The Yin-Yang Nature of Data: Policy Implications

Rahul De’ 

April 2, 2026 

Abstract 

This essay examines data governance through the lens of the ancient Chinese yin-yang philosophy, arguing that data possesses dual, complementary, and sometimes opposing characteristics that policy must balance. Rather than treating data through singular metaphors such as “oil,” “knowledge,” or “infrastructure,” it proposes yin-yang as a framework for understanding data’s multifaceted nature. The essay identifies five key dualities: property rights (open versus proprietary), data domains (human activity versus environmental phenomena), cognitive processing (symbolic versus sub-symbolic), sources (digital versus analog), and origins (authentic versus synthetic). It applies these dualities to data ownership, pre-processing ethics, provenance, purpose limitation, and retention. The analysis shows how existing regulations struggle with data’s dual nature and concludes that effective governance requires balancing opposing characteristics rather than imposing rigid, monolithic frameworks.

Introduction 

The Deep Learning movement in AI demonstrated that AI algorithms trained on large volumes of data generate significantly better models. The AlexNet system, for instance, in 2012, achieved an 11 per cent increase in image recognition performance over prior models, when trained on a massive data set. This immense success showed that deep learning systems and large data sets were key to success in building highly accurate AI systems. This trend has continued with large language models (LLMs) being built on massive text and image corpora. For instance, ChatGPT 3.5, the first widely launched model of OpenAI, was trained on a corpus of 300 billion words (filtered down from about 45TB of data), roughly equivalent to the entire content of the open internet. 

The success of deep learning triggered a veritable gold rush for data, as commercial firms and governments began to seek out, acquire, store, process, label, and generate massive data sets. The term “big data” was used to characterise the huge volumes of data being generated by digitalisation in business, society, and government. This data was obtained from many sources, took many forms, and was being created, captured, or extracted at very high speeds. In its digital form, data had immense value, though its prior, physical form was also valuable. As data became more widely available and used, academic and policy researchers began to examine its nature and usefulness and how it had to be governed as a resource. 

The objective of this essay is to examine the nature of data through a metaphorical lens and explore how policy can be set to govern this resource, particularly in the context of AI technologies. Metaphors help us reason about a phenomenon, especially when we encounter it afresh. Metaphors help make concrete concepts that are abstract and difficult to realise in language. Metaphors about data relate to its value as a concept and as a resource. For example, three ways in which metaphors drive thinking about data are as: an economic resource (“data is oil” metaphor), an epistemic resource (“data is knowledge” metaphor), and a computational resource (“data as infrastructure” metaphor). 

As an economic resource, data enables the construction of AI models that are used to create value across varied industrial and scientific applications. In this view, data is consumed by the model and is usually prepared and processed according to the needs of the algorithm or method used. As an epistemic resource, data is needed for scientific inquiry and is collected or prepared by researchers according to the requirements of the research program. Data is both an input and an output. The computational view of data sees it as a substrate, consisting of ones and zeros, that form the basis of computing. In this view, data is represented in various forms that the computing processes transform to conduct their activity. Data is neither oil with economic value nor knowledge; it is a means by which digital processing is undertaken, and thus functions as infrastructure. 

The three metaphors of data above highlight certain aspects of data and compel those using them to focus on one or the other of the core ideas. When any metaphor is invoked in the context of a policy discussion, its ambit can limit consideration of alternative perspectives. For example, a policy discussion that draws on the metaphor of “data as the new oil” sharpens the focus on the economic aspects, while abstracting away from its aspects as a knowledge resource. This can, at times, lead to contradictions, where economic resources are protected and secured, whereas knowledge resources may be freely shared and distributed. This creates a space to consider another metaphor that can help address these limitations. 

In this essay, we explore the yin-yang metaphor to understand data, its uses, collection, processing, value, and challenges. The following sections examine the implications of this metaphor for aspects of data regulation and conclude with insights for policy. 

