The overarching principle guiding the use of data and related technologies—artificial intelligence, in this context—should be to preserve, enhance, and advance human dignity.
This policy playbook offers guiding principles and practices on data governance and ethical AI to government and non-government stakeholders in Southeast Asia. It complements, draws from, and builds on the work of others across different fields of study.
Both the principles and practices take a socio-technical approach; that is, that the development and application of data-driven technologies, including AI, must be rooted in the interaction between people—along with their perspectives and knowledge systems—and technology. Consequently, the normative frameworks that govern technological development and application should also reflect this relationship.
Policy Playbook Principles
On data: End user knowledge of how their data is collected, utilized, and stored. Concomitantly, end-user ability to access, store, and even delete their data at discretion.
On AI: End user participation and local context in the research, design, development, deployment, and evaluation of AI applications.
On data: The handling of datasets and training models adheres to the highest level of integrity. Safeguards are in place from possible data breach or cyberattack.
On AI: Incorporating personal, social, and environmental harm minimization in the life cycle of AI systems and platforms.
A social contract to advance justice and redress structural discrimination by enabling users to reap the greatest benefits from the value creation of data and/or AI.
Diversification of perspectives, experiences, skills, languages, and demography in norms framing technology development.
On data: Transparency, safety, and accountability in process of data generation, collection, storage, protection, and destruction.
On AI: Fairness, safety, clarity, and accountability of AI development and deployment, as well as in algorithmic training to minimize and mitigate bias.
Map international concepts to local languages, create a corpus of interpretations
Although terms like “data,” “privacy,” and “algorithm” may carry baseline connotations and constitute part of the common policy vocabulary, they often do not translate well into other languages at the community level because there simply might not be a direct equivalent or because the concept itself might be unfamiliar. Multilingual communities in the five countries surveyed—and in many other Southeast Asian nations, for that matter—may also understand the same word differently in the various languages spoken. Language, after all, is contextual.
Explaining what these terms mean in local languages would be a first step. However, it would be more meaningful to take a converse approach by building an understanding of the concept from those languages rather than translating them from English. This way, convergences and divergences in concepts at the local, regional, and international levels could be identified. More importantly, it would contribute local knowledge, nuance, and perspective to concepts that are often assumed to be universally understood.
Collaborate (hyper-)locally on AI systems design 237
Although AI developers and local communities challenged by a particular concern may have a shared desire to resolve the problem, their differing motivations and data collection methods may complicate the process of achieving that common goal. Whereas AI engineers may be driven to apply and advance science, local communities may instead want solutions for social change. In such cases, conventional research methods that create a one-way relationship where locals inform scientists through interviews, surveys, or focus-group discussions may not be helpful.
Instead, re-centering the role of locals so that they are part of the design and development of AI solutions could contribute to more effective outcomes. One way of doing this could be to encourage the creation of community datasets, which might be protected through a data trust or co-operative (see below), to enrich the AI research process.
Explore alternative data governance approaches 238
Rather than subscribing to the binary model of data flow or data localization, Southeast Asian countries should consider, and where appropriate, adopt or promote alternative data governance approaches.
These include data stewardship arrangements such as data trusts and data co-operatives which pool data into an organization. Whereas in a data trust, trustees would exercise the data rights on behalf of the beneficiaries through a legal fiduciary relationship, a data co-operative would afford participants the opportunity to jointly and meaningfully participate in the value-extraction or monetization of their data.
Create a Data Footprint Registry
Every exchange and interaction within the digital ecosystem leaves a data footprint. Although individuals are usually informed about their data being collected or processed, often the option to destroy one’s data is hardly presented.
Giving users the full visibility of their data through a data footprint registry would allow them to understand the life cycle of their data, from generation, storage, usage, and archival, to deletion. The use of contact tracing apps during the pandemic provides initial lessons on how governments can inform users about the extent of their data footprint and explicitly offer them the choice to opt out or delete their data after a particular transaction.
Advance convergences/consolidation on an ASEAN approach toward inclusive data and ethical AI practices across all three community pillars
While ASEAN member states have begun aligning positions more closely on data governance with the release of documents such as the ASEAN Data Management Framework and the ASEAN Model Contractual Clauses, similar coordinating efforts should be replicated across the group’s two other community pillars: political-security and socio-cultural.
