Feedback on Strategy for Artificial Intelligence in Sri Lanka

A week ago, the Ministry of Technology (MoT) released a draft strategy for AI in Sri Lanka, supported by the United Nations Development Programme (UNDP) to ostensibly “to leverage AI technologies to drive innovation, economic growth and societal progress across the nation”[1].

In my capacity as Research Director at The Disinformation Project based in New Zealand[2] I’ve led grounded research into how generative AI – including multimodal platforms, and LLMs (both black-box/closed-source, and open source) – impacts malign influence operations, truth decay, information integrity as key factors eroding liberal democratic foundations. In the course of this research, I have also used the insights to study the impact of generative AI on the seed, spread, and sophistication of disinformation in Sri Lanka – which I’ve studied since 2013, and also based my doctoral research on.

The following thoughts, penned post haste (more on this later) chiefly respond to, and reflect on the MoT white paper’s presentation that it “…provides an analysis of the prevailing challenges to AI’s rollout in Sri Lanka, proposing a strategic framework to capitalize on AI for societal benefit, elevate living standards, and maintain the nation’s competitive edge. It underscores the imperative of fostering an AI ecosystem conducive to innovation, educational advancement, ethical technology practices, and synergistic alliances between the government, academia, private sector, civil society, and development partners and institutions.”

Performative consultation

The MoT’s website notes that today, 26 April 2024, is the last day public input to the white paper will be considered. This is exactly a week after the white paper was released, on 19 April. Considering the weekend (20, and 21 April), the public consultation was thus only open for five days.

This is a farce. It suggests neither the MoT nor UNDP (in the service of the Government of Sri Lanka) are interested in robust public input or consultation.

Just five days to read, reflect on, and pen a response to a substantive, significant white paper indicates that what drives this process of public input is purely performative, and a check-box against a requirement which can now be cosmetically marked as undertaken. This theatre is regrettable, especially when it is supported by the United Nations, and a line Ministry. It only ensures that critical, diverse, gendered, and grounded views, and perspectives from the public on this document, and what is signals aren’t comprehensively captured. This means that what’s proposed in the white paper risks being implemented by default, as frameworks, ideas, strategies, and potential policies projected as those which passed muster when presented to the public, because no one critiqued them robustly.

Importantly, the commendable presentation around meaningful public consultation in Section 3.3 of the white paper (page 22) is completely undermined by what’s noted above. What hope or confidence can we place in the MoT, GoSL and even the UNDP around what’s noted in just this section, if this first, consequential consultation is so ill-considered, and executed?

This white paper proposes frameworks that will impact, and in profound ways, the lives of all citizens. It is extremely regrettable that the stakeholders, including MoT, and the UNDP didn’t undertake a more robust consultation – including through the availability of this document in Tamil, and Sinhala. No public consultation in Sri Lanka can be considered meaningful or effective if the only language a document is presented in is in English.

Recommendations

Based on above, and given the impossibility of substantive, well considered feedback in the time allotted, what follows are high-level recommendations based on a hurried reading of the MoT’s white paper. I say this because what follows may be aspects that find some expressed in the white paper as it stands, but because of the need to send this into MoT, I have missed.

Furthermore, the recommendations below must be considered in the context of I’ve outlined further on in this note, as a more grounded critique of the white paper’s presentation of AI in a country enduringly, and deeply marked by a democratic deficit which has seen, and continues to foster conditions ripe for state-sponsored extrajudicial abductions, the silencing of dissent, torture, murder, and related chilling effects, with total impunity.  As the UN itself has noted, “Sri Lanka suffers from a continuing accountability deficit – be it for war crime atrocities, more recent human rights violations, corruption, or abuse of power”[3].

The white paper presents a good foundation, and conceptual scaffolding on or around which AI-based futures in Sri Lanka can be envisioned. The presentation will benefit from greater specificity on the means, and methods of implementation, addressing structural or systemic risks more, detailing funding aspects, and benchmarking global best practices.

