with higher rates of included a point on exploring the potential of big data automation and AI integration (Bughin et al., 2018b). to enhance online learning, inform an understanding The COVID-19 pandemic has only increased the pace of student behaviour, and improve the design and of automation, which may result in as many as 1 in delivery of online courses. The declaration urged that 16 workers1 requiring retraining by 2030 and a further ‘governments must develop policies and systems to decline in the availability of middle- and low-skill jobs ensure secure, appropriate and ethical use of data, (Lund et al., 2021). including safeguarding the privacy and confidentiality The impact of AI technology is not limited to the of students’ personally identifiable information’. workforce. AI has profound implications for culture, y The Beijing Consensus on Artificial Intelligence and diversity, education, scientific knowledge, and Education (UNESCO, 2019b) includes a series of communication and information, especially insofar as recommendations and considerations for AI in they concern peace, sustainability, gender equality, and Education. Demonstrating a strong focus on equity the specific challenges of Africa (COMEST, 2019). These and inclusion, one of the recommendations in are all areas of significant interest to both international the consensus is to ensure that AI promotes high- and national bodies that focus on development and quality education and learning opportunities for all, policy. Citizens are increasing their interactions with AI, irrespective of gender, disability, social or economic knowingly or unknowingly. AI has been deployed to status, ethnic or cultural background, or geographic drive cars, automate customer service, identify targets location. for military bombs, screen applicants at national ports y As part of the UNESCO Strategy on Technological of entry, direct policing efforts, determine grades, select Innovation and Education (2022-2025), in addition university entrants and scholarship recipients, and to an observatory and capacity building, the make decisions about personal finance (Engler, 2021; Organization seeks to develop standard-setting Frantzman and Atherton, 2019; Shiohira, 2021). instruments and normative tools, including guidelines and frameworks, ‘to strengthen the digital International policy guidance suggests that competencies (understanding, skills, and values) common areas should be pursued through different of teachers and learners and ensure a human- contextualized approaches such as promoting rights-based, safe, ethical, and meaningful use of 1 Eight countries were included in this analysis, namely China, France, Germany, India, Japan, Spain, the United Kingdom, and the United States, accounting for almost half the global population and 62 per cent of GDP. 7 K-12 AI curricula — A mapping of government-endorsed AI curricula “ technologies in a lifelong learning perspective’ The world’s citizens need to understand what (UNESCO, 2021a). Transversal areas of action are the the impact of AI might be, what AI can do and expansion of access to education, particularly for what it cannot do, when AI is useful and when its use marginalized groups and individuals, and the quality should be questioned, and how AI might be steered for of teaching and learning. the public good (Miao and Holmes, 2021, p. 6). y UNESCO published AI and Education: Guidance for policy-makers in April 2021 with an aim to foster AI- The Forum emphasized the centrality of human- readiness among policy-makers (Miao et al., 2021). oriented competencies, such as an understanding This report provides an orientation for its target of the ethics of AI and its social impacts, and readers on AI, including opportunities, risks, key technology-oriented competencies, such as the skills definitions, trends in AI, implications for teaching and and knowledge to use, interpret and develop AI. learning, and how education can prepare students Subject-specific and interdisciplinary approaches to AI for the AI era. It concludes with recommendations for in education were recommended, including building local policy planning. on existing ICT curricula and integrating analyses of y In October 2021, UNESCO launched AI and the opportunities and impacts of AI into humanities, the Futures of Learning,2 a project with three science and art courses (Miao and Holmes, 2021). independent but complementary strands: (1) a report proposing recommendations on AI-enabled futures This report contributes further to the understanding of learning; (2) guidance on ethical principles for the of AI in K–12 education, in particular the ways in which use of AI in education; and (3) a guiding framework students are currently being prepared for life and work on AI competencies for school students. in the AI era, by providing an analysis of the global The everyday realities of the current uses of AI and its landscape of government-endorsed AI curricula for impact on the world of work spur a sense of urgency to grade school education and their design, content and create international consensus on its acceptable roles implementation. This report is intended to inform the in society, the expected humanistic considerations creation of supportive tools and frameworks, with a view in its development and deployment, and how to to enabling the development of a guiding framework on equip children with the competences they will need AI competencies. It also forms one part of the work laid to successfully navigate the existing – not the future, out in the UNESCO Strategy on Technological Innovation in but the existing – world. The Beijing Consensus on Education (2022-2025) (UNESCO, 2021a). Artificial Intelligence and Education (UNESCO, 2019b) calls on all Member States to ‘be cognizant of the A primer on AI terms and technologies emergence of a set of AI literacy skills required for This report engages a range of concepts and terms effective human–machine collaboration, without from both AI-specialist and education-specialist fields. losing sight of the need for foundational skills such Despite the ubiquitous presence of AI in fields such as as literacy and numeracy’. The Consensus endorses a marketing, finance, and increasingly education, some ‘humanistic approach’ to ‘preparing all people with decision-makers and practitioners may be unfamiliar the appropriate values and skills needed for effective with some of the terms used in this analysis. Similarly, it human–machine collaboration in life, learning and is not guaranteed that all AI practitioners and decision- work, and for sustainable development’. To support the makers will be aware of prominent trends in pedagogy implementation of the Beijing Consensus, on 7 and referenced in the curricula. Therefore, this section provides 8 December 2020 UNESCO hosted the International a brief primer on some of the technologies, terms and Forum on AI and the Futures of Education: Developing pedagogies discussed in this text, to equip readers with Competencies for the AI Era. Participants at this event a general understanding of each main concept. First, five considered the competencies that citizens require: terms from the field of AI are explained in turn, and then the section on pedagogical concepts looks at several concepts including ‘competence-based evaluation’, ‘constructivism’, ‘constructionism’, and ‘design thinking’. 2 See https://events.unesco.org/event?id=2883602288 8 K-12 AI curricula — A mapping of government-endorsed AI curricula Nearly all of these concepts and terms have generated example, a classifier is an algorithm that is designed at least some amount of academic debate, and have to sort things into categories (e.g. ‘spam’ or ‘not both proponents and detractors, but the purpose of this spam’) using labelled data. Decision trees are a type report is not to delve deeply into conflicting viewpoints. of classification algorithm in which a series of ‘nodes’ This should not be taken as an exhaustive exploration. (decision points, represented as questions) lead to ‘branches’, where the results of different response Artificial intelligence options are separated. For example, in the MIT DAILy Curriculum, which is discussed at length later in this The term ‘artificial intelligence’ was coined in 1956 report, students create a decision tree to classify when Marvin Minsky and John McCarthy hosted the different types of pasta. One node might ask, ‘Is it Dartmouth Summer Research Project on Artificial longer than four inches?’, with spaghetti, linguine, and Intelligence (COMEST, 2019; Haenlein and Kaplan, other long pastas on one branch to the next node 2019). AI has gained popularity owing to the rise of big and macaroni, farfalle, and other short pastas on data and the exponential growth of computing power another branch. (Haenlein and Kaplan, 2019). The definition of AI has y In unsupervised learning, machine learning expanded and evolved over time (Miao et al., 2021), generates outputs based on clustering similarities in and now refers to machines that imitate some features groups of unknown and unlabelled data. of human intelligence, such as perception, learning, y Reinforcement learning is a type of ongoing ML reasoning, problem-solving, language interaction and which is trained to maximize a reward (for example, creative work (COMEST, 2019). to return the maximum amount of currency on an investment). The analysis in this report divides AI into two categories, y Neural networks are ML algorithms that are modelled ‘AI techniques’ and ‘AI technologies’. The former on animal brains. They are comprised of input layers, encompasses the methods used to build different types hidden layers and output layers. In the hidden layers, of AI, while the latter refers to the fields of study and data is processed in nodes based on its value and an products which are created by those techniques. assigned weight, and only data that passes a given threshold is allowed through. Filtered data makes its AI techniques way through one or more hidden layers to the output The AI techniques included in the curricula athat are layer. ‘Learning’ in neural networks occurs through analysed in this report are briefly described below:3 ‘back propagation’, an algorithm which seeks to minimize error by adjusting the weights in the hidden y Classical AI is rule-based and uses conditional if-then layer(s) of different nodes based on the correctness statements to generate outputs. Rule-based reasoning and influence of each node’s inputs. can be used in technologies such as chatbots (e.g. ‘If y Deep learning (DL) refers to neural networks with the input contains the words “what”, “price” and “?”, then multiple hidden layers. While ML in general relies on return the listed product price amount’). data that is structured (e.g. selected, labelled and y Machine learning (ML) refers to any type of organized into tables), DL can process unstructured computer program that can ‘learn’ without explicit data such as text and images. Neural networks and/ programming by accessing and processing large or deep learning are used in image and speech amounts of data. What is meant by ‘learn’ is that the recognition. program can produce new outputs without being y General adversarial networks (GANs) are a type explicitly ‘told’ what those outputs should be, as of machine learning which is designed to generate would be the case in classical AI. The remainder of new content, for example images.4 A GAN includes this list is comprised of some of the many different two deep neural networks. One of these generates sub‑categories of ML. content and the other evaluates it. GANs do not work y Supervised learning is a type of ML which is trained particularly well with text – yet. on known, labelled data to produce outputs. For 3 The explanations given here are derived from Miao et al. (2021), supplemented by examples and definitions from the curricula included in this report, in particular the MIT DAILy Curriculum, the AI4K12 Curriculum Framework, and the IBM Youth Challenge. 