Making AI Work in Indian Education Gurumurthy Kasinathan January 2020

IV Preface “After all, we make ourselves according to the ideas we have of our possibilities.” V.S. Naipaul There is no doubt that the technological advancement has become the game changer of our times. From the Industry 4.0 discourse launched in Germany in 2011 to the scientific advisory report presented to the former US president Barrack Obama on big data and privacy concerns in 2014, to India’s NITI Aayog Artificial Intelligence for All strategy of 2018. A lot of debates have culminated in the questions about the Future of Work in the context of the International Labour Organisation’s Centenary in 2019. Triggered by the disruptive forces of technology based start-ups and new business models, a new race for innovations and war for talents has arisen and with it, a new form of global and fierce competition. Technology has become the holy grail of progress though it did not take long to realise that there is a social dimension attached to it. The platform economy has had severe effects on the bargaining power of suppliers and workers. Data analytics opened a whole array of ethical questions regarding personal tracking and privacy. Further, technological upgrades create productivity gains by efficiency which in turn requires reduced human labour. This poses a particular threat to emerging economies, like India, which need to create new jobs on massive scale for its young and growing population. The utopia around Artificial Intelligence in the times of jobless growth presents a whole new set of challenges. Is the Indian economy ready to ride the AI wave? Who will benefit from AI: investors, big tech, users, or society as a whole? What is and can be India’s role in this global race for innovation? Is tech gender neutral? What about privacy and user protection? How to ensure decent work and social protection in this new age tech revolution? But mostly, how can we turn AI FOR ALL into a reality? To foster this debate, the FES India Office has teamed up with several experts and organisations across the country to explore ground realities with the objective to understand how technology is already unfolding in selected sectors, draft scenarios of what might happen and to ensure proper safeguards are put in place at the right time. Artificial Intelligence like any other technology is neither good nor bad. It is what we make out of it- the rules and regulations – which define the outcome of the game. Just like other countries, in India too, a mass scale application of AI is far from being established. It is still in a nascent phase and can be moulded into a success story. A success story in India AND an Indian success story for all. Patrick Ruether and Mandvi Kulshreshtha January 2020 Friedrich-Ebert-Stiftung, New Delhi Making AI Work in Indian Education

V Note of thanks Friedrich-Ebert-Stiftung(FES) India office is thankful to its partner IT for Change(ITfC) for preparing this research paper. IT for Change aims for a society in which digital technologies contribute to human rights, social justice and equity. ITfC is preparing a series of research papers for FES tracing AI’s enigma in sectors such as agriculture, ports, education and fintech. We are grateful to our colleagues at ITfC for preparing the research, drafting this paper and refining the manuscript to reflect our joint vision. We have to express our appreciation to the Editorial team at ITfC, for their constructive contribution and valuable time during the course of this research. Making AI Work in Indian Education

VI Contents List of abbreviations....................................................................................................................................... VII List of illustrations......................................................................................................................................... VII Acknowledgment......................................................................................................................................... VIII 1. Nature of education..................................................................................................................................1 1.1 Context of Indian education.........................................................................................................1 1.2 ICT in Indian education.................................................................................................................2 2. AI and education.......................................................................................................................................3 2.1‘Datafication’ in education............................................................................................................ 3 2.2 AI implementation in education....................................................................................................5 3. Issues for consideration............................................................................................................................6 3.1 Non availability of big data...........................................................................................................6 3.2 Contestation around ownership of data........................................................................................6 3.3 Proprietary ownership of algorithms..............................................................................................7 3.4 Reinforcement of bias...................................................................................................................7 3.5 Compromising education..............................................................................................................9 3.6 Exacerbation of larger socio-economic problems............................................................................9 4. Recommendations.................................................................................................................................. 10 4.1 Promote teacher and student participation in AI design............................................................... 10 4.2 Broaden understanding of AI and technology through education................................................. 11 4.3 Develop frameworks for collecting, storing and sharing data in the public interest........................ 11 4.4 Establish AI in education frameworks that are free and open....................................................... 12 4.5 Promote research on AI and education........................................................................................ 12 Endnotes....................................................................................................................................................... 13 Bibliography.................................................................................................................................................. 16 Making AI Work in Indian Education

VII List of abbreviations AI ASER FOSS ICT IDRC Artificial Intelligence Annual Status of Education Reporting Free and Open Source Software Information Communication Technology International Development Research Centre IP GDPR NITI NROER USA Intellectual Property General Data Protection Regulation National Institution for Transforming India National Repository of Open Educational Resources United States of America List of illustrations and tables Illustration 1. Model for AI in education............................................................................................................ 4 Illustration 2. AI bias......................................................................................................................................... 8 Making AI Work in Indian Education

VIII Acknowledgements This research report was commissioned by FriedrichEbert-Stiftung, India Office. Author: Gurumurthy Kasinathan Editorial assistance: Jai Vipra, Prakriti Bakshi and Amla Pisharody Graphic Support: Meenakshi Yadav and Prakriti Bakshi Making AI Work in Indian Education

