The Invisibility Hypothesis: Promises of AGI and the Future of the Global South

AI summary

Overview

Research area: AI safety and ethics, with a focus on the political economy of artificial general intelligence (AGI) and global inequality — categorized on arXiv as cs.CY (computers and society).

Technical level: Intermediate. The paper is conceptual rather than technical — it contains no models, code, or quantitative experiments — but it assumes familiarity with debates about AGI capability, AI deployment economics, and development/political-economy concepts such as informality and machine-legibility.

Scope (one sentence): The paper argues that as increasingly autonomous AI systems become the coordination layer for economic and political allocation, populations in the Global South — especially informal workers and small-scale producers — risk becoming "economically invisible" through managed exclusion rather than direct job replacement, and it sketches three possible futures for Latin America, Africa, and South Asia under such a regime.

What This Paper Is About

Discussions of AGI have concentrated on technical feasibility, timelines, and existential risk, generally assuming that its social impact will be uniform across populations. This paper instead asks what happens to populations that are poorly represented in data, weakly integrated into formal economies, and distant from global centers of technological power. Its goal is to characterize a specific failure mode — the systematic favoring of "machine-legible" individuals by AI-mediated allocation systems — and to lay out the plausible trajectories this produces for the Global South.

Key Contributions

1. Introduces the Invisibility Hypothesis, a named account of how AI-controlled economic and political management systematically favors those whose identities, transactions, and outputs can be verified, standardized, and audited.
2. Provides an operational distinction between "economically invisible" and "machine-legible" individuals, defining invisibility not metaphorically but as the absence of actionable representation within AI-governed pipelines for credit, procurement, insurance, logistics, and policy priorities.
3. Outlines three plausible futures for the Global South under an AGI regime — the Utopia (location stops mattering), the Collapse (from exploitation to irrelevance), and the Middle Ground (present inequality intensified).
4. Reframes the central risk from exploitation to exclusion, arguing that where historical inequality required human labor, an AGI regime may no longer depend on the labor, markets, or political cooperation of large populations at all.

Main Findings

• Exclusion can precede automation. The distinctive harm the paper identifies is that exclusion via AI-controlled geoeconomics can occur before full-scale automation, quietly shrinking participation and relevance even when the underlying work remains economically or socially valuable. The short-term risk for "invisible individuals" is not replacement but loss of access and relevance within credit systems, supply chains, insurance markets, and policy processes.

• Exclusion is self-reinforcing. Once denied credit, contracts, insurance coverage, or program eligibility, individuals produce fewer verifiable records and standardized signals, which further lowers their visibility and makes reentry progressively harder.

• Displacement is asymmetric across job types. Routine office work in highly-digitized environments may be displaced early, because it is already standardized into the interface that the AI understands. Many physical jobs may only be displaced if reliable, low-cost robotics actually matures.

• Capability without omnipotence. Even highly capable general-purpose systems remain constrained by physical resources, infrastructure, governments, and geography. Intelligence alone does not remove bottlenecks in energy availability, compute access, data quality, institutional trust, or political authority, and additional intelligence yields diminishing returns once coordination, infrastructure, and enforcement become the dominant constraints. A system may design optimal agricultural policies or supply chains yet remain ineffective if surrounding institutions cannot implement, regulate, or absorb the recommendations.

• The likely path is the intensified present. Of the three scenarios, the "Middle Ground" is presented as most plausible: AGI amplifies productivity and wealth for already advantaged individuals and companies while underlying social and economic issues remain largely intact. Informal economies persist but become increasingly disconnected from AI-mediated systems. Inequality deepens without a clear rupture, making this trajectory both stable and difficult to change.

• Marginalization is not uniform at the nation-state level. Exclusion increasingly operates at the level of individuals and social classes. Privileged elites within the Global South — those with access to capital, infrastructure, and transnational networks — are likely to remain integrated into global AI-driven systems, even as local inequality deepens.

• Structural asymmetries are illustrated, not measured. Figure 1 combines GDP per capita, a political corruption perception index, and global population distribution (sources: Our World in Data — Roser et al. (2023); Herre et al. (2013); United Nations (2022)) to show that the majority of the world's population resides in regions with comparatively low economic output and weaker institutional capacity. No quantitative results, benchmark numbers, or dataset sizes are reported in the paper.

