Artificial intelligence often appears seamless on the surface, capable of understanding language, recognizing objects, and even driving vehicles. But behind this apparent intelligence lies a deeply human infrastructure that is rarely acknowledged.

by Dr. Balamurugan Balusamy, Chairperson and Professor at the School of Engineering and IT, at Manipal Academy of Higher Education Dubai

[For more news, click here]

Bridging the Gap: From Physical Reality to Structured Data

AI does not emerge in isolation. It is built on vast amounts of data collected, labeled, filtered, and refined by people, many of whom remain invisible in the final system.

Most physical-world data used in AI systems is not naturally available in structured form. It must be captured from real environments, often across diverse and complex conditions. This becomes especially critical in domains like robotics, autonomous systems, and digital twins, where machines are trained to understand human-like behavior and real-world interactions. To achieve this, AI systems require enormous datasets that reflect how humans move, respond, and interact with their environment.

The Rise of the "Ghost Workforce"

A significant portion of this work is carried out by gig workers, often referred to as “ghost workers.” These individuals perform essential but underrecognized tasks such as data annotation, image labeling, content moderation, and supervised training. In some cases, they even wear sensor-equipped devices to capture human motion and behavior for training advanced AI models. Their work quietly powers systems that billions of people interact with daily, yet their presence is almost entirely absent from public narratives about AI progress.

Accelerating Innovation in Autonomous Systems

This hidden labor becomes even more evident in industries like electric vehicles (EVs) and unmanned aerial vehicles (UAVs). As companies race to develop autonomous systems, they face strict regulatory and certification requirements. These systems must be trained on massive volumes of real-world data before they can safely operate. However, collecting such data organically can take years. To accelerate this process, organizations often rely on distributed gig workforces to generate and label data at scale, filling the gap between technological ambition and real-world limitations.

Ethical Complexity in the Era of LLMs

At the same time, large language models (LLMs) have introduced a new layer of ethical complexity. Questions around data ownership, consent, and copyright remain unresolved. Can books, articles, and creative works produced through years of intellectual effort be used to train AI systems without explicit permission? Who owns the resulting model knowledge: the original creator, the data collector, or the organization that trained the system? And how much transparency should exist in the way such data is sourced?

These questions point to a larger issue: the AI ecosystem lacks universally accepted ethical boundaries for data usage. And while these debates continue at the policy level, the human cost of data creation is already being paid in practice.

Many gig workers operate under difficult conditions. They are often paid low wages, work under strict deadlines, and are subject to constant task monitoring. Some face psychological strain, particularly those involved in content moderation where they are exposed to disturbing material. In many regions, especially

in parts of the Global South, labor protections are minimal, and transparency about how data is used is limited. Consent is often unclear, and recognition is almost nonexistent.

Despite the common perception that AI systems learn autonomously, human intervention is embedded at every stage, from raw data creation to fine-tuning model behavior. AI is not purely machine-driven intelligence; it is a layered system of human decisions, human labor, and human interpretation.

Demanding Accountability and Transparency

As users, researchers, and developers of AI systems, we must begin asking harder questions. Who designs ethical frameworks for data annotation? Who ensures that consent is meaningful and not just procedural? Who takes responsibility for recognizing the invisible workforce that powers these systems? And perhaps most importantly, who protects the mental health and dignity of those exposed to the most demanding parts of this pipeline?

Without addressing these questions, the gap between visible innovation and invisible labor will continue to widen. The progress of AI risks becoming disconnected from the people who make it possible.

What is needed now is not just better models, but better design philosophies, ones that embed transparency, fairness, and accountability into the foundation of AI systems. Ethical AI must go beyond technical performance. It must include fair labor practices, clear data governance, and genuine recognition of human contribution.

AI should not only learn from humans. It should also respect them.