Nature of education 1 This paper explores the potential impact of Artificial Intelligence(AI) on educational processes and outcomes. It examines key concerns with respect to rights-based development that AI policy and programming frameworks should address to support ethical, inclusive and universal 1. Nature of education Education is accepted as a public good that needs to be universally available. As Dewey discusses in‘Democracy and Education’, education is essential for individuals to participate in a democratic society, enjoy their rights and fulfil their responsibilities.1 The primary purpose of education is to develop in each individual a sense of autonomy and agency, and to enable the individual to negotiate community life in fulfilling ways. The 2005 position paper by the Indian National Curriculum Framework on‘Aims of Education’ views education as a socially transformative project for realising constitutional aims of building a just, free and equal society of tomorrow. 2 If education has to be universally accessible and socially transformative, a strong public education system that can cater to the needs of all, specially the poor and marginalised sections of society, is essential. Privately funded education would not be able to meet the requirement of universal access. Martin Carnoy discusses how privatisation of school education in Chile and New Zealand increased stratification. 3 In India such stratification Education as a market plays out with the rich good tends to be going to well-resourced inequitable. A strong private schools and the public education poor going to“affordable system is essential private schools” of doubtful for supporting socio- quality. 4 Thus, education as economic mobility, which a market good tends to be is indispensable for social inequitable, and a strong transformation. public education system is essential for supporting socio-economic mobility, which is indispensable for social transformation. education in India. The paper is based on interviews with educationists and organisations working in the area of AI and education, and desk research conducted by IT for Change. 1.1 Context of Indian education The Indian school education system is among the largest in the world. The public education(government) system in the country, however, is weak and ill-resourced. It is battling old challenges of inadequate infrastructure, and a poorly qualified and demotivated teacher force(with teachers facing isolation and lack of trust). Teachers see their role as completing the syllabus, ensuring textbook contents are memorised by students and reproduced in written examinations, which is the primary mode of assessing learning. In this process, students often The public education fail to develop conceptual system in the country is understanding. This reliance weak and ill-resourced. on“rote-memorisation” It is battling old of content, as opposed to challenges of inadequate “meaning-making”, is a infrastructure, and a salient cause of low learning poorly qualified and levels in schools. 5 The Annual demotivated teacher Status of Education Reporting force. (ASER) reports point out that a large percentage of students in grade eight are unable to read a grade two text or solve a grade two arithmetic problem. 6 Secondly, due to reasons such as large class size and inadequate number of teachers(corresponding to the number of sections in a school), teachers find it difficult, and in addition, may not be motivated to engage individual learners and understand their learning levels, learning needs and interests. Content, pedagogy and assessment usually follow the“one-size-fits-all” approach, leading to poor learning levels. 7 Thirdly, Indian education is facing newer challenges of increasing privatisation and commercialisation of the sector, resulting in increasing stratification. Making AI Work in Indian Education

2 Nature of education 1.2 ICT in Indian education To study the introduction of AI within this context of school education in India, we need to understand the status of Information Communication Technology(ICT) programs in schools. While elite private schools have adopted ICT, the ICT@School program of the Government of India has been largely ineffective in most states of India. Several studies of these programs have indicated that teachers do not integrate ICT into their teaching, and ICT infrastructure remains unused due to this and other reasons, such as poor maintenance and support. NITI Aayog, Government of India’s policy think tank, in its discussion paper The NITI Aayog titled‘National Strategy perspective seems to cast for Artificial Intelligence’, teachers and students declares that“adoption as passive users who of new technologies is should readily consume still lacking, however,(this programs designed and is) often attributed to implemented for the unwillingness of teachers school system. and students to adopt technology.” 8 The NITI Aayog perspective seems to cast teachers and students as passive users who should readily consume programs designed and implemented for the school system. ICT programs in schools have followed the BOOT (Build, Own, Operate and Transfer) model, in which the government has not just outsourced the procurement, installation and maintenance of the ICT infrastructure to private companies, but also the curriculum and teaching of ICTs. The model has largely been a failure with ICT activities remaining distinct and dis-embedded from the regular pedagogies of the school. The private vendor finds little commercial value in integrating ICT literacy aspects into pedagogical processes, as this is resource The systemic failure and time intensive. Contrary in technology-aided to NITI Aayog’s view about learning is owing to the “unwilling teachers and lack of school-centric students”, the systemic design that can empower failure in technology-aided students and teachers learning in reality is owing alike for impact on to the lack of school-centric educational outcomes. design that can empower students and teachers alike for impact on educational outcomes. This background is relevant in unpacking AI in school education as well. Making AI Work in Indian Education

AI and education 3 2. AI and education Education has been highlighted by the NITI Aayog discussion paper, referred to earlier, as one of the five sectors where AI has high potential for solving societal needs. It suggests potential use cases of AI, including“augmenting and enhancing the learning experience through personalised learning, automating and expediting administrative tasks, and predicting the need for student intervention to reduce drop out or recommend vocational training.” 9 Indian schools are usually organised into grades based on the age of the learner. However, classrooms are not homogeneous, and learning levels can vary widely across students. Teachers, usually, do not have adequate information about the differential learning levels and diverse contexts of each student in their classroom. In this context, big data can be used to develop algorithms that can address the diverse learning needs of learners by: • Teaching through personalised education where custom content, pedagogy and assessment can be developed, based on an understanding of each student’s learning level and contexts, and her/his responses to past activities and assessments. • Providing opportunities for self-learning through adaptive practice(similar to the above, where the learner’s responses to questions and activities provides the algorithms the information required to further revise/refine the content and activities that are more in line with learner needs and contexts). • Macro diagnostics and predictive models, across groups of learners(by geography, demographic profile, grade, medium of instruction, subject and other categories) to provide inputs for policy and program. José van Dijck elaborates how“real-time data about individual learning processes help instructors monitor students’ progress and allow for corrective feedback. Personalised data allegedly provide unprecedented insights into how individual students learn and what kind of tutoring they need. And aggregated data about learning behaviour provide the input for individual ‘adaptive learning’ schemes”. 10 Thus, the starting point for AI in its current avatar is the availability of large volumes of data that code written by humans and generated by machines could process In education, big to derive correlations and data would comprise patterns for recommending information on learning action to promote learning. transactions involving In education, such big data the iterative use of would comprise information teaching-learning on learning transactions resources structured involving the iterative through learning use of teaching-learning activities as well as resources(curricular content) student responses to structured through learning these activities, which activities(pedagogy), as well can be evaluated. as student responses to these activities, which can be evaluated(assessment). 2.1‘Datafication’ in education What organisations aiming to design and develop AI applications in education require to do is capture large volumes of resources(curricular content) with related metadata––subject area, relevant grades, language, type of resource, format of file, level of resource and role of the resource(such as introducing a topic, reinforcing learning) in teaching. They would capture actual use of these resources in different learning situations through teaching-learning activities(pedagogies) and the responses of the learners to the content through assessment processes that seek to ascertain the level of learning/understanding of the student, both before and after the transaction with a resource. The student responses to a variety of questions/ quizzes would be analysed to identify conceptual gaps, misconceptions and evaluate the extent of learning (assessment), based on which, the algorithm would suggest learning experiences that could address the gap/ misconception. Students’ responses to these revised experiences would again be captured in an iterative manner. Making AI Work in Indian Education