Discussions on data and AI outside the economic realm may require greater sensitivity, but governments should nonetheless begin the process with a view to tabling informed and coherent regional perspectives, particularly in international debates on the role of technology in political and security affairs. Additionally, conversations on the impact of data and ethical AI within the socio-cultural pillar would advance ASEAN’s pledges for a more people-oriented and people-centered outlook, especially considering the pervasiveness of data-driven technologies among the region’s population.
Institutionalize Track Two inter-regional dialogues for regularized exchanges
Countries in and of the Global South are impacted by data and AI trends very differently from counterparts in the north. Non-government stakeholders in emerging regions, notably Africa and Latin America, are responding in innovative ways at the community level. There are also rich conceptual, policy, and legal debates that are provoking new and alternative framings of how data-driven technologies could be governed.
Southeast Asian countries could benefit from this diversity of thought. Regular, multilateral dialogues on these developments at the Track Two level, in particular, could facilitate greater exchange of experiences and insights among Global South countries that share developmental and technological commonalities.
Prioritize representation at international standards-setting forums
The five countries surveyed are represented at technical standards-setting bodies such as the International Telecommunication Union (ITU) and International Organization for Standardization (ISO). Two are participating or observing members of the ISO Joint Technical Committee on Artificial Intelligence, shaping AI standards, including on bias.
Resource and expertise constraints may make participation in technical bodies difficult, let alone at the highest decision-making levels. However, regular involvement in such meetings even as observers can contribute to technical proficiency over the longer term. In the meantime, identifying specific gaps and needs in these areas can help with negotiating tailored capacity-building programs supported by interested partners. Additionally, where possible, governments should also prioritize active participation in international norms and policy discussions related to data and ethical AI. Without at least representation at the table, there can be no thought partnership, let alone leadership, in these evolving conversations.
Compile impact statements
Understanding the potential and actual harms resulting from weaknesses in data protection regimes or obscure AI processes is key to mitigating and preventing them.
Documenting these harms, from personal to societal and environmental, through impact statements, can provide a clearer record of the gaps that need to be addressed. In this sense, the role of a data protection commissioner might be expanded where appropriate and supported by additional resources. In the field of AI, impact statements collated by experts from different disciplines would offer a more comprehensive picture of potential and actual harms. It would also compel greater accountability by and among AI developers.
Extend accountability obligations to government
Subject to narrow restrictions, governments should also comply with similar standards of care expected of others, particularly in protecting the personal data of its citizens. As a matter of good governance, it could also invite a credible third party to conduct environmental audits of its policy decisions on, for example, migrating to the cloud or setting up data centers.
Incentivize business-to-business capacity building
Regulatory compliance can be particularly costly and prohibitive for micro, small, and medium enterprises (MSMEs), which form the backbone of the countries’ economies.
Larger corporations might be incentivized by the government or through their corporate social responsibility programs to offer training or capacity assistance to these MSMEs, particularly in the initial stages of business. Existing models in Japan of conglomerates offering cybersecurity assistance to smaller firms might be a constructive point of reference.
Evaluate existing normative frameworks
Across the five countries surveyed, there is an urgent sense of forward momentum toward digitizing as much as possible of government, the economy, and society. However, as we assert in our introduction, there seems little appreciation of what the end goal of data-driven optimization actually is.
It might be valuable for government and non-government stakeholders, either separately or in consultation with each other, to pause and assess whether the existing normative frameworks on data and AI in which countries are operating will actually help advance national interest in the broadest sense. This will require a deeper philosophical reflection on the end objectives that countries intend to pursue through the use of data-driven technologies, beyond the achievement of a dynamic digital economy.
Reassess metrics of performance and success
Quantitative indicators can be useful to measure progress in areas such as representation, infrastructure capacity, or available talent for emerging tech sectors. However, they can also mislead and provide a false sense of growth. What are the hidden trade-offs of a rise in the number of MSMEs on e-commerce platforms, for example?
