Algorithmic power gap and potential dilemmas in rural revitalization: theoretical analysis and situational deduction

Fri, 13 Jun 2025 00:00:00 +0000

Author: Dylan Chiang

Summary

This article uses the methods of theoretical analysis and situational deduction to systematically explore how artificial intelligence (AI) causes two potential dilemmas: algorithmic bias and power gap in the context of rural revitalization. Research points out that this problem is rooted in the existing digital inequalities and social inequality structures in rural areas, and may further solidify the marginalized status of disadvantaged groups through resource allocation and public services. The conclusion is that without a prudent framework centered on local participation and data governance, AI technology may erode the goals of equity and inclusiveness in rural revitalization.

Keywords: artificial intelligence, algorithm bias, rural revitalization, digital labor, data governance

1. The risk of AI bias in rural resource allocation and public services

Artificial intelligence (AI) is regarded as a potential tool to improve the efficiency of rural governance and promote equitable distribution of resources. However, in the unique rural context of Taiwan, the introduction of AI without careful consideration of its inherent limitations and social implications may not only fail to eliminate existing gaps, but may instead create or exacerbate prejudice, posing a potential threat to rural revitalization.

The decision-making quality of an AI system highly relies on the data it learns. However, in rural areas of Taiwan, especially remote mountainous areas, outlying islands, or communities with a very high proportion of elderly people (such as Yunlin Taixi, Pingtung Manzhou, etc.), the popularity of digital infrastructure, residents’ digital access capabilities, and the activity of data production systems are still significantly “digitally weak” compared to urban areas. This directly leads to the fact that rural data that can be used for AI training may face the dilemma of “little quantity and poor quality”, or even a fundamental “absence of data”.

Just imagine, if an AI system is used to evaluate the qualifications for natural disaster relief in agriculture and fishery, its main training data comes from those farmers and fishermen who are highly informatized, good at using online systems to report disaster damage, and whose crops or breeding types are more mainstream marketing channels. Then, for those older farmers and fishermen who still rely on traditional farming methods, are not familiar with digital tools, or grow/raise products with local characteristics but small market sizes, their data may be difficult to be fully captured or correctly interpreted by the system. In this case, the AI model may make unfavorable judgments about these “data invisible people” from the beginning due to insufficient data representation and systematic bias, putting them in a more vulnerable position in resource allocation. This exactly confirms that algorithmic bias often stems from the existing “truth of inequality” in society. In rural areas, this “inequality” is first reflected in the availability of data and the ability to truly reflect the appearance of the land.

The development and verification processes of many AI models currently used in public services often take urban environments as the main scenarios, and their built-in logic and parameter settings may unconsciously have the color of “urban centralism” or “mainstream group defaults.” If this type of model is directly implemented in rural areas without sufficient local adaptation and contextualization considerations, it may be “blind” or “incompatible” with the unique social and economic structure of the countryside, multicultural features (such as the traditional norms of aboriginal tribes, the industrial characteristics of Hakka settlements, and the living needs of new resident communities) and the heterogeneous needs of residents.

For example, if an AI system used to optimize public transportation routes and schedules in urban areas is simply applied to rural areas with vast territory, scattered settlements, and special travel needs of the elderly population (such as bus services in the Chiayi Mountains), it may be unable to understand the special transportation needs of rural residents for flexible reservations, ride-sharing services, or specific periods of time (such as agricultural product harvesting periods, temple fairs), and make resource allocations that are not in line with actual benefits. This neglect of local context essentially constitutes a structural bias against the particularity of rural areas, making it difficult for AI’s “wisdom” to truly respond to the real pain points of rural areas.

When AI systems process information involving vulnerable groups in rural areas (such as the elderly living alone, single-parent families with financial difficulties, people with physical and mental disabilities, new resident women with limited digital capabilities, etc.), even if there are no clear discriminatory rules, the underlying data patterns and algorithm logic may have hidden negative labels or resource crowding effects that are not easily noticeable. Drawing on the analytical perspective of “poverty-phobia”, AI may inadvertently replicate society’s stereotypes or implicit biases against specific groups in seemingly objective data analysis.