Methodology in Plain English

This is a conceptual and analytical paper, not an experimental one. The authors reason from a hypothetical premise — that AGI may eventually be achieved — while explicitly adopting a cautious, agnostic stance on claims of its imminent arrival. They build their argument by drawing on empirical signals from contemporary AI deployment and extending those signals into potential trajectories, grounding the discussion in cited literature on AI capability, development economics, and postcolonial political economy. They define their key terms operationally (invisibility, machine-legibility) rather than metaphorically, then use those definitions to reason about how AI-mediated allocation would treat different populations. The three futures are presented as a scenario analysis spanning best case, worst case, and most likely case. The paper reports no new experiments, models, or measurements; its "evidence" is illustrative figures drawn from existing public data sources.

Why This Matters

Impact on research. The paper pushes AGI scholarship away from treating social impact as uniform and toward a distributional analysis that distinguishes between machine-legible and invisible populations. It offers a testable-sounding framing — that AI-mediated allocation optimizes for what is measurable and verifiable, not what is socially valuable — that other researchers could try to operationalize or falsify. It also foregrounds a question rarely asked directly: under what conditions are societies and individuals considered economically and politically important at all?

Real-world applications:

• Credit and development finance: designing lending and procurement pipelines that do not systematically exclude informal workers and small-scale producers who lack standardized, traceable records.
• Public service delivery and eligibility: auditing AI-mediated program eligibility and policy prioritization for populations that are weakly represented in data.
• Insurance, logistics, and supply chains: assessing who gets onboarded into AI-coordinated systems and who is quietly left outside them.
• AI governance and infrastructure policy: deciding how compute, models, and AI-mediated institutions are distributed across regions with weak institutional capacity.
• Robotics and labor planning: anticipating that physical work may be displaced later than digital work, contingent on the maturity of reliable low-cost robotics.

Industry relevance. Companies deploying AI systems in emerging markets face direct questions about whose data, transactions, and outputs their models can observe and score. Firms whose products allocate credit, insurance, logistics capacity, or access to markets effectively determine who is machine-legible and who is not — making this a commercial design question as much as an ethical one.

Future Directions

1. Operationalizing and measuring invisibility. The paper defines the concept but does not propose an empirical metric. Developing indicators — for example, rates of rejection or non-observation in AI-mediated credit, insurance, and procurement — is a clear next step.
2. Testing the reversibility of the exclusion loop. The authors describe exclusion as self-reinforcing and make reentry increasingly difficult, but leave open what interventions, identity infrastructures, or institutional designs could break that loop.
3. Timing and the robotics question. Since physical-job displacement is made contingent on reliable, low-cost robotics maturing, the relative timing of digital exclusion versus physical automation remains an open empirical question.
4. Comparative analysis within the Global South. The paper notes that privileged elites may remain integrated while others are excluded; understanding how this sub-national and class-level divergence plays out across Latin America, Africa, and South Asia is unresolved.

Target Audience

This paper is most useful for AI policy researchers and governance scholars working on global inequality and distributional impact; development economists and practitioners working on informality, credit access, and digital public infrastructure; and technology companies and regulators designing AI-mediated allocation systems for markets in the Global South. It is also accessible to readers with a general interest in AI ethics who want a structured framework for thinking about who benefits and who is bypassed as AI systems become more capable — no technical AI background is required to follow the argument.

Authors’ abstract

Discussions surrounding Artificial General Intelligence have largely focused on technical feasibility, timelines, and existential risk, often treating its social impact as being the same across different populations. Less attention has been paid to how advanced AI systems may interact with existing global inequalities. This paper examines the implications of AGI for people in the Global South, arguing that the availability of highly autonomous, general-purpose cognitive systems does not guarantee equitable outcomes. We establish that, as scientific discovery, economic coordination, and governance become increasingly automated, the relevance of human individuals may become conditional on access to infrastructure, institutional inclusion, and geopolitical circumstances rather than skills or intelligence. Under this setting, the Global South faces different pathways: in the best case, geographic location is no longer relevant as AGI fully democratizes access to knowledge and essential services for everyone in the globe; in the worst case, existing structural constraints are severely amplified, rendering already marginalized populations not merely economically invisible, but functionally irrelevant to global systems. We ground this analysis in empirical signals from contemporary AI deployment and extend to potential trajectories, highlighting both risk and opportunity pathways for Latin America, Africa, and South Asia.