By Dr Chew Han Ei, Senior Research Fellow, Head, Governance and Economy at the Institute of Policy Studies (IPS), Singapore
At a glance
AI adoption is accelerating across Southeast Asia, but uneven digital trust may create a two-speed AI economy.
Statistical modelling shows that institutional trust in digital systems is shaped by identifiable dimensions, including cybersecurity, transparency, reliability, fairness and redressability.
When trust capacity diverges across populations and firms, AI expansion risks entrenching inequality and limiting access to opportunities for households and enterprises.
Share this insight
Southeast Asia’s digital economy is projected to exceed US$300 billion in gross merchandise value in 2025, driven by continued growth in e-commerce, digital finance and platform-enabled services. Artificial intelligence is now positioned as the next catalyst for transformation, with national initiatives underway, governance frameworks taking shape, and enterprises embedding generative tools into everyday workflows.
The ambition is clear. The distribution of benefits is not.
Policy and business discourse often assumes that widespread diffusion of AI will automatically yield productivity gains across sectors and social groups. Yet those gains depend on a condition that is frequently overlooked: trust capacity.
AI adoption requires confidence that systems are secure, fair, reliable and correctable. Where such confidence is uneven, adoption will be uneven — and inequality risks becoming structural rather than incidental.
From digital trust to trust capacity
In 2023, I argued that digital trust functioned as a compass for inclusive digital development. The erosion of trust in online systems discouraged participation and weakened economic potential.
As the World Economic Forum defines it:
“Digital trust is individuals’ expectation that digital technologies and services — and the organisations providing them — will protect all stakeholders’ interests and uphold societal expectations and values.”
That diagnosis remains relevant. Despite efforts to curb harmful uses of AI, citizens continue to encounter AI-generated disinformation and synthetic manipulation that blur the boundary between online falsehood and offline harm.
As AI systems become embedded in credit assessment, supply chain management, healthcare diagnostics and public service delivery, participation depends not only on connectivity but also on whether individuals and firms perceive automated decisions as legitimate, fair, transparent and accountable.
Trust capacity captures this interaction between institutional architecture and user capability. It reflects the credibility of cybersecurity protections, the transparency of decision-making, the reliability of system performance and the availability of meaningful redress when failures occur. It also depends on users’ literacy — their ability to understand, evaluate and respond to digital systems.
When trust capacity diverges, adoption diverges.
Measuring trust: what the evidence shows
Trust capacity can be assessed empirically. Recent national-level research in Singapore examined institutional trust in digital systems using survey data and statistical modelling. Rather than treating trust as an abstract disposition, the analysis decomposed it into system-level dimensions.
Five dimensions were consistently associated with institutional trust:
- Cybersecurity — confidence that digital systems protect data against intrusion and misuse.
- Fairness — belief that systems treat users equitably and do not systematically disadvantage particular groups.
- Transparency — clarity about how data are processed and how automated decisions are made.
- Reliability — expectation that systems perform consistently and predictably.
- Redressability — the availability of meaningful channels for complaint, correction and remedy when digital systems fail.
Taken together, these dimensions accounted for nearly half of the observed variation in institutional trust across respondents. Trust is therefore structured and responsive to governance architecture and institutional design choices.
Redressability warrants particular attention in the AI era. Trust is sustained not only by performance, but also by recovery. When algorithms amplify harmful content — such as deepfake nudes or scam attempts — confidence erodes. Institutions that respond visibly and effectively to failure are more likely to restore trust than those that treat automated systems as beyond contestation.
The same research incorporated individual-level traits to examine how personal characteristics shape trust. Among these, digital literacy emerged as the strongest predictor of relational trust.
Individuals who reported greater confidence in navigating digital platforms, understanding data flows and interpreting online content were significantly more likely to express institutional trust, even after accounting for perceptions of cybersecurity, fairness and transparency.
This finding reframes the AI inclusion debate.
Mechanical trust concerns system robustness and institutional safeguards — areas that institutions can directly shape. Relational trust, by contrast, depends on how users interpret those safeguards, and digital literacy plays a decisive role in that process.