4 AI and education Recording millions of such responses from students to different resource units for a concept and to combinations of these resource units with different transaction methods, with feedback(on its effectiveness in ensuring learning) would help the algorithm identify the“effective” learning resources and methods given a learner context and purpose of learning. 11 AI in Education Illustration 1. Model for AI in education Making AI Work in Indian Education

AI and education 5 While AI may not completely replace a teacher, it is seen as having the potential to assist teachers in efficiently and effectively managing multi-level/multi-grade classrooms, by judging learning levels While AI may not of individual students completely replace a and allowing automated teacher, it is seen as development of customised having the potential educational content to assist teachers in adapted to each student’s efficiently and effectively class and learning level. managing multi-level/ Assessing work done by a multi-grade classrooms. student on each part/page of the learning material, for example, would allow real-time feedback on student performance to help the teacher appropriately tailor her guidance to the child. 2.2 AI implementation in education In India, digital education in schools is still a long way from being universalised and basic infrastructure is still not available in most schools, including electricity. 12 The availability of digital curricular content is increasing steadily owing to Open Educational Resource(OER) initiatives of the central and state governments in India. The National Repository of Open Educational Resources (NROER) and Diksha(National Teacher Platform) are wellknown OER platforms initiated by the Government of India. In the absence of meaningful ICT integration in schools, the availability of digital data covering student interactions is negligible. However, An Indian company there are private sector Education Initiatives initiatives that seek to collect (EI) has developed student data through their algorithms that can applications/platforms. For analyse student instance, an Indian company responses to develop called Education Initiatives sequence of learning (EI) has collected data from activities for a learner, millions of assessments based on their learning through its Mindspark profile. product, implemented in many states of India, over the last 18 years. Its Mindspark application has been used by over two million students, which has provided data that EI is analysing to provide learning paths based on learner contexts. 13 EI has developed algorithms that can analyse student responses to develop sequence of learning activities for a learner, based on their learning profile. 14 The algorithms interact with students through select sets of questions, and identifies their current learning levels and challenges and is able to provide a report on the learning level of each student in a short time. Based on the current learning level, the algorithm is able to suggest content for each student. The same approach can also be used to support self-learning. Another Indian company, Byju’s too claims to provide personalised learning paths based on analyses of learner contexts. 15 Google and Microsoft have projects in the state of Andhra Pradesh, the former with Vishakapatnam corporations schools and the latter with the state government, to analyse school drop-out data. Many ICT programs in Indian schools use tablets that provide combinations of pre-packaged off-line content and on-line content to students. As the use of phones and tablets increases, and these gadgets play a greater role in accessing the internet, collection of As the use of phones student demographic and and tablets increases, transaction data becomes and these gadgets play a easier and commonplace. In greater role in accessing the USA, the widespread use the internet, collection of personal digital devices of student demographic by students already allows and transaction data vendors like Google to becomes easier and collect large volumes of data. commonplace. While full-fledged personal analytics and personalised learning is perhaps still some time away, we can imagine a future where AI algorithms would predict learning paths for each learner. Making AI Work in Indian Education