4 For example, GAN technology can be used to generate images of people that do not exist (see https://www.thispersondoesnotexist.com) 9 K-12 AI curricula — A mapping of government-endorsed AI curricula AI technologies in terms of explainability and transparency. Other challenges include balancing the use of personal data The AI technologies included in the curricula that are with the individual right to privacy; the security of analysed in this report are briefly described below: data and potential exposure to cyber-crime; and the y Chatbots are computer programs designed to reinforcement of prior beliefs by AI algorithms based simulate oral and/or written conversation.5 on user interest, which can limit people’s exposure y Computer vision is the field of AI that is concerned to ideas and information and, some argue, infringe with deriving and using information gathered on an individual’s right to freedom of expression from images and visual inputs. Computer vision (UNDESA et al., 2021). drives products such as automated highlight reels, self-driving cars, and quality-control tools (for the The First Draft of the Recommendation on the Ethics identification of defects) in manufacturing.6 of Artificial Intelligence (UNESCO, 2020) highlights y Natural Language Processing (NLP) is based on some of the key ethical challenges of AI, noting combining computer science with computational impacts on decision-making, employment and labour, linguistics, an interdisciplinary field for studying social interaction, health care, education, media, human language, in order to create rule-based freedom of expression, access to information, privacy, models of human speech or text that can be used democracy, discrimination, and weaponization. by computers. This enables computers to process The Recommendation proposes that AI should be and appropriately respond to human language. This monitored by third parties to ensure it is trustworthy technology drives computer translation from one and works for the good of humanity, individuals, language to another and the ability of technologies societies, and the natural environment and its such as satellite navigation or digital assistants to ecosystems. It sets out ten principles for ethical AI: respond to verbal commands. 1. Proportionality and no do harm suggests that AI y Sensors are devices or systems that measure physical should have legitimate objectives and aims that are properties such as temperature or pressure and appropriate to the context, and based on rigorous transmit this data to other electronics (such as a scientific foundations. computer processor). Sensors are one method of 2. Safety and security suggests that AI should not gathering the data used in AI. They are a fundamental cause damage and must protect against security part of the Internet of Things (IoT), systems in which risks throughout its life cycle. actions are undertaken without human intervention 3. Fairness and non-discrimination suggests that AI based on inputs from different sensors (Mahdavinejad systems should avoid bias, and that access to AI et al., 2018). A simple example would be an IoT and its benefits should be shared at national, local irrigation system that gathers information from and international levels, and be equally distributed sensors embedded in soil and activates a watering without preference for ‘sex; gender; language; device accordingly.7 religion; political or other opinion; national, ethnic, indigenous or social origin; sexual orientation; Ethical AI gender identity; property; birth; disability; age; or other status’. As noted, AI has a wide range of applications and 4. Sustainability suggests that the social, cultural, many demonstrable benefits. For instance, AI provided economic and environmental impact of AI important insights and issued alerts early in the technologies should be continuously assessed in COVID-19 pandemic. However, the use of AI also the context of shifting goals. raises a number of ethical considerations. Bias can be 5. Privacy suggests that data for AI is collected, used, introduced into AI through the datasets used and the shared, archived and deleted in ways that protect choices of developers, leading to discrimination. Due to the individual agency of data subjects, and that elements such as the hidden layers of some types of AI, ‘legitimate aims’ and a ‘valid legal basis’ are in place the processes and factors in AI decision-making cannot for processing personal data. be seen, checked or redressed by humans, raising issues 5 See, for example, https://towardsdatascience.com/building-a-chatbot-with-rasa-3f03ecc5b324 6 For more information, see https://www.ibm.com/topics/computer-vision 7 See for example https://www.digiteum.com/iot-solutions-agricultural-irrigation-system 10 K-12 AI curricula — A mapping of government-endorsed AI curricula 6. Human oversight and determination suggests that orientation. Together, these might be called ‘AI literacy’. humans or other legal entities bear responsibility AI literacy comprises both data literacy, or the ability for AI ethically and in law. to understand how AI collects, cleans, manipulates, 7. Transparency and explainability suggests that and analyses data; and algorithm literacy, or the ability people should be aware of when decisions are to understand how AI algorithms find patterns and based on AI algorithms, and that individuals and connections in the data, which might be used for social entities should be able to request and receive human-machine interactions. This is an attempt to explanations for those decisions, including insights frame the scope, structure, and main categories of the into factors and decision trends. Explanability is emerging field of AI literacy. This term has been used to detailed further: ‘outcomes, and the sub-processes guide the study presented in this report. leading to outcomes, should be understandable and traceable, appropriate to the use context’. Pedagogical concepts and 8. Responsibility and accountability reinforces the terminologies principle of human oversight and determination, and suggests that impact assessment, monitoring, ‘Competence-based education’ (CBE) is a model and due diligence mechanisms should be in place often pursued in higher education and TVET, but it is to ensure accountability for AI systems. Auditability8 increasingly being applied in various forms to K–12 must be ensured. education. CBE is intended to transition education from 9. Awareness and literacy refers to the responsibilities models of fixed time and flexible learning, to flexible of governments as well as the public sector, time and fixed learning. In CBE models, students are academia and civil society to promote open and expected to