Data interoperability. Is it possible to describe the entire world in a single language? 

During digital transformations, data constantly flows between companies, systems, devices, and algorithms. But do they truly understand one another? The question of data interoperability isn’t just a technical one — it’s about whether we can bring meaning to the digital world. 

Spis treści   

• What is data interoperability?
• What is the idea behind the Digital Product Passport, and what stands in the way of making it a reality?
• Why can’t we create one universal data model that works for every use ? 
• Is compliance with a standard enough to make data truly understandable?
• What can we learn from ISO 20022 — a successful but incomplete standard?
• Who controls data standards, and can interoperability lead to monopolization?
• Can artificial intelligence solve the problem of data interoperability?
• Why do humans still need to be involved in training AI systems?

Academic definition vs. business reality. What is data interoperability?

At first glance, the concept of data interoperability seems simple and intuitive. According to the Polish Act on Computerization, interoperability is defined as “the ability of various entities and the ICT systems and public registers they use to work together to achieve mutually beneficial and agreed-upon goals, taking into account the sharing of information and knowledge through business processes supported by data exchange via the ICT systems used by those entities.”  – source: https://www.gov.pl/web/ia/interoperacyjnosc-w-ustawie-o-informatyzacji

Put simply, it’s about ensuring that data created in one place can be easily read and used by other systems and organizations—without the need for costly and time-consuming technical adjustments.

But doesn’t this academic vision of interoperability sound a bit too idealistic? After all, we’re dealing with a wide variety of systems, standards, users, and — most importantly— business interests. Let’s take a closer look at what really stands in the way of achieving true data interoperability. We’ll examine the key challenges and limitations.

Data interoperability in Digital Product Passports

Digital Product Passports (DPP) have recently been gaining traction. They are based on the idea that a product’s manufacturer enters detailed information about the item into a system — including data on materials, production processes, and disposal methods.

Thanks to interoperability, this same data could be instantly accessible to all market participants. Retailers, regulatory bodies, and consumers could easily read and use it. In this way, data would support supply chain management, quality control, and informed purchasing decisions. However, these ambitious plans and promising pilot projects often clash with the harsh reality of implementation. Are we truly capable of creating a universal data language that is understandable to everyone and in every situation? The attempt to achieve truly global data interoperability quickly runs into fundamental barriers — the sheer volume of data and the complexity of the world we seek to describe.

Complexity – the enemy of universal data models 

Let’s imagine the task of creating a unified data model for an operating room. At first, it might seem manageable: a list of tools, their quantities, and basic parameters. But as we start adding more elements, complexity explodes. We’d need to include a list of medical equipment, followed by information on the staff needed for standard procedures. Next, we’d have to catalogue all types of surgeries, including the tools and devices required for each. Eventually, we’d be describing the entire hospital: its structure, personnel, resources, and procedures. And that’s just one hospital.

Now imagine scaling this up to an entire healthcare system — and then across all sectors of life and the economy. The task becomes impossible. We would have to create a universal dictionary and data model that describes every aspect of reality. It would be like trying to compile a thirty-volume encyclopedia — just to find information about a single tool in a single operating room. Searching such a massive, complex dataset would become virtually impossible.

When we examine this overwhelming complexity, the idea of universal interoperability begins to feel like a utopia. That’s why ambitious initiatives like Digital Product Passports deserve a cautious — even skeptical — outlook. Their creators have noble intentions: to improve transparency and facilitate access to information. Yet the attempt to establish a single standard for an infinite variety of products may ultimately fail due to the unmanageable complexity involved.

When looking for a tangible explanation of interoperability, the idea of standards naturally comes to mind. Interoperability means that different systems can understand one another — and for that, they need shared rules. In a perfect world, data entered into one system would be automatically understood and used by another. But in practice, without a precisely defined and widely adopted standard, such data exchange is difficult — if not impossible.

Barcodes offer a helpful example. They represent a standard — an agreed-upon way of encoding information using a sequence of bars and spaces. Thanks to this, a store scanner can read a product’s price or identifier. Barcodes solve a specific, narrow problem by enabling automatic reading of certain data. But even this system has limitations. Although the encoded information is technically readable, it lacks value without understanding the context and meaning of each part. It’s like knowing the alphabet but not the language.

That’s why interoperability is more than just technical compatibility. It’s about understanding data regardless of its origin. A standard is often the result of agreements among a limited group addressing a specific issue. But real interoperability demands more — a shared understanding of data’s meaning which wouldliminate the need to constantly ask, “what did the author mean?” Standardizing formats is only the first step.