6 Issues for consideration 3. Issues for consideration There are several challenges to the use of AI in education. A report from International Development Research Centre (IDRC) identifies five types of short-term to medium-term risks associated with the application of AI. 16 Two of these risks—biases and lack of transparency in decision-making by AI applications––are inherent to the technology itself. The other three—increased surveillance and loss of privacy, automation leading to job loss and targeted misinformation—are the result of specific applications of AI in different domains. The first three apply to the education domain. In addition, predictive analytics pose a special threat to the primary purpose of education, which is to develop human agency. 3.1 Non availability of Big Data Currently, schools are dispersed across the country with most of them having limited or no access to digital technologies. Hence, the collection of student transaction data on a large AI in education has scale, required to build not kept pace with AI and refine algorithms, will in other sectors due take some time. As the NITI to non-availability of Aayog report emphasises, student transaction data any implementation of AI on a large scale required must be preceded by efforts to build and refine to digitise the curriculum, algorithms. records of teaching and student performance. This is one reason why AI in education has not kept pace with AI in other sectors, such as transport or agriculture. Nevertheless, technology companies providing solutions to schools could be in a position to gather such data. Google, for instance, is able to collect data of the teachers and of students accessing its G Suite for Education platform. It would also be able to link this data with the data collected through the use of its other applications by the same teachers and students; Gmail, Maps, Photos, videos(Youtube) etc., some of which are a part of the Education Suite itself. 3.2 Contestation around ownership of data There are contentious issues around ownership of data collected around teacher and student work. These include questions regarding who should be allowed to store the data collected, and who should and should not be allowed to access the data. In an interview undertaken for this paper, a representative from Education Initiatives argued that the company owns the data collected on student assessments, as it has been captured by their Mindspark platform. This is similar to Google and Facebook asserting control(as well as ownership) of the data collected through their platforms. The NITI Aayog report suggests that frameworks that allow for sharing of data are needed, by which monopolistic control by the data collector can be avoided. Monopolistic control over data, that too by entities primarily driven by profit motive is dangerous, as users become vulnerable to political and economic surveillance. Zuckerberg’s“privacy is dead” remark reflects a chilling reality of the surveillance state and surveillance In the case of education, capitalism, both of which are the risk of monopolistic a threat to human security control over data and democracy. 17 In the is aggravated, as case of education, this risk the data pertains to is aggravated, as the data children/minors, who pertains to children/minors, are more vulnerable who are more vulnerable to manipulation and to manipulation and not not capable of giving capable of giving informed informed consent. consent. Also, collection of data from childhood increases the danger of complete loss of privacy over time. Frameworks of“individual ownership” of data suggested by the NITI Aayog paper, however, can be ineffective. 18 For instance, the paper suggests that diagnostic centres (which collect data on health parameters of patients) should take the consent of individuals getting imaged, to aggregate and sell their data with compensation. In Making AI Work in Indian Education

Issues for consideration 7 this way, the informal market for data could be nudged towards entering the formal economy. Yet, selling individual data is complex. The terms of compensation would be impossible to decide in a fair manner. Since we have a situation of a large number of sellers and few buyers, it would mostly result in the buyer giving a‘take it or leave it’ situation similar to the situation of individuals installing apps on their phones. In addition, selling learner data in schools is fraught with ethical and pragmatic challenges. Who should sell(or license), the school or individual parents? Can they sell to multiple buyers? Should they? Do schools and school systems have the right to share, let alone sell data of their students? Should only aggregated data be made available to third parties and not individual or transaction data? Yet, it is unclear how data ownership, sharing and control issues have been dealt with, in the few implementations that have AI ambitions in Handing over data of India. Handing over data students to corporate of students to corporate entities without entities without safeguards safeguards can result in can result in public outrage, public outrage. as the education department in the state of Karnataka found out when it asked a private entity to digitise student data. 19 Another challenge will be that AI models usually require data be collated at centralised points for big data analytics. This will tend to centralise control and ownership. However, decentralisation is an important aim of education planning, as school autonomy are essential to quality education. 3.3 Proprietary ownership of algorithms Control over algorithms with the platform would subordinate the pedagogical aspirations of teachers and learners to the commercial ambitions of the provider. In their AI avatar, digital technologies are not merely “tools” that teachers and learners“use”, but “platforms” on which they interact. Tools can be subject to the control of the users, while platforms retain control of interactions. As Van Dijck says,“these emerging digital policy instruments transfer the assessment of didactic and pedagogical values from teachers and classrooms to(commercial) online platforms deploying real-time and predictive analytics techniques.” 20 Control over the algorithms with the platform would subordinate the pedagogical aspirations of teachers and learners to the commercial ambitions of the provider. The NITI Aayog report, while acknowledging that algorithms and data used in an AI powered application are key elements in ensuring operational success, suggests that“it is imperative that the Intellectual Property Regime in the context of AI be robust and enforceable for innovators to have the confidence that they will be able to make profits from and take credit for their work. This is essential for the promotion of innovation, entrepreneurship and core and applied research in the field of AI.” 21 However, this“strong IP regime” contradicts the NITI Aayog’s desire for a wide dissemination of AI to all countries. In case of proprietary AI regime, what will disseminate is the technology for“use”, on terms given by the provider. 22 3.4 Reinforcement of bias AI does not provide purpose, but only an efficiency enhancing process. Algorithms constituting the AI are not “neutral” or“objective”, as biases of the programmers get into the code and get amplified in Machine Learning (ML). This becomes evident in the outcomes, such as in the cases of criminal profiling in countries like the US, where algorithms have discriminated against AfricanAmericans, or in credit scoring and insurance premia determination. 23 By building on existing understanding of the processes and role of education, AI solutions can exacerbate existing socio-economic divides. In the Indian education scenario, the risk of bias is very high. Predominant“folk-pedagogies” 24 in India privilege education in English medium over the local/regional language, focus on“cracking the examinations” over making meaning. 25 The current syllabi overwhelmingly reflect the urban, upper caste, upper class, Hindu and male bias of the syllabi developers. 26 Derivations of such Making AI Work in Indian Education

