Author: Immanuel Putra SiregarEditor: Hosea Immanuel Latumahina
The trend of Generative AI (GenAI) adoption in Indonesia’s public and private sectors is currently being pushed massively in the name of efficiency.1–5 As an illustration, institutions (both public and private) are now deploying GenAI-based chatbots to respond to thousands of user complaints simultaneously.1–3,5 However, this efficiency ambition creates an anomaly on the ground, instead of getting easier services, people are transforming into sudden auditors who are required to verify the authenticity of photos, the truth of chatbot information, and the accuracy of problem-solving.6,1,7 Reflecting on this paradox, if GenAI was created to make life easier, why are people now getting more exhausted because they have to constantly act as “free auditors” for government and corporate systems?7–9 This phenomenon urgently needs to be discussed because it uncovers the illusion of GenAI efficiency, where this practice quietly shifts the burden of proof onto the shoulders of ordinary citizens.10–12,1 If left unchecked, this will trigger mass exhaustion (cognitive verification load) that leads to the collapse of public trust in the entire digital ecosystem.10,7,13,14,5 Therefore, this article argues that the widespread use of GenAI, which actually produces synthetic content waste and hallucinated information (AI slop) without guardrails is not just a technological miscommunication problem,7 but an institutional political-economy strategy to outsource accountability costs to citizens through unpaid work practices (invisible verification labor).7,8
The Political Economy of AI Slop: Efficiency at Whose Expense?
The massive integration of GenAI in today’s public and private service architectures cannot be separated from the logic of cutting operational costs (cost-efficiency).1,2 For state institutions and digital platforms, GenAI offers the promise of large-scale efficiency due to the sharp drop in the cost of creating automated replies (cost collapse),13 which allows them to process thousands of citizen complaints or consumer questions simultaneously in a fraction of a second with minimal human resource costs (cost burden).1 However, this efficiency narrative often hides the political-economy reality behind it, where the cost burden does not truly disappear from the system, but is externalized outside the institution’s ecosystem.7,10,12
Different from the general assumption that institutions are simply sacrificing quality control, what actually happens is a structural shift of the human verification cost burden (labour transfer).7 Several institutions have actually realized the crucial role of human judgement as a filter for GenAI output, as mandated in the nine ethical principles of Surat Edaran Menkominfo (SE Menkominfo No. 9/2023) and GenAI ethical guidelines.15,5,16 However, because the current national regulatory framework is still a soft law that is not equipped with incentive schemes or binding audit mechanisms at the operational level, this burden falls instead on the shoulders of the end users.12 As a concrete example, when a citizen reports a damaged public facility or complains about a product to customer service, they now no longer just wait for a solution from a human operator. They have to independently dissect, guess, and verify whether the repair progress photo uploaded by the officer or the regulatory answer from the bot is a real fact or just a machine hallucination.1
Bearing the Cognitive Verification Load in the Era of “Easy to Generate, Hard to Audit”
This structural asymmetry, where institutions reap efficiency while citizens bear the workload of verification, gives rise to a phenomenon that drains the working memory of netizens, known as the cognitive verification load.7,10 In an era where creating synthetic content using GenAI has become instantaneous and virtually cost-free for institutions (easy to generate),13 the process of auditing the truth of that information becomes highly complex and time-consuming for human users (hard to audit).11,12,14 This difficulty is rooted in its computational limitations: GenAI models lack the ability to independently evaluate their own factual calibration without ground truth labels (self-verification impossibility).11 As a result, every time citizens are forced to sort through machine hallucinations and official information, they are essentially being demanded to pay an AI oversight tax, namely, the extra cognitive work to verify, restore context, and correct system errors.12 The accumulation of this hidden verification burden not only triggers decision fatigue, but ultimately threatens the cognitive sustainability of society at large.7
This situation becomes even more complicated when the concept of human-in-the-loop is misinterpreted in practice. Instead of obliging institutions to put human operators on standby as final verifiers (gatekeepers) before GenAI output is thrown to the users,12,7,17 internal supervision is often degraded into mere formality.12 That responsibility is instead burdened onto users through the narrative of “improving AI literacy”.1 Demanding the public to remain vigilant and technologically literate to detect AI slop is a form of institutional wash-hands. The necessity to constantly verify this flood of automated information will ultimately only push the public towards skepticism and absolute exhaustion.18,14,13 Thus, technological literacy should be a tool for citizen empowerment, not an excuse for institutions to outsource the quality audit burden of their services.7,8,5
Digital Trust Erosion & Recommendations
As the cognitive verification load continues to accumulate, the most crucial social capital in modern society will also erode, namely digital trust.10,7,13 In the private sector, allowing AI slop clearly harms consumers’ rights to receive accurate information and proper services as mandated in Article 4 and Article 18 of the Undang-Undang Perlindungan Konsumen (UUPK).1,7,19 Meanwhile, in the public sector, outsourcing this verification betrays the social contract. Citizens have the right to accountable state services, rather than being forced to guess whether their government is hiding behind algorithmic hallucinations.5,18 If left unchecked, this asymmetry will create an ecosystem where the public is skeptical by default and rejects all digitalization initiatives, no matter how good the purpose is, because they feel their basic rights are constantly compromised.13,14,5
Therefore, the government must take structural intervention steps. The momentum of the drafting of the Presidential Regulation (Peraturan Presiden or Perpres) on AI Ethics must be utilized to firmly establish this accountability boundary. The regulation must explicitly mandate the implementation of human-in-the-loop as a compliance cost for institutions (both public and private), rather than a mere recommendation.7,12,17 The state and corporations must be held legally accountable for the flawed “hallucinations” of their GenAI, and be obligated to provide transparent non-GenAI complaint channels. Restoring AI literacy as a means of empowerment means ensuring that technology serves the citizens, rather than simply shifting the workload onto them.
Referensi