demonstrate applied knowledge, skills and accessible education and other initiatives focused values in context through assessments, and they are on the intersections of AI and human rights, in given as much additional support as needed until they order to ensure that ‘all members of society can meet the required benchmarks (NCLSorg, 2017). take informed decisions about their use of AI systems and be protected from undue influence’. At the heart of CBE is the concept of ‘competence’, a 10. Multi-stakeholder and adaptive governance and term which has evolved to describe ‘the mobilization of collaboration suggests that states should regulate knowledge, skills, attitudes and values to meet complex data generated within and passing through their demands’ (OECD, 2019, p. 5). The intended competencies territories; that stakeholders from a broad range of a curriculum are usually expressed through learning of civil organizations, and the public and private outcomes, or what a student is expected to know, sector should be engaged throughout the AI life understand and be able to do upon completion of a cycle; and that measures need to be adopted to course of study (Biggs and Collis, 1982; Cedefop, 2017; allow for meaningful intervention by marginalized Kinta, 2013). The terminology ‘learning outcome’ is a groups, communities and individuals. modification of the earlier term ‘learning objective’ which ensures that the focus of the statement is on students’ AI literacy actions or achievements rather than those of lecturers, and further are defined using measurable applications The synthesis report of the UNESCO International (Lopez et al., 2015; Sinha, 2020). The relationship Forum on AI and the Futures of Education under the between curricula, learning outcomes and competence theme of Developing Competencies for the AI Era (Miao is complex in actualization but theoretically quite and Holmes, 2020) noted that the world’s citizens direct: a curriculum describes a set of intended learning need to understand what the impact of AI might be, outcomes, and assessments of students demonstrate what AI can and cannot do, when AI is useful, when their attainment of these outcomes through the its use should be questioned, and how it might be application of knowledge, skills and attitudes/values steered for the public good. This requires everyone to within the domain or subject of study and, ideally, in achieve some level of competency with regard to AI, new domains – what Biggs and Collis’s (1982) SOLO9 including knowledge, understanding, skills, and value taxonomy refers to as ‘extended abstract’ capacity. 8 While auditability is not explicitly defined in the Recommendation, this term refers to the ability of third parties to access, review, monitor and criticize algorithms (Jobin et al., 2019). 9 SOLO stands for ‘structure of observed learning outcome’. 11 K-12 AI curricula — A mapping of government-endorsed AI curricula The frameworks and curricula examined for this developmental considerations and appropriate report also reference constructivism, constructionism, technology were critical support mechanisms. computational thinking and design thinking. One final tool which is presented in the context of some ‘Constructivism(s)’ is a broad series of concepts in of the curricula included in this study is ‘design thinking’. academia that apply to the ways in which knowledge It is presented as ‘an analytic and creative process that is created or constructed (and at times co-constructed) engages a person in opportunities to experiment, by individuals through interactions with each other create and prototype models, gather feedback, and their physical, cultural and institutional or systemic and redesign’ (Razzouk and Shute, 2012). Originally environments (Taber, 2016). The types of constructivism developed in fields such as archeology, marketing and often applied in education are built largely upon the economics (Buchanan, 1992), design thinking started work of Piaget (1972), who outlines a theory of types and emergning within industry in the early 1990s, where it forms of learning which are and are not accessible to was developed as a consumer-oriented methodology children at various stages of development; for example, to design innovative products or business models, concrete application would precede abstraction. particularly those involving technology (Hobcraft, 2017). The process of design thinking includes A related concept is ‘constructionism’, the philosophy empathizing (for example with consumers), defining that students learn best through applying knowledge a problem statement, generating ideas for solutions, to projects which hold a personal interest for them and then prototyping and testing in an iterative design (Papert and Harel, 1991). Constructionism is particularly cycle until a desirable innovation is achieved (Hasso applicable to digital curricula due to its origins in the Plattner Institute of Design, 2010). In schools, design domains of ICT and mathematics and its preoccupation thinking can offer a clear procedure for responding to a with the ways in which meaning is generated through need for digital as well as interdisciplinary activities and the process of engaging, manipulating and changing competences. digital artefacts (Kynigos, 2015). Though constructivists and constructionists have a common base, constructionists challenge the hierarchies of knowledge Existing frameworks of reference that were set out by Piaget (1972), generating on AI curricula arguments that students can productively engage with There are a few recent initiatives to map or create AI more complex concepts at younger ages through the curriculum frameworks for grades K–12. These include use of digital media and methods such as block-based three that are detailed in this section: AI Literacy: programming (Papert, 1996). Competencies and Design Considerations, the AI4K12: K-12 AI Guidelines,10 and the Machine Learning Computational thinking, or the series of mental and Education Framework. This is not an exhaustive list, as a physical processes undertaken to build a digital number of NGO, industry and academic organizations solution to a problem (identifying a problem, breaking and/or individuals have developed AI curriculum