In an AI-enabled environment, where systems are increasingly opaque and data-driven, interpretive capability becomes pivotal. Users who understand system limitations, recognise bias and are aware of avenues for redress are better positioned to calibrate trust appropriately. Others may oscillate between misplaced confidence and categorical distrust.
Trust capacity emerges from the interaction between institutional design and the distribution of digital literacy.
The AI divide is already visible
These dynamics intersect with what scholars term the AI divide: unequal access to AI systems, unequal ability to use them and unequal outcomes from engagement.
In education, AI tutoring systems can personalise learning and accelerate skill acquisition. Yet learners whose profiles align poorly with training data may receive less accurate or culturally misaligned feedback, compounding existing gaps.
In labour markets, AI-driven transformation is expected to create and displace millions of jobs. Routine and clerical roles face contraction, while workers with AI-complementary skills are better positioned to benefit. These shifts disproportionately affect certain demographic groups, including women who are overrepresented in administrative occupations.
AI-enabled risk adds another layer. Sophisticated scams and synthetic media require no voluntary adoption. Individuals with lower AI literacy may be more vulnerable to deepfakes and long-con financial manipulation, reinforcing economic insecurity.
Where literacy is strong and safeguards are credible, participation deepens. Where literacy is weak and recourse mechanisms are opaque, disengagement or harm follows.
Regional ambition meets structural risk
Across Southeast Asia, policymakers are investing heavily in AI infrastructure and governance.
Singapore’s Budget 2026 introduced National AI Missions and reinforced enterprise and workforce schemes to expand AI capability. Malaysia’s proposed AI Governance Bill seeks to regulate the full lifecycle of AI systems, clarifying accountability between developers and deployers while addressing ethical and copyright concerns. The establishment of Malaysia’s National AI Office signals institutional commitment to coordinated oversight.
These initiatives recognise that AI governance must address system robustness, fairness and accountability in parallel.
The Tech for Good Institute’s work on building a Confident Digital Society situates trust, participation and digital resilience alongside economic growth as pillars of sustainable development. AI intensifies the need to treat trust capacity as economic infrastructure, not reputational management.
Regulation can mandate transparency and formalise redress. Yet safeguards must be legible. Without distributed digital literacy, even well-designed systems may fail to generate confidence, resulting in stratified adoption.
The Path Forward: What must change now
If trust capacity shapes participation in AI systems, then how progress is defined and governed must shift accordingly.
National dashboards tend to emphasise investment flows, enterprise adoption rates and sectoral deployment. These indicators reveal little about whether citizens trust automated decisions or understand how to navigate them. Public awareness of redress mechanisms, perceived fairness of algorithmic systems and digital literacy levels should be treated as core readiness metrics. They sit closer to the conditions that determine whether adoption deepens or stalls.
For industry, legal alignment can reduce regulatory exposure, but it does not automatically build confidence. Durable adoption depends on whether systems are intelligible to users — transparent about what they do, where their limits lie and how decisions can be reviewed. Explainability cannot remain a backend technical feature; it must form part of the user experience.
Safeguards introduced without corresponding literacy risk being misinterpreted or underutilised. Literacy initiatives launched without visible institutional accountability risk breeding scepticism. The two must move in tandem. Teaching citizens and small firms how to identify bias or trace data provenance is meaningful only if institutions are demonstrably responsive when failures occur. Trust capacity develops when institutional design and user capability reinforce one another.
Southeast Asia’s AI ambitions are clear across budgets, policy papers and boardrooms. What remains uncertain is whether confidence will keep pace with capability.
Digital trust is more than a communication issue; it is an institutional variable shaping who participates and who withdraws, often before headline adoption rates reveal gaps. If trust capacity is concentrated among digitally fluent firms and individuals, AI will reproduce existing asymmetries. Broader cultivation of trust capacity, by contrast, expands meaningful participation.
The region’s AI race is underway. The key question is not who moves first, but who moves with assurance.
The views and recommendations expressed in this article published are solely of the author and do not necessarily reflect the views and position of the Tech for Good Institute.
