https:\/\/uplatz.com\/course-details\/asset-accounting-in-sap\/33<\/a><\/p>\n1.1 Disambiguation: The “Data Without Borders” Initiatives vs. The Concept<\/b><\/h3>\n
The ambiguity in the term “Data Without Borders” requires clarification. Several real-world organizations use this or similar branding, but their missions and methods differ significantly. They are, in effect, symptoms of the core problem of data siloing, not a unified technological solution.<\/span><\/p>\n\n- Statistics Without Borders (SWB):<\/b> An apolitical, pro-bono volunteer organization operating under the auspices of the American Statistical Association.<\/span>1<\/span> Its mission is to provide statistical analysis and data science services to non-profits, NGOs, and governments, with a focus on helping developing countries.<\/span>3<\/span> Its projects have included analyzing public health surveys in Sierra Leone, assessing the economic impact of the 2010 Haiti earthquake, and designing a long-term longitudinal study for Save the Children in Ethiopia.<\/span>1<\/span><\/li>\n
- “Data Without Borders” (DWB):<\/b> A 2011-era initiative founded by Jake Porway, conceived as a “data science exchange”.<\/span>4<\/span> Its goal was to bridge the “data gap” by connecting data scientists and engineers (“generous geeks”) with non-profits that were “drowning in data” but lacked the resources to analyze it.<\/span>4<\/span><\/li>\n
- Data Science Without Borders (DSWB) Africa:<\/b> A contemporary project focused on strengthening data systems and building advanced data science pipelines in Africa.<\/span>6<\/span> With pathfinder institutions in Cameroon, Ethiopia, and Senegal, DSWB aims to build local capacity and foster collaboration by harmonizing existing health datasets using common standards, such as the OMOP Common Data Model.<\/span>6<\/span><\/li>\n
- “Data without Boundaries” (DwB) Europe:<\/b> A project funded by the European Union’s 7th Framework Program (2011-2015).<\/span>7<\/span> Unlike the pro-bono organizations, its mission was to improve researcher access to <\/span>official, confidential microdata<\/span><\/i> from national statistical institutes. This project focused on developing methods for “Statistical Disclosure Control (SDC)” to create anonymized public-use files, representing an important technical precursor to modern synthetic data generation.<\/span>7<\/span><\/li>\n<\/ul>\n
These organizations all exist to circumvent the very “borders” this report investigates. They are a human-powered, manual workaround to the fact that data is siloed and its benefits are not evenly distributed.<\/span>4<\/span> The “Data without Boundaries” project, in particular, demonstrates that the problem of transnational data access is chronic, and its 2011-era SDC solution has now evolved into the more powerful generative AI techniques of today.<\/span>7<\/span><\/p>\nTable 1: Disambiguation of ‘Data Without Borders’ and Related Initiatives<\/b><\/p>\n
<\/p>\n
\n\n\n| Initiative Name<\/b><\/td>\n | Sponsoring\/Affiliated Body<\/b><\/td>\n | Primary Mission<\/b><\/td>\n | Involvement with Synthetic Data<\/b><\/td>\n<\/tr>\n\n| Statistics Without Borders (SWB)<\/b><\/td>\n | American Statistical Association (ASA)<\/span><\/td>\n | Pro-bono statistical consulting and data science services for NGOs and developing countries.<\/span>1<\/span><\/td>\n | None. Focuses on analysis of existing, real data.<\/span>1<\/span><\/td>\n<\/tr>\n\n| “Data Without Borders” (DWB)<\/b><\/td>\n | Jake Porway (Founder)<\/span><\/td>\n | A “data science exchange” to connect data scientists with non-profits needing analysis.<\/span>4<\/span><\/td>\n | None. Focuses on analysis of existing, real data.<\/span>4<\/span><\/td>\n<\/tr>\n\n| Data Science Without Borders (DSWB) Africa<\/b><\/td>\n | Africa CDC, APHRC, LSHTM, et al.<\/span><\/td>\n | Build data science capacity and harmonize health data systems in Africa.<\/span>6<\/span><\/td>\n | None. Focuses on data harmonization (e.g., OMOP CDM) of existing data.