it down into parts, building and assimilating solutions, frameworks to support their own programmes and and testing and refining them), is theorized to apply undertakings. Some of these frameworks are in use by to an array of domains outside of computer science governments, such as the Microsoft Computer Science (Lodi and Martini, 2021). The four established ‘parts’ Framework, and are included in the learning outcomes of computational thinking are sometimes cited as mapping later in this report. The three frameworks decomposition, abstraction, analysis and algorithms covered in this section were developed with the (Kush, 2019). Lee et al. (2011) studied a range of primary purpose of informing the development of AI computational thinking initiatives in grades K–12 curricula by a range of partners and are not linked to and determined that its processes could indeed specific products or courses. be deployed by students of varying demographic backgrounds. They further proposed a ‘use-modify- create’ learning progression model for engaging with computational thinking, and noted that skilled teachers, 10 See https://ai4k12.org 12 K-12 AI curricula — A mapping of government-endorsed AI curricula AI Literacy: Competencies and Design Their scoping study reveals 17 competencies and Considerations 13 design considerations. The descriptions indicate that for this proposal, competencies are universally Long and Magerko (2020) present a series of at the lower levels of a knowledge taxonomy, largely competencies and design considerations for AI literacy confined to understanding, describing, and identifying. based on a scoping study of existing research, which The competencies proposed by Long and Magerko are sought to determine emerging themes in 1) what outlined in Table 1. AI experts believe a non-technical audience should know, and 2) common perceptions and misconceptions among learners. Table 1. AI Literacy Competency Framework Competency Description / learning outcomes 1. Recognizing AI Distinguish between technological artefacts that use and do not use AI. 2. Understanding Critically analyse and discuss features that make an entity ‘intelligent’. Discuss differences between intelligence human, animal, and machine intelligence. 3. Interdisciplinarity Recognize that there are many ways to think about and develop ‘intelligent’ machines. Identify a variety of technologies that use AI, including technology spanning cognitive systems, robotics and ML. 4. General vs narrow AI Distinguish between general and narrow AI. 5. AI strengths and Identify problem types that AI does/does not excel at. Determine when it is appropriate to use AI weaknesses and when to leverage human skills. 6. Imagine future AI Imagine possible future applications of AI and consider the effects of such applications on the world. 7. Representations Understand what a knowledge representation is and describe some examples of knowledge representations. 8. Decision-making Recognize and describe examples of how computers reason and make decisions. 9. ML steps Understand the steps involved in machine learning and the practices and challenges that each step entails. 10. Human role in AI Recognize that humans play an important role in programming, choosing models, and fine-tuning AI systems. 11. Data literacy Understand basic data literacy concepts. 12. Learning from data Recognize that computers often learn from data (including one’s own data). 13. Critically interpreting Understand that data requires interpretation. Describe how the training examples provided in an data initial dataset can affect the results of an algorithm. 14. Action and reaction Understand that some AI systems have the ability to physically act on the world. This action can be directed by higher-level reasoning (e.g. walking along a planned path) or reactive impulses (e.g. jumping backwards to avoid a sensed obstacle). 15. Sensors Understand what sensors are and that computers perceive the world using sensors. Identify sensors on a variety of devices. Recognize that different sensors support different types of representation and reasoning about the world. 16. Ethics Identify and describe different perspectives on the key ethical issues surrounding AI: privacy, employment, misinformation, ‘singularity’,11 decision-making, diversity, bias, transparency and accountability. 17. Programmability Understand that agents are programmable. Source: Long and Magerko, 2020 The design considerations proposed by Long and demands and child development theory, and the Magerko (2020) focus on pedagogical and learning positioning of AI within learner contexts. The 15 specific methods, but also on social and interpersonal elements. design considerations that the researchers present are: Overall, they emphasize experiential learning and 1. Explainability: Include graphical visualizations, relevant material, an appreciation for cognitive simulations, explanations of agents’ decision-making 11 Describes the point at which AI becomes more intelligent than humans, and can be accompanied by concerns that AI would intentionally harm humans. 13 K-12 AI curricula — A mapping of government-endorsed AI curricula processes, or interactive demonstrations in order to like games or music when designing AI literacy aid learners’ understanding of AI. interventions. 2. Embodied interactions: Design interventions 13. Acknowledge preconceptions: Allow for the fact in which individuals can act as or follow the that learners may have politicized or sensationalized agent, as a way of making sense of the agent’s preconceptions of AI from popular media, and reasoning process. This may involve embodied consider how to respect, address, and expand on simulations of algorithms and/or hands-on physical these ideas in learning interventions. experimentation with AI technology. 