8 Issues for consideration content would also reflect these biases. The teacher profile also reflects similar biases, with teachers in government and aided schools, largely belonging to a class and caste different from their students. Data sets embodying biases are bound to lead to predictive models that reinforce social discrimination, further impoverishing learning possibilities of students, especially those belonging to marginalised sections. This would bely the transformative potential of education in India. inappropriate curricular India is a stratified and feudal society, with widespread marginalisation due to caste, gender, religion, class. The caste system’s danger is its requirement that the son of a manual scavenger would be best placed to become a manual scavenger. Such existing biases and harmful prejudices may get amplified through AI. Rather than tackle the structural deficiencies in education that promote rote/drill learning, materials and a privileging of aptitudes over inclinations, AI may simply legitimise them. Data sets embodying biases are bound to lead to predictive models that reinforce social discrimination, further impoverishing learning possibilities of students, especially those belonging to marginalised sections. This would bely the transformative potential of education in India. The NITI Aayog vision of using AI for identifying students for whom vocational education is appropriate, could even end up reinforcing vocational education for those sections of society that have been traditionally marginalised into traditional low paying occupations, segmenting new age service sector jobs as more appropriate for the meritocrats. AI would continue, in a more sophisticated manner, to do what the caste system did in terms of sifting people into different categories based on socioeconomic backgrounds and assigning them education and vocational possibilities that further reinforce the stratification. By reproducing a future based on the past, AI discourages aspirational advancements. AI sifts students on the basis of success and failure in certain subjects or areas, which perpetuates already existing caste, gender and class biases Algorithmic sorting of students for vocational education resulting in access to only low paid jobs Illustration 2. AI bias Making AI Work in Indian Education Algorithmic sorting of students for vocational education resulting in access to higher paying professions

Issues for consideration 9 3.5 Compromising education AI has focused on individual learning. However, education psychology, research and theory emphasises that learning is not an entirely individual phenomenon and requires the learner to interact with other learners and the teacher, who can help extend learning. 27 The reduced emphasis in AI on social aims of education(which require interactions amongst learners) would tend to diminish the socialtransformation role of education by emphasising a process termed as“learnification.” Van Dijck, J.,& Poell, T. deplore how: data-driven, personalized education initiatives focus on learning rather than education, and on processes rather than on teachers and students. The(social) activity of learning is broken into quantifiable cognitive and pedagogical units, such as instruction, short quizzes, assignments, deliberation with other students, and tests. The‘learnification’ model is predicated on… learning rather than its long-term outcome, which is, in most schools, to provide an education. 28 What is most needed in education is the development of a sense of discrimination, critical thinking and openended exploration, often touted as“21st century skills.” The danger is that this is what AI-based education will weed out. 3.6 Exacerbation of larger socio-economic problems Technology led capitalism, whose power will exponentially increase with AI, is driven by overexploitation of natural resources. Pre–AI automation has been reducing employment opportunities, which AI will severely aggravate. Automation is already leading to an increase in income and wealth inequalities, as resources are diverted more and more to the richer classes(who benefit from automation) and away from the others(as unemployment keeps real wages suppressed). Thus, the“efficiency” frameworks fostered by the current AI paradigm will exacerbate the larger challenges humanity faces today—environmental degradation, unemployment and inequality. Making AI Work in Indian Education

10 Recommendations 4. Recommendations The biggest contribution that AI should make to Indian education is to promote“learner-centred” learning, where curricular content and The biggest contribution teaching methods would be that AI should make adapted by each teacher to to Indian education is the needs and contexts of to promote“learnereach learner, moving from centred” learning, the dominant“one-size-fitsmoving from the all” approach. AI could do dominant“one-size-fitsthis by suggesting diverse all” approach. and contextual content and pedagogy possibilities, encouraging teachers to expand their limited content and pedagogy practices. However, an enabling policy environment is essential so that teachers are empowered to understand the various options with respect to content and pedagogy available and decide what to use and how. Else, AI would become another force disempowering the teacher to accept AI chosen content and pedagogy choices in their practice. Given the innumerable and urgent social concerns with respect to AI in education, a comprehensive policy framework in the domain becomes imperative. 4.1 Promote teacher and student participation in AI design AI models need to promote teacher participation in design and implementation and not treat the teacher as an uninformed consumer of AI products and services. It would be useful here to consider the failure of the BOOT model of ICT integration, which treated the school and teachers as mere consumers of proprietary digital products. On the other hand, the participatory model of the IT@Schools program in Kerala, where teachers led the integration of ICT in schools, is a successful alternative. 29 The Kerala model of integrating ICT through teachers is now gradually being taken by up other states as well, thus avoiding the outsourcing of core curricular and pedagogical processes to external business entities. Learning from this experience is important for designing and deploying AI programs in education. The purpose of education is to build agency and a sense of participation, a principle that AI programs would need to adhere to. Teacher educators opening up a desktop computer, in a workshop on Digital Literacy in the District Institute of Education and Training, Dharwad, Karnataka, 2013 Photo Credit: IT for Change Making AI Work in Indian Education