The Yin-Yang Metaphor 

The concept of yin and yang from ancient Chinese texts and philosophy refers to opposing yet complementary ideas or objects. The world consists of yin and yang elements – light and dark, fire and water, positive and negative, soft and hard, feminine and masculine. A core tenet of this philosophy is that harmony depends on the coexistence of both the yin and the yang, as they are interrelated and interdependent. The dual aspects of nature affect our understanding of biology, the environment, technology, and the social world. The duality emphasises that suppression of one aspect leads to the dominance of the other, even though both are necessary for the whole. Further, the whole – the combined forces of both – is greater than the sum of its parts. 

The dual aspects of yin and yang, subsequently referred to as yin-yang, are useful for understanding data. First, they permit visualising and recognising the duality of data without inordinate emphasis on one aspect over the other. Second, they remind us of the complementary nature of contrasting, or even opposing, properties that combine to form a whole. Third, they nudge policy discourse towards engaging with these contrasts, thus encouraging empathy and balance. Fourth, they discourage drawing harsh or rigid boundaries, particularly when fluidity and permissiveness are in order. 

This metaphor does not exclude existing metaphors of “data as the new oil” or as “infrastructure;” rather, it is proposed as a complement to enhance our understanding of data and its governance. Viewing data through the yin-yang lens provides a conceptual scaffolding to comprehend issues of governance and stewardship. It broadens existing perspectives on how data is seen, managed, understood, and regulated. 

Yin-Yang of Data 

Property Rights 

The growth of open source software since the early 2000s also spread the idea of open data, or data that is freely shared. Early efforts to “open” data came from governments and civil society organisations seeking broader access to information hitherto available only to a select few through cumbersome processes. Over time, various countries and regions passed laws to ensure public data was made openly available. Today, nearly 200 countries have regulations that enable and promote open data. Open data is expected to be timely, available and accessible, shareable, complete, primary, and licensed under unrestricted conditions (including provisions for modification). 

The contrast to open data is proprietary data, which is controlled, protected, restricted in its use and modification, and limited in its ability to be transmitted, shared, and accessed. Data privacy laws in many countries prohibit the public release of data of a personal nature, referred to as Personally Identifiable Information (PII), and even impose severe penalties for unauthorised disclosure. In current practice, the value of data for AI development has encouraged firms, organisations, and even governments to treat it as a private resource. 

The yin-yang nature of data is evident in this dimension of property rights: data can be both open and proprietary, sometimes within the same domain of use. For example, there have been long-standing efforts to create electronic health records (EHRs) for citizens of India to improve healthcare delivery and also inform policy design. Because EHRs contain PII related to patients, they must be protected from unauthorised access. At the same time, they must also be made accessible to AI models for processing and policy modelling. Regulations surrounding EHRs must therefore be conscious of both these aspects. 

A similar tension exists in mapping initiatives such as OpenStreetMap. Groups and organisations collect data related to objects, landmarks, establishments, buildings, parks, and other locations, which are then uploaded to these sites. These maps are open and contain a wealth of information about a region or neighbourhood. Businesses and public organisations use this data to provide services to residents. However, these maps carry security risks, as they may be used by malicious actors, including states, to target and harm people. In such a case, authorities often remove data points and shut down sites to prevent their misuse. The yin-yang nature of these maps points to both their material benefits and risks, while also highlighting their social value for the communities using them as local, situated resources. 

Data Domains 

Data captured through devices and tools may reflect human domain activity or pertain to phenomena in the environment and nature. Human-domain data are derived from industrial, social, and economic activity. These may include social media content (including text, audio, and video), cultural production such as movies, written material, audio, or media sources reporting on aspects of society. Data may also be obtained from economic activity, including commercial transactions, banking, transfers, taxation, and trade. Further, data may be obtained from formal sectors, such as manufacturing, mining, education, healthcare, or transportation. These activities constitute the human domain of needs, tasks, processes, and everyday commerce. 