For example, if an AI system is used to screen social welfare or emergency assistance application eligibility, if in its training data, the cases of “successful” receiving subsidies are more likely to present some “typical” dilemma narratives or easily quantifiable indicators of difficulty. Then, for those disadvantaged cases in rural areas who are in the same difficult situation, but whose plight is more complex and difficult to present through standardized data, or who do not meet the mainstream society’s imagination of “ideal help seekers”, the AI system may give them a lower priority or even exclude them from potential service targets because their characteristics deviate from the “normal” model. This kind of implicit discrimination based on algorithms will make those disadvantaged groups in the countryside who are already in “invisible corners” face the risk of further marginalization and deprivation of resources in the AI era, running counter to the goal of rural revitalization to promote inclusion and equity.

2. The operation and inequality of algorithmic power in rural context

The application of AI systems in rural areas not only brings potential risks of bias, but may also create new asymmetries in power relationships. The design, deployment and operation process of algorithms are often dominated by specific technical elites, policy makers or business institutions, and the rural community itself has relatively weak voice and control in it. This power gap may make AI a tool to strengthen existing inequalities and even create emerging relationships of dominance.

The complexity and technical threshold of AI systems constitute an insurmountable barrier for many rural residents. Whether it is the underlying logic of AI decision-making, the way data is collected and used, or the explanation and appeal channels for system output, rural communities are often in a disadvantaged position due to information inequality. For example, when a local government introduces an AI system to assist agricultural land use planning or disaster potential assessment, ordinary farmers may have difficulty understanding how the system makes judgments, let alone effectively questioning its decisions or participating in system improvements. This is consistent with what Veale et al. have observed in the public sector. Even civil servants may lack a full understanding of the operation of AI systems, resulting in over-reliance or selective adoption. In rural contexts, this information gap may be more serious.

In rural areas of Taiwan, many elderly people or residents with relatively low education levels have limited digital literacy. If the operation interface of the AI system is not friendly enough, or the relevant training and instructions are insufficient, they are more likely to be marginalized in the AI era. AI technology that seems to be designed to “empower” rural areas may become a new “empowerment” mechanism if it lacks sufficient consideration and support for the abilities of local users, reducing rural residents from technology users to passive data providers and decision-making recipients.

The design and deployment process of AI systems often lacks substantial participation of rural communities. Decision-making power is often concentrated in the hands of central government, local government technocrats, or external AI solution providers. The local knowledge, practical needs and value concerns of rural residents, local organizations, and even grassroots civil servants may be ignored or simplified in the technology-led development process. This echoes Sampath’s concerns about the insufficient participation of local communities in the development of AI in countries in the Global South and their vulnerability to dominance by external forces.

For example, in the process of promoting “smart agriculture”, if the development of AI systems focuses too much on pursuing the single goal of increasing output or reducing costs, while ignoring the multiple values that small farmers attach to ecological sustainability, agricultural product quality, or rural cultural inheritance, it may lead to a disconnect between technical solutions and local needs. Many rural areas in Taiwan have unique natural environments and cultural features (such as the traditional areas of aboriginal tribes and the cultural landscape of Hakka villages). If the planning of the AI system fails to fully incorporate the opinions of local residents and respect their lifestyle and value choices, it may lead to questions about “Whose countryside? Whose AI?” and even rights conflicts over “Who does AI serve?” This risk of being “determined” will seriously weaken the subjectivity of rural communities and the momentum for local independent development.

When AI systems in rural areas mainly rely on external sources (such as large urban technology companies or multinational enterprises) to provide technology, platforms and data analysis services, we need to be alert to the risks of “technological colonization” that it may bring. Externally led AI solutions may carry with them the values and business logic of their developers (usually urban elites or Western cultural backgrounds), which may not be consistent with the local context of rural areas. This analogy to the dilemma of “digital colonization” or “data dependence” that countries in the Global South may face as Sampath refers to it may also appear in different forms in urban-rural relations.

For example, if the health passbook AI system in rural areas is trained mainly based on urban population data, its disease diagnosis model or treatment recommendations may ignore the impact of rural-specific genetic background, living habits or environmental factors on the disease, thus affecting its applicability and accuracy in rural areas. Furthermore, over-reliance on standardized, externally provided AI systems may also lead to the marginalization or even elimination of valuable knowledge accumulated in rural areas (such as traditional farming methods, local herbal knowledge, operational experience of community mutual aid networks, etc.). Zajko emphasized that sociology should pay attention to how AI affects power relations. In the rural context, this also includes an examination of the ebb and flow of power between local knowledge systems and external technological systems. One of the goals of rural revitalization is to activate local characteristics and resources. If the introduction of AI fails to effectively integrate with local knowledge and instead crowds out local wisdom, it will run counter to the original intention of rural revitalization.