More specifically, the GoSL/MoT/UNDP should,

1. Provide more specific targets and metrics for measuring progress around what’s proposed in the white paper. While the draft lays out a good high-level vision for country, and framework around AI’s use, having more quantitative goals, qualitative assessments, and well-defined milestones would make the strategy more actionable. For example, the white paper should specify targets for the number of AI engineers to be trained[4], the amount of AI research funding, key infrastructure improvements etc.
2. Focus more, and expand on the role of the private sector. In any country, the private sector is critical in driving AI adoption, and innovation with the state leading more in regulatory frameworks, and the conceptualisation of meaningful (policy) guardrails. The white paper should more details on policies, programmes, and incentives to spur greater private sector involvement, and investment in AI, including around R&D. Public-private partnerships should be a key focus area, and here too, the white paper can benefit from references to how other countries in South Asia have gone about similar national frameworks. News reports note, for example, that in India ““Over 10,000 Graphics Processing Units (GPUs) will be procured in the public-private partnership to aid a computing system for a high-end Al ecosystem, which will also be conducive to the design of an Al marketplace”[5].
3. Address the significant risks, and challenges of AI more thoroughly, especially around job reconfiguration, losses, and displacement. While the white paper is justified in recognising the potential of AI to aid Sri Lanka’s development, the negative impacts AI will very likely have on the country’s workforce remain under-appreciated. It’s essential to plan a risk mitigation strategy around jobs that will be lost to, or in some way, re-oriented because of AI’s adoption, and adaptation. MoT/GoSL/UNDP should think of ways to manage this significant societal transition, including worker retraining, and upskilling programmes – as well as more robustly, and openly outlining the risks posed by unemployment due to AI. Addressing this proactively, and honestly is important.
4. Provide an estimated budget and funding sources for the various initiatives outlined. Given Sri Lanka’s fiscal constraints especially after 2022’s collapse, innovative, sustainable, long-term, and large-scale financing approaches leveraging international aid, private capital etc will be needed. The costs of AI adoption at a country-scale could potentially run into billions of dollars, including robust research, trust, and safety oversight[6]. Nowhere does the white paper mention how what’s proposed will be adequately, and meaningfully financed. The stress on meaningfully is because the white paper notes “the government has earmarked LKR 1.5 billion in the 2024 budget to foster early AI successes while simultaneously strengthening the essential foundations for AI”. Considering international examples of state-led investments in AI, this figure, though seemingly very large in Sri Lankan rupees, is likely to be wholly inadequate for what’s proposed. Accordingly, ballpark funding estimates are required, broken down by sector, actor, subject, and domain.
5. International benchmarking analysis comparing Sri Lanka’s current AI readiness to peer countries (especially in South Asia). This measure will help calibrate the ambition of MoT’s AI strategy for Sri Lanka, and identify priority gap areas to address based on what has worked (or not!) in other South Asian countries, including, but not limited to India’s pioneering initiatives in this regard[7].
6. Create an AI Regulatory Sandbox to test new AI applications in a controlled, contained environment. Especially given country’s democratic deficit, and regulatory frameworks that are unlikely to be first or even best in class, this will help Sri Lanka (and successive governments) balance oversight with encouraging innovation as AI technologies also evolve rapidly, and often, in ways that are unforeseen. Regulatory flexibility is important, and having a sandbox approach allows for flexibility, creative adaptation, robust testing, and risk mitigation without holding country, and citizens hostage to policies rolled out at pace, and scale without due regard for unintended, and deleterious consequences to (the country’s fragile) democracy, and social cohesion.
7. Emphasise open data initiatives more to fuel an AI innovation ecosystem. Ensuring complete compliance with the Personal Data Protection Act (PDPA), the white paper should detail concrete steps to improve data collection, and data sharing across government entities, and also outline how government datasets will be opened up to researchers, and the private sector. If data is the new oil for AI, measures should be taken to protect Sri Lanka’s data sovereignty, and at the same time, establish ways that AI can fully harness the potential of public data, for public good.

A ‘gaze from nowhere’, and threats to democracy

“A conquering gaze from nowhere”, proposed by the scholar Donna Haraway[8] is used in feminist studies to understand power, and its location(s), as well as the situated nature of critiques. In the context of the MoT’s white paper, this critical framework helps explain a fundamental flaw in presentation.

Nowhere in the white paper are aspects like, and related to Sri Lanka’s socio-political precarity, democratic deficit, and structural racism mentioned. The white paper presents AI’s possible, and potential future(s) in an imagined country, dislocated from grounded realities, and the lived experience of vulnerable sections of society, including, but not limited to the lived experiences of women (reflecting ingrained misogyny at many touchpoints including with the state sector), persecuted minorities, and others who are routinely discriminated against, like the disabled.

Nowhere in the white paper is there a robust recognition of how AI will exacerbate, complicate, amplify, and accelerate threats to Sri Lanka’s fragile democracy – which are already significant[9].

Nowhere in the white paper is there a more general recognition of threats to democracy because of AI’s transnational/extra-territorial development, aside from illiberal domestic adoption, and adaptation – in what’s an increasingly volatile, uncertain, complex, and ambiguous (VUCA) global stage, and Sri Lanka’s a geo-political, and strategic importance in South Asia.