<\/span>6<\/span><\/td>\n<\/tr>\n\n| “Data without Boundaries” (DwB) Europe<\/b><\/td>\n | European Union (FP7)<\/span><\/td>\n | Improve researcher access to confidential, official microdata from national statistical institutes.<\/span>7<\/span><\/td>\n | No, but a direct technical precursor. Focused on “Statistical Disclosure Control” (SDC) to create “anonymized microdata”.<\/span>7<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n <\/p>\n 1.2 Defining the “Borders”: The True Barriers to Global Data Collaboration<\/b><\/h3>\n <\/p>\n The true “borders” are not geographical but a fragmented and increasingly hostile patchwork of legal, economic, and social restrictions on data. Synthetic data is proposed as a technological “passport” to navigate this landscape.<\/span><\/p>\n\n- Legal Border 1: Data Sovereignty & Localization<\/span>
\n<\/span>The most formidable barriers are legal. A rising tide of digital nationalism has formalized the concept of data sovereignty\u2014the principle that data is subject to the laws of the nation in which it is collected.8 This principle is enforced through data localization mandates, which require organizations to store and\/or process data within a country’s physical borders.11 Nations like China, Russia, and South Africa have all implemented such requirements, effectively “cutting the ‘world’ out of the ‘World Wide Web'”.12<\/span><\/li>\n- Legal Border 2: The Compliance Gauntlet (GDPR)<\/span>
\n<\/span>Simultaneously, regulations with extraterritorial reach, most notably the EU’s General Data Protection Regulation (GDPR), govern the data of their citizens regardless of where in the world it is processed.15 This creates a “patchwork of privacy laws” 17 with “overlapping and conflicting requirements”.14 Consequently, simple cross-border data transfers, even when using approved mechanisms like Standard Contractual Clauses (SCCs), have become a complex, high-risk, and operationally burdensome legal endeavor.14<\/span><\/li>\n- Economic & Operational Borders<\/span>
\n<\/span>These legal borders impose severe economic friction. Localization mandates force multinational organizations to “duplicate infrastructure” 12 and build “bespoke data storage centers” in multiple jurisdictions.14 This is not only inefficient but “resource-intensive” 12, increasing costs that are often passed on to consumers.14 These burdens have a “disproportionate effect on smaller businesses,” actively “thwarting growth opportunities” and stifling innovation.14<\/span><\/li>\n- The Security Paradox<\/span>
\n<\/span>A critical paradox has emerged: policies enacted in the name of national security and privacy may, in fact, decrease data security. Centralized, heavily resourced data centers are often more secure than the “minimally-resourced local facilities” that companies are forced to establish to comply with localization laws.14 These local outposts are “more likely to permit network intrusions and data compromises”.14 This paradox creates the central business case for a Privacy-Enhancing Technology (PET): a tool that can satisfy the goal of privacy and security while bypassing the inefficient and insecure policy of localization.<\/span><\/li>\n- Social & Trust Borders<\/span>
\n<\/span>Finally, there is a deep-seated social barrier. Public opinion surveys reveal a significant distrust of international data sharing.18 A 2015 US survey, for example, found that while 71% of adults supported sharing their data with US university researchers, that support plummeted to just 39% for “university researchers in other countries”.18 This “trust border” is a significant social hurdle that a technological solution must also be prepared to address.<\/span><\/li>\n<\/ul>\n <\/p>\n 2.0 Synthetic Data as a Proposed Passport: Generation and Principles<\/b><\/h2>\n <\/p>\n To bypass these borders, organizations are turning to synthetic data. This technology is proposed as a “passport”\u2014a new, clean, and authentic-looking document that retains all the <\/span>statistical characteristics<\/span><\/i> of the original data but severs any link to a real person.<\/span><\/p>\n <\/p>\n 2.1 Defining the “Passport”: What is Synthetic Data?