In short, if the application of AI in rural areas fails to face up to and actively deal with the unequal operation of algorithmic power caused by problems such as technical thresholds, insufficient participation, and external leadership, AI will not only be difficult to become a tool to promote equitable rural development, but may deepen the existing power structure, causing rural communities to face new dilemmas and challenges in the digital age.

3. Potential impact on rural revitalization goals

Taiwan’s rural revitalization policy, whether it is rural regeneration in the early days or the local revitalization emphasized in recent years, has as its core goals the promotion of balanced regional development, revitalization of the local economy, and improvement of quality of life, with a special emphasis on social equity and inclusiveness, striving to allow everyone in the countryside to share the fruits of development. However, the risks of bias hidden in the rural applications of AI analyzed above are inconsistent with the operation of power. If they are not effectively controlled, they may have a direct and far-reaching negative impact on these noble policy goals.

Social equity and inclusiveness are the cornerstones of rural revitalization policies, which aim to narrow the gap between urban and rural areas and ensure equal opportunities among different groups within the countryside. However, the existence of algorithmic bias may cause AI systems to unconsciously replicate or even strengthen existing social inequalities in resource allocation and public service decisions, thus eroding the basis of policy fairness.

For example, if a set of AI is used to evaluate the priority of subsidy for “local creation” proposals, Systems whose training data or built-in evaluation indicators are overly focused on quantifiable economic benefits, technological innovation, or past successful urban cases may make rural proposals that are more focused on cultural heritage, ecological protection, community building, or caring for the disadvantaged (for example, a A traditional craft revitalization project led by tribal elders, or a local ingredient promotion cooperative initiated by mothers of new residents), due to their “non-mainstream” or “difficult to quantify” characteristics, are at a disadvantage in the ranking, causing subsidy resources to flow to projects that are more in line with mainstream imagination. Not only does this fail to promote diversity and inclusiveness in rural development, but it may stifle grassroots innovation that truly has local characteristics and social value.

Furthermore, as mentioned above, the potential discrimination of AI systems against rural vulnerable groups (such as the elderly, indigenous people, and economically disadvantaged households) may cause them to encounter hidden obstacles in accessing social welfare, medical resources, employment opportunities, or financial services. For example, if an AI system used to match long-term care resources in rural areas underestimates the actual needs of elderly people living alone in certain areas with inaccessible transportation due to data bias, or fails to fully consider the special family structure and care culture of indigenous tribes due to blind spots in the model design, it may lead to a misallocation of long-term care resources, causing the elderly who are most in need of help to be “forgotten.” This difference in resource allocation based on algorithms directly violates the social fairness principles of “taking care of the disadvantaged” and “leaving no one behind” emphasized in the rural revitalization policy, and may form a “fairness paradox in the AI era” - that is, technology in the name of pursuing efficiency and objectivity has instead created new injustices.

(2) The algorithmic power gap further marginalizes disadvantaged rural communities

The algorithmic power gap, that is, the insufficient voice and control of rural communities (especially disadvantaged groups) in the design, deployment, operation and supervision of AI systems, may further marginalize them in the process of rural revitalization and widen existing social gaps.

When the introduction of AI systems is mainly led by external technical elites or higher-level governments, and rural residents lack effective participation channels and capabilities, their needs and voices are easily ignored. For example, in the process of promoting “smart tourism”, if the establishment of AI systems (such as attraction recommendation, tourist behavior analysis) is only from the perspective of improving tourism output value, and fails to fully consult local residents’ opinions on tourism development models, environmental carrying capacity, and cultural impact, it may lead to uneven distribution of tourism benefits, and even trigger conflicts between local residents and tourism development, deteriorating the good intentions of rural revitalization.