In discussions of standardization in specific domains, ISO 20022 is hard to ignore. This global standard was created to harmonize data exchange in payments. Its goal is to allow financial institutions to work together and transfer funds efficiently. It also aims to increase transaction transparency and provide richer information about senders and beneficiaries.

The introduction of ISO 20022 has brought clear benefits. The standard defines additional data fields, such as amount, currency, and beneficiary details. As a result, transactions have become more transparent and easier to trace. ISO 20022 makes them more structured and easier to verify for legal and regulatory compliance.

Yet even a sophisticated standard like ISO 20022 has not achieved full system interoperability. Within the standard, there’s room to attach additional transaction documentation in nearly any format. This flexibility, while seemingly beneficial, has led to the emergence of “subgroups” of the standard. For instance, banks aligned with HSBC have agreed on a specific way to include trade documentation in an ISO 20022 message. Other banking groups follow different conventions. As a result, when a transaction from an HSBC-aligned bank reaches another group’s bank, the system may struggle to interpret the attached documents — even though the core structure of the ISO 20022 message is understood.

Efforts to achieve interoperability within specific sectors can lead to power concentration. When a single organization defines the main data models, it gains influence over the entire market — and may impose its terms on others. A good example is SWIFT — the dominant communication network for banks. It played a key role in introducing ISO 20022 for international payments.

As the main promoter of this standard, SWIFT not only facilitated its implementation but also gained influence over its future development. Any change to ISO 20022 requires banks to adapt their systems, incurring costs and creating dependency on a single operator. To reduce this complexity, the market has introduced “hubs” — intermediaries that translate data between different standards. In finance, correspondent banks have fulfilled this role for years.

While hubs offer a practical solution to interoperability challenges, they are not without drawbacks. They introduce added costs, delays, and complexity in data flow. Achieving full alignment with a single global standard often proves too expensive and difficult. As a result, centralized solutions or hub networks emerge, balancing flexibility with control.

Given the vast complexity of data, many pin their hopes on artificial intelligence (AI). Machine learning models can recognize patterns and structure information. At first glance, this seems like a promising path to unprecedented levels of interoperability. We might imagine AI analyzing vast datasets from different sources and autonomously creating universal models and dictionaries — bypassing the limits of human understanding and manual data integration.

But this vision should be approached with caution. There is a real risk that relying solely on AI to create data models for interoperability could do more harm than good. Imagine a situation where AI generates a complex data model, showing how elements such as customers, orders, and products are interconnected. The model could become so elaborate and vast that even domain experts can’t understand it.

This raises the question of verification. Who will check the accuracy and relevance of AI-generated models? If a model is too complex for humans to grasp, it may contain errors, inconsistencies, or arbitrary connections — leading to misinterpretation of data. Relying on such unverifiable models in business processes could have serious consequences and become a “road to nowhere.” Instead of simplifying data exchange, we might create even more confusion — based on seemingly intelligent but ultimately ungovernable structures.

Read more in our article: How to prepare data for the AI era?

The risk of bias — when AI repeats our mistakes 

Human involvement is essential when implementing AI-based systems — especially during training and validation phases. We are already seeing the rise of new job roles focused solely on reviewing AI-generated content, often reduced to binary judgments: “yes” or “no.” This manual data labeling helps steer AI and ensures high-quality performance.

But human involvement brings a fundamental limitation: AI can never be smarter than the dataset it was trained on — and thus no smarter than the people who created or reviewed that data. If the training data is incomplete, erroneous, or reflects biased opinions, those flaws will inevitably be transferred to the AI model.

Selecting the right group of people to train AI is particularly challenging when building universal data models. How do you ensure a diverse, representative group in terms of knowledge, experience, and perspectives? If only experts or individuals with similar views are involved, the AI will inherit their biases and blind spots. In such a case, we won’t create an objective, universal tool for data interpretation. Instead, we’ll end up with a system that reflects only a narrow slice of reality — potentially reinforcing existing flaws.

Conclusion 

Data interoperability is far more than just technical compatibility. It is the pursuit of a shared language that enables genuine understanding between systems, organizations, and technologies. But the more complex the world we try to describe, the quicker we hit the limits of universal models.

The example of Digital Product Passports shows that while transparency and easy access to data are appealing goals, their realization is hampered by diverse needs, data formats, and interpretations. Neither standards nor AI alone can guarantee success. Human oversight and agreement on data meaning are still essential.

What matters is not just whether data is exchanged, but how it is understood. Effective data exchange requires flexibility, collaboration, and conscious intent. Ultimately, it is shared understanding — not format — that determines the true value of data.