<\/b><\/h3>\n <\/p>\n Synthetic data is “artificially generated information” <\/span>19<\/span> or “imitation data” <\/span>20<\/span> created by computer algorithms.<\/span>21<\/span> It is engineered to “mimic” <\/span>19<\/span> and “resemble” <\/span>20<\/span> the statistical properties, patterns, correlations, and structure of a real-world dataset.<\/span><\/p>\nThe core premise is a paradigm shift from traditional anonymization (like masking or randomization), which merely alters <\/span>existing<\/span><\/i> records.<\/span>24<\/span> Synthetic data is “created entirely from scratch”.<\/span>24<\/span> This “no one-to-one relationship” between a synthetic record and a real individual is the foundation of its claim to be “fully compliant with data privacy regulations” <\/span>16<\/span> because it “contains no actual personal or sensitive information”.<\/span>16<\/span> It aims to be a “perfect proxy” for the original data, severing the link to individual identity while preserving the aggregate insights.<\/span>24<\/span><\/p>\nThis “passport” can come in several forms, depending on the use case:<\/span><\/p>\n\n- Fully Synthetic Data:<\/b> The entire dataset is algorithmically generated, and no real data is present. This method is considered to have the lowest re-identification risk and is the primary type used for privacy-preserving analytics and sharing.<\/span>23<\/span><\/li>\n
- Partially Synthetic Data:<\/b> Only specific sensitive values (e.g., names, social security numbers) are replaced with synthetic ones, while other non-sensitive real data (e.g., transaction amounts) remain.<\/span>23<\/span> This approach is used to maintain higher analytical validity but comes with a “higher disclosure risk” as some true values remain.<\/span>23<\/span><\/li>\n
- Non-representative (“Dummy”) Data:<\/b> This data is structurally similar to the original (e.g., same column names, data types) but is not statistically representative. It is “dummy data” useful for software testing or code reproduction, but not for analysis.<\/span>28<\/span><\/li>\n<\/ul>\n
<\/p>\n 2.2 The “Passport” Office: How Deep Generative Models Create Synthetic Data<\/b><\/h3>\n <\/p>\n The technology used to create synthetic data has evolved from simple “rule-based approaches” <\/span>23<\/span> and “traditional statistical models” (like those explored in the DwB project <\/span>7<\/span>) to “state-of-the-art” <\/span>23<\/span> deep generative models.<\/span>30<\/span> This is not just an increase in power; it is a fundamental shift in capability. While traditional methods required a human to <\/span>pre-define<\/span><\/i> the statistical distributions to be mimicked, generative models <\/span>learn<\/span><\/i> the “patterns, correlations and statistical properties” <\/span>24<\/span> themselves, capturing complex, high-dimensional, and non-obvious relationships <\/span>32<\/span> that no human could define.<\/span><\/p>\nTwo types of deep learning models dominate this field:<\/span><\/p>\n\n- Technique 1: Generative Adversarial Networks (GANs)<\/span>
\n<\/span>A GAN is a “strong class of generative models” 33 comprised of two competing neural networks 22:<\/span><\/li>\n<\/ul>\n\n- The Generator:<\/b> Its job is to “create synthetic data samples” <\/span>33<\/span>, often by taking random noise as input.<\/span><\/li>\n
- The Discriminator: Its job is to act as a “distinguisher” 33, analyzing a piece of data and determining if it is “real” (from the original dataset) or “fake” (from the generator).<\/span>
\n<\/span>The two networks are trained in an “adversarial” process. The generator’s sole goal is to “produce data that attempts to fool” the discriminator.35 This competition forces the generator to become progressively better at producing “realistic synthetic data” 35 that is statistically indistinguishable from the original.36<\/span><\/li>\n<\/ol>\n\n- Technique 2: Variational Autoencoders (VAEs)<\/span>
