Privacy – or security or any other desirable, ethereal property – by design sounds like a great thing to do. Alas, design is complicated and hard to guide or control as a process. One common misunderstanding has become obvious in current efforts to develop and deploy contact tracing technology contributing to epidemic control. Some of these efforts such as DP^3T, TCN, or Apple’s & Google’s announcement promote privacy to the top of their list of objectives and requirements. This is wrong. It would be appropriate in an R&D project developing experimental technology, but contact tracing is an actual, real-world application and must fulfill real-world requirements. Premature optimization for technical privacy protection does not help its cause.
First and foremost, an application needs to support a basic set of use cases and provide the necessary functions in such a way that the overall approach makes sense as a solution of the given problem(s). For contact tracing, essential use cases include:
- contact identification,
- contact listing, and
- contact follow-up.
In addition, any large-scale application of contract tracing needs to support:
- safeguards against abuse, and
- monitoring and governance.
Each use case entails requirements. Contact identification must be sufficiently reliable and comprehensive; it must also take place quickly after an infection has been detected. Contact listing needs some form of risk assessment, a method to notify contacts, and a way to justify mandatory quarantine requests. Contact follow-up needs some idea how and when to interact with listed contacts. Underlying the whole design must be some conception of which contacts matter, what an identified contact implies, what to do with or require from identified contact persons, and what to achieve through contact tracing. There needs to be some definition of success and failure for the system and individual cases, and some monitoring of how well the system operates. One also has to think about possible abuses and misuses of the system such as evasion, manipulation, or exploitation, and find ways to prevent them or to deal with them when they occur.
Such questions are to be answered in the high-level design of a contact tracing system. They can and should be pondered at the level of paper prototypes, forcing developers to get specific but allowing quick modification and user testing. Technology has to be considered at this point primarily as a constraint: What is realistically possible or available and would the design be feasible to implement? However, some fundamental design decisions have to be made at this level after evaluating alternatives, for example, which parts of the system to automate and which ones to leave to humans, or which technologies, platforms, and devices to consider and use.
Like any design process, this high-level system design may take any number of iterations before converging toward something that might work when implemented. New questions will likely come up in the process. If, for example, the system is to leave tracing to humans, how much time can they spend per case, how many of them will be needed, how will they work, and which types of data and support would really help them?
Secondary requirements like performance or privacy can and should already be considered at this stage. Privacy by design means just that, to consider privacy protection as dimensions of the design spaces from the beginning on. However, privacy is a dependent design dimension and like all other requirements it is subject to trade-offs. Dependent means that any design decision can affect the privacy properties of a system. One cannot delegate privacy to a system component or function that would take care of it comprehensively regardless of the design of any other aspect of the system. Trade-offs occur when once has to choose between design alternatives; each option will likely have some advantages over the others but also some disadvantages, so that one has to compromise and keep a balance.
Misunderstanding privacy by design as privacy technology über alles, demonstrated by current proposals for privacy-preserving contact tracing, is a recipe for disaster. Starting with perfect technical privacy protection as the primary requirement constitutes a premature optimization that de-prioritizes all other requirements and design dimensions, delays important design decisions while arbitrarily constraining them without impact assessment, and prevents well-considered trade-offs from being made. The most likely result is a system that performs well at most in the privacy dimension, reflecting the priorities of its designers.
As a symptom, none of the proposals for privacy-preserving contact tracing has yet answered question like the following: How does it assure the reliability of the data it collects or produces? Which failure modes and error rates does it produce? How is the system to be monitored for problems and abuses? In which institutional framework is it designed to operate? How does it distribute responsibilities between involved parties? How are outputs of the system to be interpreted and used in the real world, which consequences should they have and which ones are not desirable? How can its operation become transparent for its users? Should participation be mandatory or voluntary and how can the design be optimized for either case? If participation is mandatory, how would this be enforced, how can the system be made universally accessible for everyone, and how may people evade it? If voluntary, which incentives does the system create and which features let users trust or distrust the system? Such questions need to be discussed and addressed long before the technical minutiae of data protection.
Placing technical privacy protection in the center of attention can make sense in a research project, where one develops new technology to evaluate its properties and capabilities. The stakes are low in such a project, where the results are prototypes and research papers. Developing a real-world system, especially one to be used at the intended scale of contact tracing apps, requires a broader perspective and comprehensive requirements analysis.
P.S. (2020-04-18): Government Digital Services of Singapore with their TraceTogether app apparently got their requirements analysis and design process right:
One thing that sets TraceTogether apart from most private efforts to build a Bluetooth contact tracer, is that we have been working closely with the public health authorities from day 1. (…) The team has shadowed actual real-life contact tracers in order to empathise with their challenges.
P.S. (2020-04-19): The closest to a requirements document I have seen so far is this: Mobile applications to support contact tracing in the EU’s fight against COVID-19, Common EU Toolbox for Member States (via).
P.S. (2020-04-22): The Ada Lovelace Institute published a quick evidence review report titled: Exit through the App Store? A rapid evidence review on the technical considerations and societal implications of using technology to transition from the COVID-19 crisis, which makes a number of reasonable recommendations.
Ein Kommentar zu „Privacy by Design or Poor Requirements Engineering?“
Die Kommentarfunktion ist geschlossen.