Based on a paper by Sarah Kreps, and Doug Kriner in the Journal of Democracy[10], there are four key threats rendered more sharply, urgently, and enduringly by AI. Not a single aspect is considered by the MoT’s white paper. All of them have a profound, enduringly important impact on equitable, sustainable development, and human rights.

1. Threats to democratic representation: The white paper does not discuss how (generative) AI in particular could be used to generate fake “constituent sentiment” at scale, potentially skewing the perceptions of policymakers around constituent preferences, presentations, and priorities.
2. Threats to democratic accountability: The draft fails to address how (generative) AI could be used to create, and disseminate targeted disinformation, propaganda, and malign influence operations making it increasingly more difficult for citizens (and voters) to hold elected officials accountable based on verified, accurate information.
3. Threats to democratic trust: The white paper does not consider how the proliferation of AI-generated content – including by state-sponsored political actors, and/or their proxies, based on what’s already being done, and present in Sri Lanka’s electoral landscape alone – will invariably serve to erode trust in what academics call societal ‘semantic foundations’. This can be more simply presented as the engineered erosion of public trust in media reporting, and official communications which are crucial for civic engagement, political participation, and respect for shared, liberal democratic values – especially in a country with very weak social cohesion, and a very high distrust in government.
4. Mitigating the threats: While MoT’s draft discusses the need for responsible AI development, and public engagement, it does not provide specific measures to mitigate the threats of generative AI to democracy, such as the creation, and/or adaptation of tools to identify AI-generated content, self-regulation by generative AI platforms, and actors, and, importantly, the promotion of digital literacy among the public (and political elites/policymakers) in order to help them discern authentic from inauthentic content, and the growing risks to polity, society, and democracy by not being able to do so.

Human rights in Sri Lanka

The MoT’s white paper, continuing the “gaze from nowhere” has zero recognition of Sri Lanka’s awful state of human rights, as a defining feature of the country’s democratic deficit. What’s proposed is akin to retrofitting nitrous oxide to the private busses that ply on our roads, and doing nothing to address how they are driven. It is a recipe for disaster, and risks the accelerated decay of social cohesion through human rights violations that will expand, and increase significantly in scale, scope, and significance because of AI’s introduction, and entrenchment in the country’s discriminatory, racist, and violent official structures, systems, and policies (as well as unintended consequences arising from this technology’s access by illiberal state sponsored actors, and their proxies).

The following points are based on the 2023 US State Department Country Report on Sri Lanka[11], which features a critique mirrored through, and well-established in reporting from Amnesty International, Human Rights Watch, the International Crisis Group, as well as domestic civil society organisations, and even the Human Rights Commission of Sri Lanka.

Simply put, the MoT (and especially the UNDP) must honestly confront the Sri Lanka’s enduring human rights challenges head-on, and embed strong human rights safeguards into what’s proposed as AI governance frameworks. Failure to do so risks exacerbating existing inequities, structural racism, systemic violence, and abuses as AI systems, and AI-based touchpoints get deployed at scale in governance, and public service delivery.

To this end, significant short-comings in the MoT’s draft include, but are not limited to,

