Clinical AI works in Indonesian hospitals but flounders outside them: promising diagnostic and telemedicine results are undermined by patchy broadband, fragmented regulation and limited workforce readiness, risking foreign vendor dependence and lost economic gains unless coordinated policy and investment follow.

Main Finding

Indonesia has proven clinical successes with AI in healthcare (notably diagnostics, telemedicine, and chronic disease management) but overall readiness is moderate (AI healthcare maturity score 52/100). Major bottlenecks — fragmented regulation, limited governance (transparency/explainability), unequal infrastructure and digital access, and weak workforce AI literacy — threaten equitable scale-up and risk technological dependency unless coordinated policy, investment, and capacity-building are rapidly implemented to meet Golden Indonesia 2045 goals.

Key Points

• Performance highlights
  • Diabetic retinopathy screening: accuracy 89.3% (95% CI 86.7–92.1%), sensitivity 91.7%; high patient (87.5%) and physician (90%) acceptance.
  • Cervical cancer VIA-based AI: sensitivity 80%, specificity 96.4%, AUC 0.85.
  • Pediatric dentistry AI: 90–97% accuracy; high usability.
  • Telemedicine for chronic disease: medication adherence increased from 6.8 to 8.9 (p < 0.001) over 3 months.
  • Mental-health chatbots reduce symptoms in some studies but are limited to screening/triage roles.
• Readiness and gaps
  • National maturity score: 52/100 vs Singapore 92, Malaysia 78.
  • Regulatory landscape: Law No.17/2023, PP No.28/2024, and UU PDP (2022) provide partial frameworks but lack AI-specific provisions on liability, audit standards, consent, transparency, and algorithmic fairness.
  • Governance: Delphi consensus produced 24 governance indicators, but transparency items (model cards, training-data summaries) lacked expert consensus.
  • Infrastructure disparities: internet coverage urban 82% vs rural 35%; medical equipment urban 78% vs rural 28%; data systems urban 75% vs rural 22%.
  • Workforce competence: ~58.7% of healthcare workers lack sufficient AI competence; 82.5% have basic computer skills but <50% can interpret AI outputs; in 3T regions digital readiness ~15% vs 72% in urban centers.
  • Adoption barriers: professional resistance (autonomy, job fears), liability uncertainty, limited change-management capacity, and scarce implementation science evidence.
• Equity risks
  • Without targeted policies, AI adoption may widen urban–rural and socioeconomic health inequalities (first-, second-, and tertiary-order digital divides).
• Strategic recommendations (high level)
  • Enact AI-specific regulatory clarity (liability, auditing, consent, fairness).
  • Strengthen SATUSEHAT integration and governance indicators including transparency and explainability standards.
  • Invest in broadband, power, equipment, and data systems with rural prioritization.
  • Scale targeted workforce training (technical operation + critical appraisal of AI outputs) and implementation science.
  • Foster domestic AI capability to avoid dependency on foreign tech and better match local epidemiology.

Data & Methods

• Study type: Analytical narrative review with thematic synthesis.
• Search strategy: Systematic searches (PubMed, Google Scholar, Garuda, SINTA) for 2020–2025 using terms related to AI, healthcare, Indonesia, SATUSEHAT, telemedicine, digital health.
• Sources: Peer-reviewed articles, conference proceedings, grey literature, national policy documents (Law No.17/2023, PP No.28/2024, UU PDP 2022), SATUSEHAT reports, and Delphi governance studies.
• Screening/results: 68 records screened -> 42 selected for in-depth review.
• Comparative benchmarking: Regional/international frameworks (EU AI Act, Singapore, Australia) and maturity catalogues.
• Thematic domains: clinical effectiveness, regulation, infrastructure, human resources, equity, and alignment with RPJPN 2025–2045.

Implications for AI Economics

• Investment needs and returns
  • Significant public and private capital required for digital infrastructure (broadband, power, equipment), data systems, and workforce upskilling — especially to close urban–rural gaps. These are preconditions to realizing AI-driven productivity and health gains.
  • Targeted, context-appropriate AI (screening, telemedicine, chronic care support) shows cost-effectiveness potential by decentralizing specialist services and improving adherence; these could reduce downstream treatment costs and increase health-system efficiency if scaled equitably.
• Market development and innovation strategy
  • Regulatory uncertainty (liability, auditing, transparency) depresses investor confidence and slows market formation. Clear, risk-based regulation will lower transaction costs and attract both domestic and foreign investment.
  • Nurturing domestic AI capacity (data, models, companies) reduces dependency risks, captures local value chains, and better aligns products with Indonesian epidemiology and constraints — improving social returns compared to importing turnkey solutions.
• Human capital and labor economics
  • Up-skilling healthcare workers creates complementary capital to AI, increasing labor productivity and enabling higher-value clinical roles; conversely, failure to invest risks deskilling or displacement and political resistance.
  • Training programs should combine technical AI operation with decision-making, uncertainty calibration, and medico-legal literacy to maximize complementary productivity effects.
• Equity and welfare economics
  • Without pro-poor deployment, AI can exacerbate welfare inequalities: urban populations and private facilities will likely capture most early benefits. Public financing and targeted subsidies are needed to realize equitable welfare gains.
  • Prioritizing high-impact, low-cost interventions (e.g., AI-enabled screening in remote areas) can deliver large marginal social returns and reduce inequality in health outcomes.
• Governance, externalities, and market failures
  • Information asymmetries (black-box models) and negative externalities (algorithmic bias, failures) justify regulation, standards, and transparent audit mechanisms to correct market failures and sustain trust — essential for demand-side adoption.
  • Public goods investments (data infrastructure, shared annotated datasets, model validation labs) can lower entry barriers for startups and reduce duplication.
• Policy instruments and financing models
  • Hybrid financing: public investment in infrastructure and data governance + incentives (grants, prizes, procurement) to stimulate domestic AI solutions targeted at underserved regions.
  • Public–private partnerships and conditional procurement (requiring explainability, local-data training, capacity-building) can align private incentives with public objectives.
• Research and evaluation economics
  • Implementation science and real-world cost-effectiveness studies are needed to assess scalability, inform resource allocation, and guide prioritization of interventions with highest social return.

Overall, from an AI-economics perspective, Indonesia’s clinical AI successes present high social-return opportunities, but realizing those returns at scale requires upfront public investments, regulatory certainty, human capital formation, and equity-focused deployment policies to correct market failures and avoid widening health and economic disparities.