14. New perspectives: Introduce perspectives that are 3. Contextualizing data: Encourage learners to not as well-represented in popular media (e.g. less- investigate who created the dataset, how the data publicized AI subfields, balanced discussions on the was collected, and what the limitations of the dataset dangers and benefits of AI). are. This may involve choosing datasets that are 15. Low barrier to entry: Consider how to communicate relevant to learners’ lives, are low-dimensional and AI concepts to learners who do not have extensive are ‘messy’ (i.e. not cleaned or neatly categorizable). backgrounds in mathematics or computer science 4. Promote transparency: Promote transparency in (e.g. by reducing the prerequisite knowledge/skills, all aspects of AI design (i.e. eliminating black-box relating AI to prior knowledge, and addressing functionality, sharing creator intentions and funding/ learners’ insecurities about their ability). data sources, etc.). 5. Unveil gradually: To prevent cognitive overload, give AI4K12: Five Big Ideas and K–12 AI Curriculum users the option to inspect and learn about different Guidelines system components; explain only a few components at a time; or introduce scaffolding that fades as the The AI4K12 Initiative was launched by the Association user learns more about the system’s operations. for the Advancement of Artificial Intelligence (AAAI), the 6. Opportunities to program: Provide ways for Computer Science Teachers Association (CSTA), and AI4All individuals to program and/or teach AI agents. Keep in 2018 as a joint working group that seeks to develop coding prerequisites to a minimum by focusing national guidelines for teaching K–12 students about AI on visual/auditory elements and/or incorporating (AAAI, 2018). strategies like Parsons problems and fill-in-the-blank code. This group brought together academics, researchers and 7. Milestones: Consider how perceptions of AI are teachers to work towards a comprehensive AI framework affected by developmental milestones (e.g. theory of based on ‘five big ideas’: 1) computers perceive the world mind development), age, and prior experience with using sensors; 2) agents maintain representations of the technology. world and use them for reasoning; 3) computers can learn 8. Critical Thinking: Encourage learners to be critical from data; 4) intelligent agents require many types of consumers of AI technologies by questioning the knowledge to interact naturally with humans; and, at the intelligence and trustworthiness of AI applications. very centre, 5) AI can impact society in both positive and 9. Identities, values and backgrounds: Consider how negative ways. The ‘Five Big Ideas in Artificial Intelligence’ learners’ identities, values, and backgrounds affect poster resource has been translated into 15 languages their interest in and preconceptions of AI. Learning to date,12 and formed at least part of the basis for the interventions that incorporate personal identity or development of curricula in multiple contexts, including cultural values may encourage their interest and several of the curricula researched for this study. motivation. The working group was convened to unpack each of these 10. Support for parents: When designing for families, ideas into a curriculum framework divided into four parts, help parents scaffold their children’s AI learning for grades K–2; 3–5; 6–8; and 9–12. To date, curriculum experiences. guidelines for the first three ‘big ideas’ have been drafted 11. Social interaction: Design AI learning experiences and are currently available for public comment.13 that foster social interaction and collaboration. 12. Leverage learners’ interests: Exploit current issues, everyday experiences, or common pastimes 12 See https://ai4k12.org/resources/big-ideas-poster 13 See https://ai4k12.org/gradeband-progression-charts 14 K-12 AI curricula — A mapping of government-endorsed AI curricula In the guidelines, each ‘big idea’ is subdivided into concept components and associated learning learning concepts, which are further split into concept outcomes for ‘Big Idea 1: Perception’ are summarized components. For example, the learning concepts, in Table 2. Table 2. ‘Big Idea 1: Perception’ concepts and learning outcomes Learning Concept Learning outcome progression concepts components K–2: Identify human senses and sensory organs. Living 3–5: Compare human and animal perception. things 6–8: Give examples of how humans combine information from multiple modalities. 9–12: N/A K–2: Locate and identify sensors (camera, microphone) on computers, phones, robots, and other devices. Sensing Computer 3–5: Illustrate how computer sensing differs from human sensing. sensors 6–8: Give examples of how intelligent agents combine information from multiple sensors. 9–12: Describe the limitations and advantages of various types of computer sensors. K–2: N/A Digital 3–5: Explain how images are represented digitally in a computer. encoding 6–8: Explain how sounds are represented digitally in a computer. 9–12: Explain how radar, lidar, GPS, and accelerometer data are represented. K–2: Give examples of intelligent vs non-intelligent machines and discuss what makes a machine intelligent. 3–5: Use a software tool such as a speech transcription or visual object recognition demo to exhibit Sensing vs machine perception, and explain why this is perception rather than mere sensing. perception 6–8: Give examples of different types of computer perception that can extract meaning from sensory signals. 9–12: Explain perception algorithms and how they are used in real-world applications. K–2: Give examples of the features that one would look for if one wanted to recognize a certain class of objects or entities (e.g. cats) in an image. Feature 3–5: Illustrate how face detection works by extracting facial features. extraction 6–8: Illustrate the concept of feature extraction from images by simulating an edge detector. 9–12: Explain how features are extracted from waveforms and images. Processing K–2: Describe the different sounds that make up one’s spoken language, and for every vowel sound, give a word containing that sound. Abstraction 3–5: Illustrate how sequences of sounds can be recognized as candidate words, even if some sounds pipeline: are unclear. language 6–8: Illustrate how sequences of words can be recognized as phrases, even if some of the words are unclear. 9–12: Illustrate the abstraction hierarchy for speech understanding, from waveforms to sentences. K–2: Demonstrate figure/ground segmentation by identifying the foreground figures and the background in an image. 