Data may also be drawn from the environment, including the earth, the oceans, rivers, the atmosphere, and outer space. While these data sources are of scientific and commercial interest to humans, their origins are not tied to human activity, except insofar as humans collect and aggregate the data. 

AI models draw on data from both domains to build models for prediction and generation. The relevance of such data may vary over time, though much of it can be mined for historic and periodic trends. The yin-yang perspective shows that drawing only from one domain does not deplete the other, though resources for extracting, processing, and storing data have to be allocated across both domains, creating trade-offs that require careful management. 

Large Language Models such as ChatGPT and Gemini are trained mainly on data from the human domain, including books, articles, and web pages. The text in these sources was written and processed by humans for different activities and includes some data on the environmental domain. Consequently, model outputs often reflect human values, assumptions, and biases, while giving comparatively less attention to environmental and ecological concerns. In addition, they cannot directly access data from environmental sources unless it is explicitly provided. 

Cognitive Processing 

Data is processed by humans in a symbolic or sub-symbolic manner. Symbolic data consists of words, numbers, icons and other understandable symbols that humans use for reasoning. Mathematical symbols, for example, enable complex reasoning based on logic and rules, while language reasoning is based on words and their meaning. By contrast, sub-symbolic processing is based on sensations: sights, sounds, touch, smell, and taste. These sensory inputs are processed by humans at a more fundamental level, corresponding to the sense-response mechanisms of the body. Symbolic reasoning occupies a higher level in the cognitive hierarchy and requires inputs from the sub-symbolic mechanisms. 

Data for AI is drawn from both symbolic and sub-symbolic processing methods. Sub-symbolic data is typically processed by deep learners, although reasoning about it requires conversion to symbolic form. From a yin-yang perspective, data for both methods has to be accumulated, labelled, processed and stored differently. They also raise distinct concerns relating to accountability, security, explainability, and fairness. 

Multi-modal AI models are often trained on both symbolic and sub-symbolic data (text, video, images, and audio), for which training data has to be labelled and stored appropriately. Current models are often dominated by symbolic data, which provides the context for inferences. For “common-sense” reasoning, models must be provided with a balance of symbolic and sub-symbolic data to ensure that the inferences are accurate and unbiased. For example, water is “wet” and a bottle is “dry”, but a bottle holding water may be wet or dry, depending on the situation (the base is wet, though the rim is dry). Future AI models will have to resolve these common-sense issues by drawing on both kinds of data. 

Digital and Analog Sources 

Data may be obtained from digital or analog sources. When captured through digital infrastructure devices, such as sensors, data is recorded in digital form, as the process of digitising is built into the design of the devices. When data is obtained from analog sources, such as print media or images on paper or film, it must first be converted into digital form, a process that requires considerable effort and intervention. It also brings into focus questions of permission, copyrights, and ownership associated with the original analog materials. 

Digital data is typically non-rivalrous, whereas analog data is rivalrous, meaning that use by one person or entity precludes use by others (e.g., a book being read by one person cannot be simultaneously read by another). The cost of reproducing digital data is nearly zero, whereas that of analog is non-zero (possibly finite). Similarly, the capacity to reproduce digital data is effectively unrestrained, while it is limited for analog materials. 

The yin-yang nature of data sources highlights these dichotomies of use and also invites a more holistic view. Ancient, written documents, for instance, carry meaning and cultural values that have deep significance in many cultures. For example, paper land records maintained over centuries in the state of Karnataka possess great social and historical significance, much of which was lost when they were digitised to create the electronic land records system. A holistic approach would require being sensitive to such values while enhancing practical use through digitisation. 

