Facets of Software Doping

Gilles Barthe, Pedro R. D'Argenio, Bernd Finkbeiner, Holger Hermanns

TL;DR

This paper introduces and explores the concept of "software doping," where software intentionally behaves in a way that benefits a manufacturer or vendor at the expense of the user or society. The authors propose initial formal characterizations to distinguish doped software from clean software, aiming to lay the groundwork for specifying and enforcing regulations against such practices.

What Problem Does This Paper Address?

The paper addresses the growing phenomenon of "software doping," where software, particularly embedded software, is designed to exhibit behavior that is not in the best interest of the owner or society. This can manifest in various ways, such as printers refusing to work with non-manufacturer ink cartridges, or vehicles intentionally polluting more outside of specific test conditions. While these practices are becoming more prevalent with the increasing embedding of software in everyday devices, there's a lack of systematic investigation and formalization from a software engineering perspective. The authors aim to bridge this gap by providing a conceptual account of software doping and proposing formal definitions to identify it.

Methods

The paper takes an informal, conceptual approach to defining software doping. It starts by reviewing real-world examples of software doping to build intuition. Based on these examples, the authors propose a series of characterizations for what constitutes "clean" software, which by negation, defines "doped" software. These characterizations evolve from a simple notion of consistent behavior across parameter values to more nuanced definitions that consider "reasonable distances" from standard inputs and outputs, especially in the context of continuous or imprecise physical quantities. The core method is to define clean software through its expected behavior and then identify doping as any deviation from this clean behavior that serves a hidden interest.

Datasets

Not discussed in the paper. The paper relies on real-world examples and synthetic examples inspired by these cases rather than empirical evaluation on specific datasets.

Key Insights and Ideas

  • Software Doping is Intentional and Harmful: The core idea is that software doping is not accidental but a deliberate design choice to favor a specific party (manufacturer, vendor) over the user or society.
  • Beyond Simple Bugs: Doping is distinct from typical software bugs; it's a feature designed to achieve a specific, often detrimental, outcome.
  • User vs. Societal Interest: Software doping can harm either the user's interest (e.g., printer ink) or society's interest (e.g., emissions scandals), or both.
  • The Challenge of Intentionality: Determining whether a behavior is intentional doping or a genuine glitch is difficult, leading the authors to focus on observable behavior rather than the manufacturer's intent.
  • Formalizing "Clean" Software: The paper's main contribution is proposing formal characterizations of clean software, which can then be used to identify doped software. These characterizations evolve to handle different scenarios, including those with continuous physical quantities.
  • "Reasonable Distance" as a Key Concept: For systems involving physical quantities, the concept of "reasonable distance" from standard inputs/outputs is introduced to define acceptable deviations, distinguishing them from intentional doping.

Key Results

The paper does not present quantitative results in the traditional sense. Its key "results" are the proposed characterizations of clean software:

  • Definition (2): A program is clean if for every standard parameter, it exhibits the same visible outputs when supplied with the same inputs. (This is a basic consistency check).
  • Definition (3): A program is clean if for every standard parameter, it exhibits the same visible outputs when supplied with any possible input complying with a given standard. (This is a weaker notion, focusing on consistency within a standard domain).
  • Definition (4): A program is clean if for every standard parameter, whenever it is supplied with any input that deviates within "reasonable distance" from a given standard input, it exhibits a visible output which does not deviate beyond a "reasonable distance" from the specified output corresponding to such standard input. (This is the most nuanced definition, accounting for physical systems and acceptable tolerances).

These definitions serve as the primary outcome, providing a conceptual framework for identifying software doping.

Limitations and Future Work

  • Informal Definitions: The authors acknowledge that the proposed definitions are still informal and that a rigorous formalization is needed.
  • Intentionality: While they acknowledge the difficulty of proving intent, their current definitions focus on observable behavior, which might not capture all nuances of intentional doping.
  • Formalization with Hyperproperties: The authors are actively working on formalizing these definitions, expecting them to fall under the class of "hyperproperties." This formalization is intended to help specify contracts and regulations for embedded software, enabling rigorous discrimination between doping and acceptable deviations.
  • Analysis Methods: The formal characterizations are expected to provide foundations for developing analysis methods (verification or testing) to detect doping.

Summary updated

Abstract

This paper provides an informal discussion of the formal aspects of software doping.
BibTeX
@inproceedings{barthe2016facets,
  author = {Gilles Barthe and Pedro R. D'Argenio and Bernd Finkbeiner and Holger Hermanns},
  title = {{Facets of Software Doping}},
  booktitle = {ISoLA},
  year = {2016},
  doi = {10.1007/978-3-319-47169-3\_46},
}
Figure