Recommendations 11 Education should also support the learner to become a “responsible citizen.” This means educational processes must enable the learner The use of AI needs to question dominant to be appropriate to perspectives and beliefs. educational contexts and Going against the grain may its direction designed in often be essential for positive consultation with and progress in society. Hence, participation of teachers the use of AI needs to be and educators. appropriate to educational contexts and its direction designed in consultation with and participation of teachers and educators. Leslie Loble, Deputy Secretary in the New South Wales Department of Education, Australia, emphasises that it is crucial that educators are Teachers and school in the driver’s seat when leaders must play a it comes to designing and critical role in defining developing AI-based systems. a clear purpose for AI Teachers and school leaders in the classroom and be must play a critical role in trained to understand defining a clear purpose and utilise it effectively. for AI in the classroom and be trained to understand and utilise it effectively. Leslie adds that students must also be involved in decisions about the use of these technologies and educated about the ethical frameworks that accompany their use. Their future will depend on the policies and approaches that are adopted now. 30 4.2 Broaden understanding of AI and technology through education If AI has the potential to harm the processes and purpose of education, the long-term solution is education itself. There is an urgent need to introduce AI in secondary and tertiary education, to build awareness of its implications on education, as well as on society. If schools and teachers have to be in control of AI implementation, education about AI becomes critical. If schools and teachers have to be in control of AI implementation, education about AI becomes critical. Currently, AI is largely the preserve of a microscopic section of society–– researchers, think tanks, few large digital corporations, first world governments. The impact of AI, however, is society-wide and its negative impact is likely to be felt more by developing countries and marginalised communities across the world. 31 Solutions to the problems that AI will bring have to be sought in the political realm rather than in the technological realm. Bringing about widespread awareness of AI and its potential and problems is an urgent imperative towards raising political awareness. Undergraduate courses should build critical perspectives on AI in their foundation courses in science, sociology, economics and political science, going beyond the hype that characterises popular discourse on AI. This will be necessary to design technology towards public welfare. Secondly, appropriateness in technology adoption is an important political question that education must deliberate. This has been raised even decades ago by people like Gandhi, Schumacher and Lewis Mumsford. At this point, where the use of AI may bring about even greater power imbalances, this question becomes critical to address. 4.3 Develop frameworks for collecting, storing and sharing data in the public interest Frameworks that ensure beneficial use of community data that is collected, without harming students and teachers, will need to be established as a pre-requisite to allowing Data must be aggregated data collection in schools. and anonymised before Data must be aggregated and anyone beyond the anonymised before anyone school/teachers can beyond the school/teachers access it. While data can access it. While data would need to be would need to be collectively collectively owned by owned by the school, teachers the school, teachers and and parents(as guardians), parents(as guardians), the school would need to be the school would need to its custodian. be its custodian. The danger of dilution of school autonomy and teacher/ learner agency from centralisation of data collection and storage can be avoided by considering federated implementations of AI, with data and algorithms controlled Making AI Work in Indian Education

12 Recommendations at the school level and at the district level federation of schools, rather than being entirely centralised. The notion of‘community ownership’ of data referred to in the report of the Justice Srikrishna Data Protection Committee, should be further explored to support decentralised ownership, control and storage. 32 Recently, NITI Aayog has suggested that fintech data should be shared with competitors to avoid data monopolies/oligopolies. 33 Another option could be data cooperatives set up by the public education system, where public data sets are developed and data access, storage and use are based on maximising public interest rather than profit. 4.4 Establish AI in education frameworks that are free and open Like any curriculum framework or model, algorithms must be released as open source, so that they can be subject to public review and audit. This will allow them to be scrutinised for the curricular assumptions they make, the educational aims they serve and the biases they hide. It is also necessary that teachers should have the possibility to understand the workings of AI they use. If the AI is private/proprietary, it would not be possible for the teacher to find this out even if she wants to. Open AI can provide an opportunity for teachers(or interested others) to try to understand the working of the algorithms and provide inputs for further tweaking. Kerala favoured FOSS and OER in its ICT programs in education, and this has allowed teachers to be cocreators and active participants. This design ensured scaling/sharing as well as Free and Open avoiding rentals on software data sets and Free and content. Free and Open and Open source AI could be an alternative to algorithms can help the NITI Aayog perspective create a collaborative of seeing digital technology environment and inhibit provisioning as an essentially data monopolies/ commercial process. Also, oligopolies. Free and Open data sets and Free and Open source algorithms can help create a collaborative environment and inhibit data monopolies/oligopolies. 4.5 Promote research on AI and education Any change in education has a huge impact as it affects future generations. The IDRC report warns that“if we continue blindly forward, we should expect to see The use of AI in increased inequality alongside education should be economic disruption, evidenced based, social unrest and in some through research and cases, political instability, pilots. with the technologically disadvantaged and under-represented faring the worst.” 34 Hence, the use of AI in education should be evidenced based, through research and pilots. Research should go beyond first-order effects, such as increased efficiencies or accuracy of diagnosis, to include broader social effects. Practices that support participatory design and development of inclusive AI applications need to be studied, including understanding ways by which these practices can counter bias and make AI relevant to marginalised communities. Research on the benefits and risks and benefits of open AI(for example, sharing AI resources, datasets) is needed. Research should connect supply side questions (how best to provide open access to AI algorithms, tools and datasets) with deepening understanding of the engagement necessary to ensure that open AI resources are available for(re)use and adaptation by diverse populations(and not just by those who are already skilled and resourced). While research would help us deepen our understanding of AI in education, an appropriate policy environment is critical to ensuring that AI can help education to deliver on its promise of being a force for social transformation, rather than of bondage. Making AI Work in Indian Education

