Digital Contact Tracing: Large-scale Geolocation Data as an Alternative to Bluetooth-based Apps' Failure

José González Cabañas, Ángel Cuevas, Rubén Cuevas, Martin MaierarXiv 2021 · arXiv

TL;DR

Bluetooth-based COVID-19 contact tracing apps have failed due to low adoption rates. This paper proposes an alternative using large-scale, existing geolocation data from BigTech companies, combined with a novel protocol to ensure privacy for infected individuals and prevent misuse by authorities or tech giants.

What Problem Does This Paper Address?

The COVID-19 pandemic highlighted the critical need for effective contact tracing. While digital solutions were explored, Bluetooth-based mobile apps, despite their privacy-preserving designs (like DP-3T), have largely failed to achieve the necessary high adoption rates (estimated at 60% or more) to be effective. This low uptake means they miss a significant portion of potential exposures, rendering them insufficient for pandemic control. Furthermore, Bluetooth's limited range (under 2 meters) misses airborne transmission risks. This paper addresses this critical gap by proposing a system that leverages existing, widely adopted technologies and data sources to overcome the adoption barrier, while also considering airborne transmission and robust privacy guarantees.

Methods

The proposed solution shifts from Bluetooth proximity detection to leveraging existing large-scale geolocation data from "Location Providers" (LPs) like Google and Facebook. The core of the method is a novel protocol involving four key players:

  • Health Authority (HA): Identifies infected individuals and initiates the tracing process.
  • Location Provider (LP): Owns and processes vast amounts of user location data.
  • Identity Provider (IDP): Provides anonymized mobile numbers (user IDs) to the HA.
  • Independent Third-Party Authority (ITPA): Acts as a trusted intermediary for key exchange.

The protocol works as follows:

  1. The HA sends a list of anonymized infected user IDs (mixed with random IDs) to the LP.
  2. The LP uses its geolocation data to identify potential "risk contacts" for these IDs, considering airborne transmission distances. It also aggregates Point of Interest (POI) data.
  3. Crucially, the LP encrypts the risk contact lists and POI data for each group using separate keys and sends this encrypted information back to the HA.
  4. The HA requests decryption keys for only the "infected" groups from the ITPA, which in turn requests them from the LP.
  5. The HA can then decrypt the risk contact information for infected individuals, allowing them to be notified.

Privacy is enhanced through K-anonymity (mixing infected IDs with many random ones), cryptography (encryption of sensitive data), and non-repudiation auditing to deter malicious behavior.

Datasets

The paper does not explicitly train or evaluate its proposed system on specific datasets. Instead, it relies on publicly available data for smartphone penetration, Android OS market share, Facebook Monthly Active Users (MAU), and Bluetooth app installations across 18 countries (cited from sources like StatCounter, Pew Research, and the World Bank) to demonstrate the adoption gap of current solutions and the potential reach of BigTech data.

Key Insights and Ideas

  • The adoption barrier is the Achilles' heel of Bluetooth contact tracing: The paper strongly argues that the failure of current apps lies not in their technical design but in their inability to achieve critical mass adoption.
  • Leveraging BigTech's existing infrastructure is a pragmatic solution: Instead of building new systems, the paper proposes tapping into the billions of users already tracked by major tech companies, bypassing the adoption problem.
  • Privacy can be preserved through protocol design: The proposed multi-party protocol, with its emphasis on encryption, anonymization, and a trusted third party, aims to protect infected individuals' identities and prevent misuse of data by both LPs and HAs.
  • Airborne transmission is a critical consideration: The system is designed to account for transmission beyond close proximity, a limitation of Bluetooth-based approaches.
  • Non-repudiation and auditing are key for trust: The inclusion of auditing mechanisms provides a deterrent against malicious use of the system by any party.

Key Results

The paper does not present quantitative results from a deployed system. Instead, its "results" are primarily:

  • Demonstration of the adoption gap: Table I clearly shows that even in countries with high smartphone penetration, Bluetooth contact tracing app installations and estimated active users are significantly lower than BigTech platform penetration (e.g., Facebook, Android). This visually supports the core problem statement.
  • A technically sound architectural proposal: The paper outlines a detailed protocol and architecture that, in theory, addresses the identified problems of low adoption and privacy concerns.

Limitations and Future Work

  • No empirical validation: The authors acknowledge that they have no evidence of whether their system will solve the contact-tracing problem and that it is a "technically sound alternative worth exploring."
  • Reliance on BigTech cooperation: The success of the system hinges on the willingness and ability of BigTech companies to participate and share data, which may be a significant hurdle.
  • Complexity of the protocol: The multi-party protocol, while designed for privacy, introduces complexity in implementation and coordination.
  • Definition of "risk contacts" and POI analysis: The paper states that the specific algorithms for identifying risk contacts and the definition of POI categories should be determined by epidemiologists and are outside the scope of this work.
  • Potential for targeted attacks: While countermeasures are proposed, the paper acknowledges that a malicious HA could theoretically attempt targeted attacks by manipulating requests, though auditing mechanisms are in place to detect this.

Future work would involve implementing and testing this proposed architecture to empirically validate its effectiveness and privacy guarantees.

Summary updated

BibTeX
@misc{cabanas2021digital,
  author = {José González Cabañas and Ángel Cuevas and Rubén Cuevas and Martin Maier},
  title = {{Digital Contact Tracing: Large-scale Geolocation Data as an Alternative to Bluetooth-based Apps' Failure}},
  year = {2021},
  eprint = {2101.07024},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2101.07024},
}

Loading related content...

Figure