What is even more worrying is that technical thresholds and information asymmetry may put vulnerable groups in rural areas in a position of being “powerless to resist” when faced with AI decision-making. They may not understand how AI can make decisions that go against them and lack effective avenues for appeal or redress. For example, if an AI system demarcates the traditional territory of an indigenous tribe as an area unsuitable for the development of a specific industry, but its decision-making basis is contrary to the tribe’s traditional knowledge and wishes, and if tribal members lack sufficient technological literacy and legal resources to challenge this decision, their development rights and interests may be infringed. In this case, the AI system not only failed to empower the countryside, but instead became a new dominant force, making disadvantaged communities even more speechless in the rural development agenda, and making their unique culture and needs more difficult to see and respect. The solidification of this power structure will reduce rural revitalization to a game for a few elites, and will not truly benefit all rural residents, especially those disadvantaged groups that need help most.

In short, algorithmic bias and power gap are like two sides of the coin of rural AI applications, jointly posing severe challenges to the goals of fairness, inclusiveness and local subjectivity in rural revitalization. If we fail to face up to and actively respond to these potential impacts, the introduction of AI technology may drift further away from the original intention of rural revitalization.

4. Conclusion and Policy Implications

(1) Summary of main arguments

Through a review and theoretical analysis of relevant core literature, this study aims to explore the algorithmic bias and power gap dilemmas that artificial intelligence (AI) may cause in resource allocation and public service decision-making in the context of rural revitalization. Based on the previous arguments, this study puts forward the following main arguments: The potential and risks of applying AI to rural revitalization coexist. Algorithmic bias and power gap are potential dilemmas that must be seriously faced. Although AI technology is highly expected to bring opportunities for efficiency improvement and resource optimization for rural development, its actual application in rural areas will inevitably encounter the risk of bias posed by data poverty and distortion, contextual blind spots in model design, and implicit discrimination against disadvantaged rural groups. At the same time, factors such as technical thresholds, insufficient participation mechanisms, and external technological dominance may also form new asymmetries in the operation of algorithmic power within villages. If these biases and power gaps are not effectively identified and managed, AI will not only be unable to achieve its original intention of promoting rural revitalization, but may instead become a new driver of exacerbating existing inequality and eroding social fairness.

These dilemmas are not purely technical defects, but complex problems deeply rooted in social structure, power relations and governance practices. The root cause of algorithmic bias is not just the “dirtyness” of the data or the “imperfection” of the model, but more deeply reflects what Zajko calls the “real basis of inequality” in society. The long-standing urban-rural gap, low digital life, uneven distribution of resources, and the disadvantaged position of specific groups in the social structure in rural areas may all be copied, continued, or even amplified unconsciously by AI systems. Similarly, the operation of algorithmic power is closely related to the existing local political and economic structure, the transparency and accountability of the governance model, and the participation ability of civil society. As Sampath observes in the Global South, without a sound governance framework and full respect for local contexts, the introduction of technology may actually reinforce existing power imbalances. Therefore, it will be futile to try to “repair” prejudice or “balance” power from a technical perspective only, while ignoring the underlying social structural factors and governance mechanisms.

(2) Policy Enlightenment

In order to deal with the algorithmic bias and power gap that may be caused by AI in rural applications, and to ensure that technological development can truly serve the goals of equity and inclusion in rural revitalization, this study proposes the following policy implications based on the analysis of core literature and considering the special context of rural Taiwan:

Data governance and local participation: Constructing a people-centered rural data ecosystem Data is the cornerstone of AI, but the poverty and bias of rural data are the source of AI bias. Therefore, the first priority is to establish a rural data governance framework that is ethical and emphasizes local participation. This includes:

Improve data representativeness and quality: Resources should be invested to improve digital infrastructure in rural areas, and more inclusive data collection methods (such as combining oral interviews, community field surveys, etc.) should be designed for digitally vulnerable groups such as the elderly, indigenous people, and new residents to ensure that their experiences and needs can be included in the data set. For example, when building an agricultural AI system, in addition to standardized production and sales data, “atypical” data such as small farmers’ local farming knowledge and traditional crop varieties should also be included.
Emphasis on data sovereignty and local community participation: We should learn from international concepts such as “Indigenous Data Sovereignty” and explore the establishment of a mechanism for rural communities (such as tribes, rural and fishing village community development associations) to participate in data collection, management, interpretation and application. For example, promote the establishment of “community data cooperatives” or “local data trust” models, so that rural residents can have greater informed consent and control over data involving themselves or their communities, and prevent data from being unilaterally seized or improperly used by external agencies.
Develop ethical guidelines for rural data: Based on the particularity of rural AI applications, clear ethical guidelines for data collection and use should be formulated to clarify data ownership, privacy protection, and bias risk assessment specifications to ensure that data applications are premised on promoting rural well-being.