Origins 

The origins of data also shape its yin-yang characteristics. Authentic data, which has known and accountable sources, can be distinguished from synthetic data. Authentic data has a representational quality: it stands for and describes a phenomenon. For example, data about economic behaviour reflects the economy and reveals its properties (it is not the economy). Similarly, data in the form of images of cancerous tissue are representations of the cancer. Synthetic data, on the other hand, does not correspond to any actual event, phenomenon, or object from which it was primarily sourced. It is created or simulated as an independent artefact, though it could pertain to the properties and characteristics of a real phenomenon. For example, a synthetic stock-market dataset may reflect the behaviour of actual stocks while being generated from patterns observed in real market data rather than drawn directly from original transactions.

The widespread use of deepfake images, videos, text, and audio has brought

this yin-yang nature of data into sharp focus. A highly realistic video of a political leader draws attention both to its apparent authenticity – how exactly it renders the person’s appearance – and to the fact that it may be entirely fabricated.

Policy Implications of the Yin-Yang Metaphor 

As data cannot be understood by a monolithic view owing to its yin-yang duality, it gives rise to several implications for policy, regulations, and firm practices. The following sections highlight a small subset of the issues that arise in governing data. While not exhaustive, they illustrate how the yin-yang metaphor reveals nuances that might otherwise be overlooked. 

Data Ownership 

Recent court challenges brought by artists, authors, and filmmakers concerning ownership of styles, authored content, and re-created images have drawn attention to the complexities of data ownership. 

Although copyright laws protect creators of original content, works synthesised from original art are not easily classified as copyrighted (by the original creator). Further, data localisation and sovereignty claims also become more complex when applied to synthetic or environmental domain data; for example, data collected on the atmosphere over the Indian subcontinent defies geographic or sovereign boundaries. 

Following the yin-yang metaphor, two insights emerge: 

• Data may be required to be hosted in a region; however, its use in training models may not be similarly constrained. Even when models are required to be trained in the same region, their subsequent use may occur across regions. 

• Even as copyright and intellectual property regulations are amended 

to include styles, movements, genres, schools, and techniques, there 

must remain an understanding that new models and styles will evolve from AI-human collaborations. Regulations must balance protection with continued scope for innovation. 

Data Processing 

Data pre-processing involves cleaning, preparing, labelling, standardising, and possibly anonymising data for use in AI models. Labelling and preparing data is a multi-billion-dollar industry employing thousands of people around the world. Many ethical concerns have been raised regarding labelling conditions, including the employment of low-wage labour and exposing labellers to disturbing content. Data preparation must also account for annotator and contextual bias and varying standards of interpretation across regions. For example, synthetic data generated from mislabelled or biased source data may include those errors and biases. Data from the human domain may require greater effort in preparation and labelling than that from the environment domain. Moreover, because human domain data is embedded within social and cultural contexts, extraction may lead to processing errors.

Policy implications from a yin-yang perspective include: 

 Several jurisdictions are considering regulations requiring synthetic data to be identified as such, the purpose being to inform downstream users whether the data is synthetic or authentic. Such regulations should also consider the trade-offs defined by the purpose for which the data are used and the cost of enforcing such regulations. For example, synthetic data used to train models for healthcare provision would warrant disclosure, but verifying authenticity may incur costs that exceed the benefits in some contexts. This trade-off will be defined by a Pareto-optimal frontier, where unit costs of compliance would have to be plotted against benefits. 

 Regulators are also seeking to flag data origins and sources. When data is obtained from analog sources, there is a requirement to demonstrate that it was obtained legally, as unauthorised use of printed material could be in violation of copyright rules. A yin-yang perspective informs us that printed, analog sources are different from originally digital content in both form and the values associated with them. Ancient, printed texts represent cultural and civilisational consciousness, much of which is lost when digitised. The layered, palimpsest texts from ancient Egypt offer one such example. Policy frameworks must remain sensitive to these issues. 

Data Provenance 

Regulations increasingly require data provenance authentication, particularly for human domain applications. In medical diagnosis, for example, users want to know the data on which the models were trained to be assured that the inferences are accurate and reliable. The yin-yang nature of data complicates such assurances. Data has to be anonymised to be used, limiting the ability to trace its origins.