Endnotes 13 Endnotes 1 John Dewey, Democracy and Education,(New York: Macmillan Company, 1916), https://en.wikisource.org/wiki/Democracy_and_Education. 2 “National Focus Group on Aims of Education” NCERT, March 2006, http://www.ncert.nic.in/new_ncert/ncert/rightside/ links/pdf/focus_group/aims_of_education.pdf. 3 Martin Carnoy,“School Choice? Or Is It Privatization?” Educational Researcher 29, no. 7(October 2000), https:// www.researchgate.net/profile/Martin_Carnoy/publication/250183359_School_Choice_Or_Is_It_Privatization/links/53d28d760cf2a7fbb2e9a6b7/School-Choice-Or-Is-It-Privatization.pdf. 4 Geetha B. Nambissan,“Low-Cost Private Schools for the Poor in India: Some Reflections,” in India Infrastructure Report 2012(New Delhi: Routledge, 2013), http://www.idfc.com/pdf/report/2012/Chapter_8.pdf. 5 Emphasised in policy and curricular documents, such as National Curricular Framework, 2005. 6 Annual Status of Education Reports, can be accessed through http://www.asercentre.org/Keywords/p/346.html. 7 Content(‘what’ of learning), pedagogy(‘how’ of learning) and assessment(did learning happen?) are the three components of the educational process. 8 NITI Aayog,“Discussion Paper: National Strategy for Artificial Intelligence,” June 2018, 35, http://www.niti.gov.in/ writereaddata/files/document_publication/NationalStrategy-for-AI-Discussion-Paper.pdf. 9 Ibid., 20. 10 José Van Dijck and Thomas Poell,“Higher Education in a Networked World: European Responses to U.S. MOOCs,” International Journal of Communication 9, no. 2015(September 2015): https://pdfs.semanticscholar.org/c9d0/a7933dad4c445740997a232721eb34bf9633.pdf. 11 Along with the profiles of the teachers and the learners. 12 The Right to Education Act, 2000, prescribes minimum infrastructure that every school in India must have. It is estimated that around 10 percent of schools in the country meet these requirements. 13 Based on interactions with Data Science team from EI. 14 Based on interview with EI team member working in their AI program. 15 Author could not get an interview with the Byju’s team to get more information about their program 16 Matthew Smith and Sujaya Neupane,“Artificial Intelligence and Human Development: Toward a Research Agenda,” IDRC, April 2018, https://www.idrc.ca/en/stories/artificial-intelligence-and-human-development. Making AI Work in Indian Education

14 Endnotes 17 The shipping connectivity index is based on the five components of the maritime transport sector: number of ships, their container carrying capacity, maximum vessel size, number of services and number of companies that deploy container ships in a country’s ports. 18 Barbara J. Evans,“Much Ado about Data Ownership,” Harvard Journal of Law& Technology 25, no. 1(Fall 2011): http://jolt.law.harvard.edu/articles/pdf/v25/25HarvJLTech69.pdf. 19 See https://bangaloremirror.indiatimes.com/bangalore/others/education-dept-strikes-out-digital-deal/articl show/63048569.cms, parents expressed apprehension about sharing of their children’s data with a private entity. This agreement was subsequently rescinded, on public protest. 20 José Van Dijck and Thomas Poell,“Social Media Platforms and Education,” ed. Jean Burgess, Alice Marwick, and Thomas Poell, The SAGE Handbook of Social Media: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3091630. 21 NITI Aayog,“Discussion Paper: National Strategy for Artificial Intelligence,” June 2018, 60, http://www.niti.gov.in/ writereaddata/files/document_publication/NationalStrategy-for-AI-Discussion-Paper.pdf. 22 Ibid., Data democratisation(pg 48) v/s IP(pg 46). 23 Kirsten Martin,“Ethical Implications and Accountability of Algorithms,” Journal of Business Ethics, 2018, https://link. springer.com/article/10.1007/s10551-018-3921-3. 24 Jerome Bruner,“The Culture of Education,”(Cambridge, MA: Harvard University Press, 1996), http://www.cs.kent. ac.uk/people/staff/saf/share/great-missenden/reference-papers/brunerFolkPedagogy.pdf. 25 An important cause for rote memorisation as the predominant mode of“learning”. 26 For instance, though agriculture is the livelihood of a majority of the population of India, agricultural practices, ideas have very little place in text books. 27 The theory of social constructivism suggests that knowledge is constructed through interaction with others. 28 José Van Dijck and Thomas Poell,“Social Media Platforms and Education,” ed. Jean Burgess, Alice Marwick, and Thomas Poell, The SAGE Handbook of Social Media: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3091630. 29 Gurumurthy Kasinathan.“Policy Brief- ICTs in School Education- Outsourced versus Integrated Approach,” August 2009, IT for Change, https://itforchange.net/policy-brief-%E2%80%93-icts-education-outsourced-versus-integrated-approach 30 Leslie Loble,“Learning to Live in the Time of AI,” UNESCO Courier, August 17, 2018, https://en.unesco.org/courier/2018-3/learning-live-time-ai. 31 Matthew Smith and Sujaya Neupane,“Artificial Intelligence and Human Development: Toward a Research Agenda,” IDRC, April 2018, https://www.idrc.ca/en/stories/artificial-intelligence-and-human-development. 32 B. N. Srikrishna and Data Protection Committee,“A Free and Fair Digital Economy Protecting Privacy...,” Government of India, July 27, 2018, http://meity.gov.in/writereaddata/files/Data_Protection_Committee_Report.pdf. Making AI Work in Indian Education

Endnotes 15 33 Yogima Sharma,“Niti Aayog Bats for Ending Data Monopoly” ET Tech, May 17, 2019, https://tech.economictimes. indiatimes.com/news/technology/niti-aayog-bats-for-ending-data-monopoly/69366570. 34 Matthew Smith and Sujaya Neupane, 20182019, https://www.cbp.gov/border-security/ports-entry/cargo-security/ ctpat. Making AI Work in Indian Education