The pursuit of algorithm transparency, accountability and explainability: opening the “black box” of AI decision-making The opacity of algorithms is a breeding ground for power gaps and biases. When the public sector introduces AI systems in rural areas, it must strive to promote transparency, accountability and explainability:

Promote the transparency of public sector AI systems: The government should establish a login and disclosure mechanism for AI systems. For AI systems used in rural resource allocation or public services, they should proactively disclose their design purpose, main functions, training data sources (without infringing privacy), and known limitations and potential risks.
Establish bias impact assessment and appeal mechanism: Before the deployment of AI systems and during operation, bias impact assessments should be conducted regularly, with special attention to their potential differential impact on rural vulnerable groups. At the same time, a convenient and effective complaint and relief channel should be established so that rural residents can receive immediate response and processing when they believe they have been unfairly treated by AI decision-making.
Exploring local interpretability methods: Given the difficulty that rural residents (especially the elderly) have in understanding complex technical explanations, communication methods that go beyond purely technical explanations should be explored. For example, visual tools, local languages, case descriptions, or translation through local opinion leaders (such as village chiefs and tribal leaders) can be used to make AI’s decision-making logic easier to understand and trust among rural communities.

Empowerment and critical thinking cultivation: improving AI literacy in rural areas Facing the advent of the AI era, it is crucial to improve the understanding and critical thinking of rural residents and grassroots workers about AI. This is the basis for narrowing the gap in algorithm power:

Promote AI literacy education in rural areas: Combining existing resources such as community colleges, villagers’ gatherings, and farmers’ and fishermen’s association promotion systems, we can set up easy-to-understand AI science courses and workshops for rural residents of different ages and backgrounds, so that they can understand the basic principles, common applications, potential risks, and their own rights of AI.
Strengthen the AI awareness of grassroots civil servants: Provide AI-related professional training to grassroots civil servants serving in rural areas, so that they can not only operate AI systems, but also understand the limitations of the system, identify potential biases, and play the role of “translator” and “gatekeeper” between AI decision-making and actual local needs to avoid blind obedience or abuse of AI.
Cultivation of critical participation skills: Encourage rural residents not only to be passive AI users, but also to become participants with critical thinking, who can actively explore problems in the local application of AI systems and actively make suggestions for improvement.

Encourage the development of context-sensitive and demand-oriented AI: let AI truly serve rural areas

Avoiding “one-size-fits-all” technical solutions and encouraging the development of AI applications that can truly respond to the local needs of rural areas is the key to eliminating blind spots in the model context and leveraging the positive benefits of AI.
Support bottom-up AI innovation: Resources and platforms should be provided to encourage rural communities, local organizations, local community colleges or returning youth to proactively propose and participate in the development of AI solutions with local characteristics and context sensitivity in response to practical problems faced by rural areas (such as elderly care, characteristic agricultural development, cultural inheritance, environmental monitoring, etc.).
Emphasis on participatory and collaborative design of AI design: In the early stage of the design of rural AI systems, the participation of multiple stakeholders such as rural residents, grassroots workers, and local experts should be included, and through workshops, focus discussions, etc., their needs, experiences, and values can be fully integrated into the system design to avoid the disconnect between technology and needs.
Establish a pilot and evaluation mechanism for rural AI applications: For new rural AI applications, small-scale pilots should be conducted first, and an effectiveness evaluation and ethical review mechanism involving the participation of local communities should be established. Continuous revision and optimization should be based on actual feedback to ensure that AI applications meet the real needs and long-term well-being of rural areas.

The importance of sociology and cross-disciplinary cooperation: integrating multiple perspectives to jointly govern AI The complexity of AI governance, as well as its deep-rooted nature in social structures, make it difficult for a single discipline or department to deal with it alone. Zajko emphasizes the important role of sociology in understanding and responding to bias and inequality in AI, a perspective that is particularly instructive for rural AI governance:

Strengthen the role of social sciences in AI governance: In the formulation of rural AI policies, systematic design and development, impact assessment, and ethical review, experts and scholars in social science fields such as sociology, anthropology, geography, and law should be actively included so that they can provide in-depth analysis and suggestions from multiple perspectives such as social structure, power relations, cultural context, and ethical norms.
Promote cross-field dialogue and cooperation between government, academia, industry and civil society: A regular cross-field communication platform should be established to allow AI technology developers, policy makers, rural practitioners, local residents’ representatives and academic researchers to have full dialogue, consultation and cooperation on the vision, challenges and paths of rural AI development, and jointly shape an AI governance framework that is responsible, inclusive and conducive to sustainable rural development.