1. The complete lack of safeguards against AI-enabled surveillance and repression: Given the reports of arbitrary arrests, detention, and harassment of activists, journalists, and minorities by state security forces, against what is a surveillance, and securitised state that features the invasive, rapacious collection of data on citizens clandestinely, the MoT white paper must propose strict guidelines, and oversight to prevent the misuse of AI for further human rights abuses and repression of dissent.
2. Insufficient focus on ensuring equal access and non-discrimination: The State Department’s human rights report highlights discrimination against ethnic and religious minorities, women, LGBTQI+ individuals, and persons with disabilities. This is very well-known, and hardly news to anyone in the country (or keen observers abroad). The proposed national AI strategy should prioritise inclusive development, rights-based/rights-first approaches to AI governance, and meaningful measures to mitigate algorithmic bias, and the resulting offline, and online discrimination in, and through AI systems designed for the public.
3. Not addressing the potential impact on worker’s/employee rights: The State Department report flags key issues with freedom of association, collective bargaining, and acceptable conditions of work. Recalling what was noted earlier around the mitigation of what are likely to be significant disruptions to Sri Lanka’s job market, including – and especially – the state sector, the MoT’s white paper must consider how AI will exacerbate these challenges (e.g., job displacement), and propose measures to protect worker’s rights in grounded AI futures.
4. Failure to emphasise the rule of law and accountability: Given the significant concerns about impunity for human rights abuses, and the near complete lack of accountability (with the resulting culture of impunity), the MoT’s AI strategy must reflect the importance of robust legal frameworks, independent oversight, and grievance redressal mechanisms to check against AI-related human rights violations – given the inevitability such violations will occur, and in Sri Lanka’s context, add to systemic, and structural violence sponsored by the state.
5. Lack of human rights-based/human-rights first approach: Overall, the white paper doesn’t adequately stress a human rights-based approach which places the protection, and promotion of the rights, and especially those which impact vulnerable groups, at the centre of the proposed AI vision for Sri Lanka. The sections titled ‘Creating a Safe and Trustworthy AI Ecosystem for Sri Lanka’, ‘Legal and Regulatory Framework, and ‘Responsible AI Development’ (pages 21 to 23) – while commendably comprehensive in what’s outlined as guiding principles, and international best practice frameworks – is again a “gaze from nowhere”. Nowhere do these sections explicitly acknowledge fundamental human rights issues in Sri Lanka, such as discrimination against ethnic and religious minorities, extrajudicial killings, and torture by state security forces. While the sections mention the importance of the rule of law, MoT’s draft doesn’t outline any meaningful mechanisms for holding the government, and state actors (defined by decades of human rights violations) accountable, and for violations that in the future will likely be committed using AI systems, including through the partisan, parochial capture, misuse, and abuse of AI architectures designed for public good, and service delivery.

Given Sri Lanka’s tragic, decades-old history of state-sponsored abuses, it is crucial to involve civil society, human rights defenders, independent media, researchers, and especially affected communities in shaping AI policies, and regulations – in order to ensure they meaningfully prioritise human rights.

The MoT’s draft features none of this.

The Online Safety Act (OSA), and its impact on AI futures

As a subject/domain expert in policies dealing with online harms, I’ve talked, and written extensively about the devastating impact of the Online Safety Act (OSA) on Sri Lanka’s democratic fabric, and future[12].  The very high potential for the weaponisation and instrumentalisation of the OSA through (generative) AI architectures is a real, urgent, and growing threat to human rights, governance, and social cohesion.

Under the OSA, AI tools can be used at scale to flag content and accounts for takedown requests, which will likely include or involve the targeting of independent, investigative journalists, academics, activists, and political adversaries. There’s an asymmetrical risk to those who are from minorities communities, including Tamils, and Muslims, who engage in critical dissent. The OSA allows so-called, and possibly self-styled “experts” assisting investigations to access user data, and intercept electronic communications with a magistrate’s warrant. Coupled with AI capabilities, this will enable the unprecedented expansion of what’s already expansive government surveillance, and the pervasive profiling of citizens through their online communications, and activities. It is highly likely that the data gathered will be used in discriminatory ways. The Act undermines the necessity in Sri Lanka to create, and maintain anonymous or pseudonymous accounts for activism, and critical dissent. The AI architectures proposed by the MoT AI can be (mis)used to identify, and report such accounts at scale, adding to risks faced by activists holding government accountable.

These are risks, and threats that run completely counter to constitutional safeguards, human rights, and additional protections afforded by progressive legislation like the PDPA. The OSA, combined with artificial intelligence, and machine learning tools risk being used by the state – in what may well be architectures primarily meant for, and supported by the likes of the UNDP for social development, and democratic participation – for mass surveillance, profiling, and highly targeted suppression of dissent at individual, and institutional levels.

The MoT’s white paper must recognise these threats, and risks, and recommend strong, effective safeguards to prevent the  weaponisation of (generative) AI enabled by the OSA.

Concluding thoughts

I recognise, and value the time, and effort spent by the authors of the MoT’s draft around what’s the first comprehensive crack at one of the most significant issues of our times, and of inter-generational consequence. Any omission of material, and evidence in the white paper which addresses what I’ve critiqued above is due to the pressure of submitting this response before the expiration of the deadline. I reiterate that it is really regrettable that the MoT/GoSL/UNDP didn’t allow for more time before closing off submissions, facilitating a more considered, substantive response that would also have done more justice to the intellectual input of those who crafted this draft.

It is important Sri Lanka gets AI right. Nothing short of our democracy is at stake. The white paper’s emphasis that “The strategy must also address AI-related risks, balancing the protection of individual rights and societal harmony with the support for responsible AI development” is timely, important, and valued. It is in the lack of how this vision, and goal is fleshed out, and realised in the rest of the white paper that I have focussed this note around.