\n<\/span>A VAE consists of two networks 35:<\/span><\/li>\n<\/ul>\n\n- The Encoder:<\/b> This network “summarizes the characteristics and patterns of real-world data”.<\/span>35<\/span> It does this by compressing the input data into a “lower-dimensional latent space,” which is a probabilistic representation of the data’s key features.<\/span>33<\/span><\/li>\n
- The Decoder: This network “attempts to convert that summary into a lifelike synthetic dataset”.35<\/span>
\n<\/span>The generative step is to sample a point from the probabilistic latent space 33 and feed it to the decoder. This generates a new data sample that follows the learned patterns but is not a simple copy of the original input.<\/span><\/li>\n<\/ol>\nIn this new paradigm, the generative model itself\u2014the GAN or VAE trained on the original, sensitive data\u2014becomes a critical asset. It is a “blueprint” <\/span>35<\/span> or “summary” <\/span>35<\/span> of the sensitive data. This “concentrated information” <\/span>38<\/span> also makes the model a significant new liability and an attractive target for cyberattacks.<\/span><\/p>\n <\/p>\n 3.0 The Regulatory Ambiguity: Is This Passport Legally Valid?<\/b><\/h2>\n <\/p>\n The central premise of synthetic data as a global collaboration tool rests on a single, powerful assumption: that the resulting dataset is “anonymous” and therefore not subject to the cross-border transfer restrictions of regulations like GDPR. This assumption is a dangerous oversimplification.<\/span><\/p>\n <\/p>\n 3.1 The “Anonymization” Claim<\/b><\/h3>\n <\/p>\n Synthetic data is heavily marketed as a “compliance-friendly alternative” <\/span>39<\/span> that is “fully compliant with data privacy regulations like GDPR, CCPA, and HIPAA”.<\/span>16<\/span> The logic is that because it “contains no personally identifiable information (PII)” <\/span>16<\/span> and no “one-to-one relationship” with real individuals <\/span>24<\/span>, it is legally and functionally anonymous. It is positioned as a superior alternative to traditional de-identification, which often fails to protect against re-identification and results in “decreased utility and statistical relevance”.<\/span>25<\/span><\/p>\n <\/p>\n 3.2 The Critical Legal Nuance: The <\/b>Process<\/i><\/b> vs. The <\/b>Product<\/i><\/b><\/h3>\n <\/p>\n The legal analysis of synthetic data is not a single question but two distinct ones. Failure to separate them is the most common and critical error in this domain.<\/span><\/p>\nPart 1: The Generation Process is Fully Regulated<\/span><\/p>\nAn organization cannot claim immunity from data protection law simply because its end product is synthetic. To create the synthetic data, one must first access, analyze, and train a model on the original, real, sensitive dataset.40 This “initial processing” 40 of personal data is fully subject to all data protection laws.41<\/span><\/p>\nUnder GDPR, this means an organization <\/span>must<\/span><\/i> have a “lawful basis” (such as legitimate interest or explicit consent) to process the original data for this purpose.<\/span>40<\/span> It also means a “Data Protection Impact Assessment” (DPIA) is almost certainly required <\/span>before<\/span><\/i> the generation can even begin.<\/span>41<\/span> This “original sin” of processing PII means synthetic data is <\/span>not<\/span><\/i> a tool for organizations to analyze data they have no legal right to acquire. It is a tool for <\/span>data sharing<\/span><\/i> by organizations that already have a legal basis to <\/span>hold<\/span><\/i> the data.<\/span><\/p>\nPart 2: The Output Product’s Ambiguous Status<\/span><\/p>\nThe second question is whether the resulting synthetic dataset is legally “personal data.” The answer is a complex and non-guaranteed “it depends.”<\/span><\/p>\nThere is <\/span>no one-size-fits-all answer<\/b>.<\/span>41<\/span> Regulators, in an “orthodox” approach, start with the <\/span>presumption<\/span><\/i> that if the source data was personal, the synthetic output <\/span>remains personal data<\/span><\/i>.<\/span>41<\/span><\/p>\nThe <\/span> | | | | |
![\"\"](\"https:\/\/uplatz.com\/blog\/wp-content\/uploads\/2025\/11\/Synthetic-Data-for-Global-Collaboration-1024x576.jpg\")
<\/p>\n