3–5: Illustrate how the outlines of partially occluded (blocked) objects in an image differ from the full Abstraction shapes of the objects. pipeline: 6–8: Describe how edge detectors can be composed to form more complex feature detectors, e.g. for vision letters or shapes. 9–12: Demonstrate how perceptual reasoning at a higher level of abstraction draws upon earlier, lower levels of abstraction. K–2: Describe some things an intelligent agent must ‘know’ to make sense of a question. 3–5: Demonstrate how a text-to-speech system can resolve ambiguity using context, and how the error Types of rate increases with ungrammatical inputs. domain 6–8: Classify a given image and then describe the kinds of knowledge a computer would need in order Domain knowledge knowledge to understand scenes of that type. 9–12: Analyse one or more online image datasets. Describe the information that the datasets provide and how this can be used to extract domain knowledge for a computer vision system. K–2: Discuss why intelligent agents need to understand other languages. 3–5: Discuss how domain knowledge must be broad enough for all the groups an application is intended to serve. Inclusivity 6–8: Describe how a vision system might show cultural bias if it lacked knowledge of objects not found in the culture of those who created it. 9–12: Describe some of the technical difficulties in making computer perception systems function well for diverse groups. Source: AI4K12 (2020) Each big idea is broken down in a similar manner outcomes, the curriculum guidelines offer examples of with a concrete learning outcome pathway from early the ‘enduring knowledge’ that students are expected primary school to high school. In addition to these to retain, for example: ‘Sounds are digitally encoded 15 K-12 AI curricula — A mapping of government-endorsed AI curricula by sampling the waveform at discrete points (typically represents the affective domain (why it matters) and several thousand samples per second), yielding a ‘hands’ represent the psychomotor domain (what you series of numbers’ or ‘The spoken language hierarchy can do with it) (Gazibara, 2013; Singleton, 2015; Sipos is: waveforms  articulatory gestures sounds et al., 2008). This integration has also expanded the  morphemes  words  phrases  sentences.’ concept of competence to include social and emotional Sometimes the learning outcomes and enduring skills (European Parliament and Council of the European knowledge are further unpacked, as was this second Union, 2006; Mulder, 2007). example: ‘To go from noisy, ambiguous signals to meaning requires recognizing structure and applying In addition, Lao (2020) draws on: domain knowledge at multiple levels of abstraction. y theories of constructionism, or the idea that learning is A classic example: the sentences “How to recognize enhanced when undertaken through the construction speech” and “How to wreck a nice beach” are virtually of an item that has personal meaning for the students; identical at the waveform level.’ y computational thinking, a proposed reframing of familiar competence concepts to apply concretely Occasionally, activities are suggested. For instance, to to the programming world: technical concepts, explain decision trees at the grade 3–5 level, ‘the “guess programming practices, and perspectives on an the animal” game, troubleshooting problems, and the individual’s relationships with technology; Pasta Land activity’ are recommended. y a model for understanding the learning outcomes for computational thinking lessons, divided into The big ideas are mutually reinforcing. For example, abstraction, or the ability to apply concepts to new ‘Big Idea 3’ leverages the knowledge from sensing use cases; automation, or utilizing a computer to components to facilitate a discussion of differences in increase efficiency in repeated tasks; and analysis, or how people and computers learn. Similarly, it builds reflection on a student’s assumptions and methods of on the knowledge of processing components to implementation (Lee et al., 2011). equip learners to label a dataset for machine learning, y Use-Modify-Create (UMC), a tiered progression often train classifiers and engage AI concepts such as employed in computational thinking lessons, in which decision trees, neural networks, supervised learning, students first engage with existing software, and unsupervised learning, and reinforcement learning. then modify it to fit new needs, and finally create new software (Lee et al., 2011). The Machine Learning Education Framework The Machine Learning Education Framework (outlined Although it never mentions competence-based in Table 3) consists of six ‘minimally required courses education, the Machine Learning Education Framework for ML-engaged citizens’, and is targeted to a ‘tinker/ (Lao, 2020) follows the well-known CBE framework of consumer’ audience (Lao, 2020, p. 61). In her framework, knowledge, skills and attitudes (which have in other Lao makes the argument that understanding bias contexts included items such as abilities and/or values) and the social implications of AI are fundamental (Brewer and Comyn, 2015; CANTA, 2014; European requirements for all skills. Parliament and Council of the European Union, 2006). CBE has in the past been criticized by some for its lack of attention to the meaning of the task for students and a reductionist view of competence which, while firmly rooted in the context of performance, is less sensitive to individual factors like prior experience and the flexibility to tap into external resources, e.g. the knowledge of teammates (Rutayuga, 2014). However, the gradual integration of theories such as constructivism and experiential learning (Brunner, 1990; Kolb, 2015; Piaget, 1972; Williams, 2017) has resulted in a competence-based framework that focuses on ‘head, heart and hands’, in which the ‘head’ represents the cognitive domain (what you know about it), the ‘heart’ 16 K-12 AI curricula — A mapping of government-endorsed AI curricula Table 3. The Machine Learning Education Framework with learning outcomes and definitions Knowledge Know what machine learning is (and is not). Understand the entire pipeline of the creation of ML 1. General ML knowledge* systems. Identify when to use a range of ML methods across the breadth of the field (e.g. k-nearest 2. Knowledge of ML methods neighbours, CARTs or decision trees, neural networks, ensemble methods). Understand how different methods work. Understand that systems can be biased, and the different levels and ways in which bias can be 3. Bias in ML systems* introduced. Understand that ML systems can have widespread positive and negative impacts. Consider the 4. Societal implications of AI* ethical, cultural and social implications of what they do. Skills 1. ML problem scoping Determine which problems can and should be solved by ML. 2. ML project planning Plan a solution which is sensitive to both technical and contextual considerations. 