Alternative metrics of success in a data-driven environment plotted against the socio-cultural or biodiverse landscape of Southeast Asian nations could include the number of languages and dialects mapped to preserve the diverse heritage in each country, but also to provide more responsive AI solutions to various accented speech. It could measure the acreage of land reserved for environmental conservation or the tonnage of marine plastic waste cleared from coastal waters. The use of data for social good might reorient the metrics of success, reinterpret development for the better, and redress some structural inequities replicated by data-driven technologies.
Mainstream alternative perspectives, including respect for self-governance of data
In light of increasing reports of algorithmic bias and harms, the revival of traditional notions such as Ubuntu in Africa or buen vivir in Latin America that advocate a relational rather than a purely rational approach to AI is a call to reframe the current discourse on AI ethics. These frameworks are community- rather than state-centric and they center the voices of the marginalized, including indigenous communities, with the aim of utilizing data and AI to achieve social progress in harmony with nature.
There is, in parallel, an indigenous data sovereignty movement which sits awkwardly alongside open data initiatives. While the former drives at the right of indigenous people to own and control data about their communities and land, the latter offers a compromise of sorts, allowing sustainable development of indigenous resources within the parameters of the state.
These different perspectives of data governance and, by extension, AI ethics offer a valuable tapestry of knowledge systems from which Southeast Asian countries could learn. They afford conceptual and policy options to the region in its pursuit of agency. More importantly, they prompt stakeholders to reflect more thoughtfully about how ancestral knowledge might be repurposed for a data-driven present and future.
Expand expert stakeholder engagement
Discussions on data and AI often involve scientists, industry, academics, policymakers, and lawyers. But interpreting data and minimizing related harms require broader lenses. Analyses could be contextually enriched and improved upon by having historians, linguists, community activists, philosophers, and religious leaders at the table.
Launch all-female hackathons
Done right, this exercise would not only encourage more girls to code and compete but afford a different, if not safer, space for participants to network and grow the community through mentorship. Of value would be the results from these hackathons that could highlight gendered perspectives in coding solutions for everyday applications. Research has shown that automatic captions are more accurate with male rather than female voices.239 Having women and girls hack these and other flaws could yield innovative fixes.
Introduce security- and dignity-by-design in education and training
Advancing a socio-technical approach to AI requires the integration of security and dignity at the very start of the AI development cycle. The concept of security-by-design has been widely adopted as a risk-management approach. Adding the consideration of dignity to the equation in engineering and computational science courses trains students to prioritize the user as a human being first, rather than as a disembodied consumer of technology.
Incorporating the dignity angle prompts a closer analysis of requirements, available data, and plans to mitigate the potential occurrence of harms.
Establish an AI bias bank
Generally, algorithmic bias is often attributed to computational factors like the quality or quantity of the datasets or the fairness of machine learning algorithms. But bias can also be the outcome of human and systemic factors or the culmination of all these factors. With the availability of off-the-shelf AI solutions, it is becoming easier for individuals and organizations to install applications without fully understanding their risks and harms.
Creating an algorithmic bias bank—a database of use-cases of how bias was produced in the past—can raise public consciousness about the risk, explain more simply how complex codes and algorithms may result in bias, and promote productive exchanges on how to prevent it. By creating space for conversations surrounding AI, the lack of knowledge as well as negative perceptions are mitigated, instilling public trust throughout the process.
237 Yen-Chia Hsu, Ting-Hao ‘Kenneth’ Huang, Himanshu Verma, Andrea Mauri, Illah Nourbakhsh, Alessandro Bozzon, “Empowering Local Communities Using Artificial Intelligence,” Patterns 3, no. 3 (2022), https://doi.org/10.1016/j.patter.2022.100449.
238 Exploring Legal Mechanisms for Data Stewardship, Ada Lovelace Institute and UK AI Council, London: Ada Lovelace Institute, March 2021, https://www.adalovelaceinstitute.org/report/legal-mechanisms-data-stewardship/.
239 Rachel Tatman, “Gender and Dialect Bias in YouTube’s Automatic Captions,” Proceedings of the First ACL Workshop on Ethics in Natural Language Processing (April 2017), Association for Computational Linguistics: 53–59, http://www.ethicsinnlp.org/workshop/pdf/EthNLP06.pdf.