Synthesised data that is used to complement authentic data to build models may also be difficult to label for provenance, as its source population, on which it is based, may have mixed origins. Similarly, some types of environment-domain data, such as satellite images, do not lend themselves easily to territorial ownership or provenance claims.

Purpose 

Regulations often impose that data be legally collected for a specified purpose, indicated in advance, and its use restricted for that purpose alone. India’s Digital Personal Data Protection (DPDP) Act, 2023, includes such provisions. A yin-yang view of data shows that such a regulation is practically difficult to implement and also that such provisions can be sustained only for a short duration. Data is often collected for a specific purpose that is met in the immediate term, yet the same data can be used later for other beneficial purposes. For example, health data collected for diagnosis and treatment could be used later for improving hospital efficiency. 

It is rarely possible to anticipate all future uses of data stored for posterity. Therefore, it may be wiser to regulate its use for the present and stipulate that future uses be disclosed and publicised. In the example above, if health data was subsequently used in improving hospital efficiency, such use should be duly notified to data principals and owners. 

Retention 

Most regulations of data around the world now state that data should be retained or stored on servers for the shortest possible time for which they are required. At the same time, other legal frameworks – including those covering health care, taxation, civic administration, or property – often require data to be maintained and stored for a minimum length of time, and in some cases forever, as with land records. These contradictory requirements point to the yin-yang nature of data and complications that arise when data with different retention requirements are mixed in AI models. There is also the related issue of models retaining the data indirectly for as long as they exist. In such situations, regulations have to be crafted to address the core issues of data use and privacy, rather than specify time durations. Human domain data, for instance, can be stored in a time-boxed manner, expiring after a certain period, whereas environment domain data may be so tagged to be held in perpetuity. Such approaches would have to be designed keeping in mind other existing and overlapping laws. 

Conclusion 

The yin-yang concept from Chinese philosophy encourages an understanding of the world that recognises duality and contrasts, and to adjust our own views of it accordingly. AI, computational models, and data are evolving around the world at a frenetic pace. It is imperative that governing bodies recognise and address the dual aspects of data while regulating it, ensuring that data and models are guided towards benefit and social good, while mitigating potential harms. 

Policy makers may keep in mind a few key aspects of the yin-yang view of data while designing regulations and guidelines. 

One key method to balance the yin and yang aspects of data is to follow the Pareto optimal principle. Long established in economics, it encourages policymakers to evaluate how different stakeholders are affected by regulatory choices and to identify situations in which benefits can be achieved without making others worse off. For example, if a country has to decide between local data hosting or enabling data hosting to be on the cloud, unrestrained by the location, then it will have to decide on some key parameters that are impacted by this decision. One such parameter is the value generated by local hosting providers, possibly their combined contribution to the domestic product. This value would have to be balanced against the convenience of doing business globally. A Pareto-optimal frontier would examine how increasing localisation could worsen the value of international trade, or vice versa. 

Second, it is important to realise that there can be no absolute and permanently binding regulations, especially in the human domain. As technology advances and diffuses through society, people’s habits and patterns of use will likely change and evolve. Rigid policies from the past may not effectively guide regulations regarding usage and adoption over time. At the same time, theory has shown that it is better for policies and regulations to be fixed and stable, so that businesses can adjust their own procedures to survive in the market. The yin-yang principle highlights both permanence and evolution and suggests that though policies should be stable, their evolution should address the needs of various stakeholders. 

Third, effective governance depends on broad public understanding of data and its role within digital systems. While it may not be possible to anticipate all the implications of emerging technologies, policymakers, businesses, and citizens must continue to examine and understand the evolving properties of data. The yin-yang perspective reminds us that data will continue to exhibit complementary and sometimes competing characteristics. It is important, then, to spread understanding and comprehension of the various aspects and properties of data, amongst all users and stakeholders.