5. Research Prospects

First, there is a call to deepen on-the-ground empirical research: from theoretical discussion to field cultivation.

The existing literature mostly discusses AI bias and governance challenges based on general principles or specific foreign cases. However, rural society in Taiwan has its own unique history, culture, industrial structure, and social network characteristics (for example, the traditional domains and governance models of aboriginal tribes, the industrial culture of Hakka settlements, the population structure and livelihood methods of western coastal agricultural and fishing villages, and the significant digital activities between different regions, etc.). How these local contexts interact with the introduction of AI technology to shape the specific forms of algorithmic bias, the transformation of power relations, and the actual impact on rural revitalization remains to be answered by in-depth local empirical research.

Therefore, future research should actively invest in rural fields in Taiwan and use qualitative research methods (such as in-depth interviews, focus groups, participant observation) and quantitative research methods (such as questionnaires, data analysis) to combine multi-level discussions.

For example, we can conduct in-depth case studies on AI application cases that have been implemented in rural areas (such as smart agricultural guidance systems, remote long-term care resource matching platforms, local revitalization subsidy evaluation tools, etc.), and interview system developers, policy makers, grassroots executives, and affected rural residents to understand their understanding of the AI system, usage experience, perceived benefits and difficulties, as well as their agency and negotiation process. Secondly, the bias impact assessment and power interaction analysis focus on specific cases. In terms of specific empirical research directions, in the future, more detailed bias impact assessment and power interaction analysis can be conducted for specific rural AI application cases:

Bias impact assessment: You can learn from Curto et al.’s analysis method of “poverty phobia”, but further combine it with local data and situations to examine whether there are systemic biases in specific AI systems against rural elderly, indigenous people, new residents, low-income households, or digitally disadvantaged groups. For example, analyze whether the decision-making results of an AI model used for agricultural loan approval have a disproportionate negative impact among farmers with different socio-economic backgrounds or farming scales.
Power interaction analysis: You can use Zajko’s sociological perspective and Veale et al.’s insights into public sector AI practices to conduct an in-depth analysis of how power relationships between different actors (such as the central government, local governments, technology providers, local elites, ordinary residents, representatives of disadvantaged groups, etc.) are performed, negotiated, and reconstructed during the design, deployment, and operation of rural AI systems. For example, it explores how the voices of local communities are heard or silenced in the cracks between the technocratic system and commercial interests in smart rural policies promoted “from above.”

Through the above-mentioned empirical research, we can not only gain a deeper understanding of the real impact of AI in the rural context of Taiwan, but also provide a solid academic foundation and practical reference for formulating more targeted and effective algorithmic governance policies and developing AI applications that better meet local needs, thus ensuring that AI technology can truly become a positive force in promoting inclusive, equitable and sustainable development in rural Taiwan.

References

Curto, G., Jojoa Acosta, M. F., Comim, F., & Garcia-Zapirain, B. (2024). Are AI systems biased against the poor? A machine learning analysis using Word2Vec and GloVe embeddings. AI & Society, 39(2), 617-632.
Gehl Sampath, P. (2021). Governing artificial intelligence in an age of inequality. Global Policy, 12, 21-31.
Zajko, M. (2022). Artificial intelligence, algorithms, and social inequality: Sociological contributions to contemporary debates. Sociology Compass, 16(3), e12962.
Kukutai, T., & Taylor, J. (2016). Indigenous data sovereignty: Toward an agenda. ANU Press.
Veale, M., Van Kleek, M., & Binns, R. (2018). Fairness and accountability design needs for algorithmic support in high-stakes public sector decision-making. In Proceedings of the 2018 CHI conference on human factors in computing systems (pp. 1-14).

[Author information] Dylan Chiang, National Institute of Development and Mainland China, Chinese Culture University, Taiwan, email:

Social Justice | Dylan Chiang