16 Bibliography Bibliography Burt, Andrew, and Samuel Volchenboum.“How Health Care Changes When Algorithms Start Making Diagnoses.” Harvard Business Review. May 08, 2018. https://hbr.org/2018/05/how-health-care-changes-when-algorithms-start-making-diagnoses. Bruner, Jerome. The Culture of Education. Cambridge: Harvard University Press, 1996. http://www.cs.kent.ac.uk/people/ staff/saf/share/great-missenden/reference-papers/brunerFolkPedagogy.pdf. Carnoy, Martin.“School Choice? Or Is It Privatization?” Educational Researcher 29, no. 7(October 2000). https:// www.researchgate.net/profile/Martin_Carnoy/publication/250183359_School_Choice_Or_Is_It_Privatization/links/53d28d760cf2a7fbb2e9a6b7/School-Choice-Or-Is-It-Privatization.pdf. Cave, Stephen.“To save Us from a Kafkaesque Future, We Must Democratise AI.” The Guardian. January 04, 2019. https://www.theguardian.com/commentisfree/2019/jan/04/future-democratise-ai-artificial-intelligence-power. Chang, Lulu.“Mark Zuckerberg Wants to Share His Daughter Via VR.” Digital Trends. December 11, 2015. https://www. digitaltrends.com/social-media/mark-zuckerberg-daughter-vr/. Dewey, John. Democracy and Education. New York: Macmillan Company, 1916. https://en.wikisource.org/wiki/Democracy_and_Education. Dijck, José Van, and Thomas Poell.“Higher Education in a Networked World: European Responses to U.S. MOOCs.” International Journal of Communication9, no. 2015(September 2015). https://pdfs.semanticscholar.org/c9d0/a7933dad4c445740997a232721eb34bf9633.pdf. Dijck, José Van, and Thomas Poell.“Social Media Platforms and Education.” Edited by Jean Burgess, Alice Marwick, and Thomas Poell. The SAGE Handbook of Social Media: 579-91. https://papers.ssrn.com/sol3/papers.cfm?abstract_ id=3091630. European Commision.“Ethics guidelines for trustworthy AI” December 2018. https://ec.europa.eu/newsroom/dae/document.cfm?doc_id=57112. Evans, Barbara J.“Much Ado about Data Ownership.” Harvard Journal of Law& Technology 25, no. 1(Fall 2011). http:// jolt.law.harvard.edu/articles/pdf/v25/25HarvJLTech69.pdf. Loble, Leslie.“Learning to Live in the Time of AI.” UNESCO Courier. August 17, 2018. https://en.unesco.org/courier/2018-3/learning-live-time-ai. Martin, Kirsten.“Ethical Implications and Accountability of Algorithms.” Journal of Business Ethics, 2018. https://link. springer.com/article/10.1007/s10551-018-3921-3. Muralidharan, Karthik.“Disrupting Education? Experimental Evidence on Technology...” December 2016. https://www. nber.org/papers/w22923. Nambissan, Geetha B.“Low-Cost Private Schools for the Poor in India: Some Reflections.” In India Infrastructure Report Making AI Work in Indian Education

Bibliography 17 2012. New Delhi: Routledge, 2013. http://www.idfc.com/pdf/report/2012/Chapter_8.pdf. “National Focus Group on Aims of Education.” NCERT. March 2006. http://www.ncert.nic.in/new_ncert/ncert/rightside/ links/pdf/focus_group/aims_of_education.pdf. NITI Aayog.“Discussion Paper: National Strategy for Artificial Intelligence.” June 2016. http://www.niti.gov.in/writereaddata/files/document_publication/NationalStrategy-for-AI-Discussion-Paper.pdf. Sharma, Yogima.“Niti Aayog Bats for Ending Data Monopoly- ETtech.” ET Tech. May 17, 2019. https://tech.economictimes.indiatimes.com/news/technology/niti-aayog-bats-for-ending-data-monopoly/69366570. Smith, Matthew, and Sujaya Neupane.“Artificial Intelligence and Human Development: Toward a Research Agenda.” IDRC. April 2018. https://www.idrc.ca/en/stories/artificial-intelligence-and-human-development. Srikrishna, B. N., and Data Protection Committee.“A Free and Fair Digital Economy Protecting Privacy...” Government of India. July 27, 2018. http://meity.gov.in/writereaddata/files/Data_Protection_Committee_Report.pdf. Making AI Work in Indian Education

About the authors Gurumurthy Kasinathan is founder and director of IT for Change. He leads projects in the area of education, including in research, demonstration projects, systemic teacher education reform and policy advocacy. His areas of expertise include open educational resources, ICT integration in education and school leadership. The views expressed in this publication are not necessarily those of the Friedrich-Ebert-Stiftung. Imprint © 2020 Friedrich-Ebert-Stiftung India Office K-70-B, Hauz Khas Enclave| New Delhi-110016 India Licensed under a Creative Commons License Attribution-ShareAlike 4.0 International(CC BY-SA 4.0) Responsible: Patrick Ruether| Resident Representative Mandvi Kulshreshtha| Program Adviser T:+ 91 11 26561361-64 www.fes-india.org FriedrichEbertStiftungIndia To order publication: info@fes-india.org Commercial use of all media published by the Friedrich-Ebert-Stiftung(FES) is not permitted without the written consent of the FES. https://www.fes-asia.org/ Friedrich-Ebert-Stiftung(FES) is the oldest political foundation in Germany. The foundation is named after Friedrich Ebert, the first democratically elected president of Germany. FES is committed to the advancement of both socio-political and economic development in the spirit of social democracy, through civic education, research, and international cooperation. This publication is part of a series on Artificial Intelligence in India, a collection of papers developed under the Socio-Economic Transformation program coordinated by the FES India office. The Wicked Problem of AI governance_ITfC