3. Creating ML artefacts Use tools to create appropriate artefacts. 4. Analysis of ML design Describe the explicit and implicit design intentions of an ML system. Critically analyse the interactions and results* intentions against how the system can and should be used. 5. ML advocacy* Critically discuss ML policies, products and education. 6. Independent out-of-class Students seek learning experiences outside the classroom. learning Attitudes 1. Interest Students are engaged and motivated to study the topic. Students contribute to and learn from a community of peers and/or broader online 2. Identity and community communities who are interested in ML. 3. Self-efficacy Students are empowered to build new, meaningful things. 4. Persistence Students continue and expand their engagement with ML. *The starred items are the six required courses outlined in the framework. Source: Lao, 2020 Lao (2020) also presents a rubric for evaluating ML y Primary and middle school: Engage applications learning programmes against this framework, setting that allow students to complete specific tasks up the basis for a set of standards at the exit level which using ML, e.g. exploiting neural network and GAN could be built upon. For example, the four ‘top scores’ applications to create art or music, or deploying in the rubric for the four learning outcomes under reinforcement learning to play games, etc.) ‘Knowledge’ are: 3. Bias: Graduates of this course are able to describe how ML systems may come to be unpredictably 1. General knowledge: Graduates of this course biased against specific groups throughout can give a precise definition of machine learning each step of the ML pipeline. They can critically and provide a detailed description of the steps of incorporate the practices of ethical thinking and the ML pipeline with technical and socio-ethical design in their own work. considerations for each step. 4. Social implications: Graduates of this course 2. Knowledge of ML methods: Graduates of this course recognize that it is necessary for the creators of ML are able to accurately discern when to use a range technologies to consider the societal implications of machine learning methods across the breadth of their work. They are able to apply ethical and of the field. They are able to describe core technical cultural perspectives and concepts to the analysis concepts of these methods and comfortably use/ of ML artefacts in comprehensive, interrelational implement them in appropriate applications. (Lao and sensitive ways (i.e. considering privacy, then lists her views on appropriate methods for security, the potential for abuse, and the balance different educational levels: of benefits and harm; and assessing ethnographic y High school and above: K-nearest neighbours, reception and disparate impacts using tools such CART/DT, regression, convolutional neural as stakeholder analyses, ethical matrices and networks; unsupervised methods such as k-means model cards). clustering, principal component analysis, GANs, and embeddings; RNNs/LSTMs; reinforcement learning; transfer learning; and ensemble methods. 17 K-12 AI curricula — A mapping of government-endorsed AI curricula Methodology Data collection refers to structured programmes of learning that cover topics in the field of AI and engage with AI-related Two surveys were distributed, the first to learning outcomes. representatives of 193 UNESCO Member States and the second to over 10,000 private- and third-sector Of the 193 Member States contacted through the actors.14 The surveys asked the respondents to official UNESCO channels of correspondence, a total of report on AI curricula for students in K–12 general 51 responded, indicating at least a general interest in education. Appendix A provides the questions in the the topic. Among them, 29 countries and one territory survey sent to the representatives of Member States. completed the full survey. It was modified only very slightly for the private- and y Representatives from 10 countries reported no AI third‑sector actors. curricula in their country. These are: Bahrain, Canada, After the surveys were returned, the team emailed Colombia, Costa Rica, Estonia, Guinea, Macedonia, the additional questions on learning outcomes, Maldives, Singapore, and the Ukraine. implementation and preparation to the respondents y Representatives from 20 countries and one territory who had indicated that they did have AI curricula at responded that they were aware of at least one AI some stage of development. curriculum that was developed and endorsed by government or is under development. These are: In addition, semi-structured key informant interviews Algeria, Armenia, Austria, Belgium, Canada (Yukon were held with Member State representatives, non- Territory), France, Germany, Jordan, Republic of Korea, profit leaders and developers, academics and industry Kuwait, Lao People’s Democratic Republic, Peru, professionals to gain further clarity on the curricula Portugal, Qatar, the Republic of Bulgaria, Saudi Arabia, and their deployment in schools. Interviews probed Serbia, Slovenia, Syria, and the United Arab Emirates. the motivations for developing the AI curricula, and the reasons for their decisions around implementation In addition, a total of 31 NGOs, academics and industry methods and pedagogies. partners responded to the non-governmental survey and indicated that they had an AI curriculum. Finally, a mapping exercise was undertaken for those curricula which had been drafted or published and All Member State representatives and organizations were available for review. The exercise focused on the

Use Quizgecko on...
Browser
Browser