Submissions of this nature, especially in Sri Lanka, are often perceived as personal affronts to individuals who have crafted what’s critiqued. This is not my aim. The critique is around (significant gaps in) policies, and frameworks proposed, and not the persons who drafted the white paper. I trust this critique is received in the spirit it was intended all parties concerned, reflecting the fundamental nature of any public consultation.

I hope the MoT is able to iteratively build on this white paper in a more meaningful manner, locate it in the socio-political, cultural, and contextual realities of Sri Lanka, and align it towards the fullest realisation of our democratic potential – which I have no doubt the authors are invested in, the UNDP supports, and we all collectively work towards.

---

[1] Crafting a Comprehensive Strategy for Artificial Intelligence in Sri Lanka – Engaging Public Input for Inclusive Development, https://mot.gov.lk/blog/crafting-a-comprehensive-strategy-for-ai-public-input

[2] https://www.thedisinfoproject.org

[3] Accountability central to Sri Lanka’s future – UN Human Rights report, https://www.ohchr.org/en/press-releases/2023/09/accountability-central-sri-lankas-future-un-human-rights-report#:~:text=GENEVA%20(6%20September%202023)%20–,Office%20report%20published%20on%20Wednesday.

[4] Taken from Realizing Sri Lanka’s AI Vision: Preliminary Matrix of Actions, page v.

[5] Government announces India AI mission: What it is and more, https://timesofindia.indiatimes.com/gadgets-news/government-announces-india-ai-mission-what-it-is-and-more/articleshow/108322391.cms

[6] New Report Details Costs and Structure of a National AI Research Resource, https://hai.stanford.edu/news/new-report-details-costs-and-structure-national-ai-research-resource

[7] National Strategy for Artificial Intelligence, https://www.niti.gov.in/sites/default/files/2023-03/National-Strategy-for-Artificial-Intelligence.pdf

[8] Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective, https://www.jstor.org/stable/3178066

[9] Generative AI, and Sri Lanka’s Online Safety Act, https://www.linkedin.com/pulse/generative-ai-sri-lankas-online-safety-act-sanjana-hattotuwa-ph-d–0lz0c/?trackingId=fJnIXbTeR6WqS%2FWy8lGbhw%3D%3D

[10] How AI Threatens Democracy, https://www.journalofdemocracy.org/articles/how-ai-threatens-democracy/

[11] https://lk.usembassy.gov/2023-country-reports-on-human-rights-practices-sri-lanka/

[12] The Rise of the Panopticon State, https://groundviews.org/2024/02/12/the-rise-of-the-panopticon-state/

The Online Safety Act: A Trojan Horse for Authoritarianism, https://groundviews.org/2023/09/23/the-online-safety-act-a-trojan-horse-for-authoritarianism/

The Online Safety Act’s impact on advertising industry, https://www.linkedin.com/pulse/online-safety-acts-impact-advertising-industry-hattotuwa-ph-d–igbzc/?trackingId=fJnIXbTeR6WqS%2FWy8lGbhw%3D%3D

Meeting Sri Lanka’s trade unions around Online Safety Act (OSA), https://www.linkedin.com/pulse/meeting-sri-lankas-trade-unions-around-online-safety-hattotuwa-ph-d–769tc/?trackingId=fJnIXbTeR6WqS%2FWy8lGbhw%3D%3D

Generative AI, and Sri Lanka’s Online Safety Act, https://www.linkedin.com/pulse/generative-ai-sri-lankas-online-safety-act-sanjana-hattotuwa-ph-d–0lz0c/?trackingId=fJnIXbTeR6WqS%2FWy8lGbhw%3D%3D

The OSA’s first application as red flag, https://www.linkedin.com/pulse/osas-first-application-red-flag-sanjana-hattotuwa-ph-d–chhzc/?trackingId=fJnIXbTeR6WqS%2FWy8lGbhw%3D%3D

The American Privacy Rights Act of 2024 (APRA) & Sri Lanka’s Online Safety Act (OSA), https://www.linkedin.com/pulse/american-privacy-rights-act-2024-apra-sri-lankas-osa-hattotuwa-ph-d–khvjc/?trackingId=fJnIXbTeR6WqS%2FWy8lGbhw%3D%3D

Sri Lanka’s Online Safety Act (OSA): The government’s lies, falsehoods, & deception, https://www.linkedin.com/pulse/sri-lankas-online-safety-act-osa-governments-lies-hattotuwa-ph-d–dvupc/?trackingId=fJnIXbTeR6WqS%2FWy8lGbhw%3D%3D