Context-Based Reasoning on Privacy in Internet of Things

Nadin Kökciyan, Pinar YolumIJCAI 2017 · DOI

TL;DR

This paper proposes a novel approach for IoT devices to autonomously reason about privacy by inferring their current context and applying context-specific norms. It leverages argumentation to enable agents to decide whether to share information, addressing the scalability challenges of traditional privacy policies in the heterogeneous and dynamic IoT landscape.

What Problem Does This Paper Address?

The Internet of Things (IoT) is rapidly expanding, connecting a vast number of diverse devices that collect and share information. This proliferation raises significant privacy concerns: it's unclear what data is shared, with whom, for what purpose, and under what conditions. Traditional privacy policies, typically established between a user and a single entity, are not scalable to the IoT environment due to its dynamic nature (devices appearing and disappearing), sheer scale (requiring countless policy agreements), and heterogeneity (devices from different vendors with varying capabilities and trustworthiness). The authors argue that instead of relying on external policies, each IoT entity should autonomously reason about privacy based on its context and the information it gathers.

Methods

The core of the proposed approach is to represent each IoT entity as a software agent. These agents perceive their environment, communicate with other agents, and reason about the information they possess. The key innovation lies in how agents infer their "context" and make privacy decisions:

  • Contextual Integrity (CI) Adaptation: The authors adapt Helen Nissenbaum's theory of contextual integrity. Instead of focusing on user roles within a single organization, they define contexts using relationships between entities, which is more suitable for the distributed nature of IoT.
  • Dynamic Context Inference: Agents can be in multiple contexts simultaneously, with varying degrees of belief. They infer active contexts based on incoming information and their existing knowledge.
  • Argumentation Framework: The paper employs an argumentation-based approach (specifically, the ASPIC framework) to enable agents to reason about their knowledge base (facts and norms) and belief base (information gathered from other agents).
  • Decision Making Process: When an agent receives a request to share information, it first identifies its active contexts. It then generates questions to gather missing information from other agents, using a trust model to weigh the reliability of information from different sources. Finally, it uses the argumentation framework to derive the strongest decision (share or not share) based on the accumulated evidence and context-specific norms.

Datasets

Not discussed in the paper. The paper focuses on a theoretical model and demonstrates its applicability through a running IoT scenario.

Key Insights and Ideas

  • Autonomous Privacy Reasoning: The most significant insight is shifting the burden of privacy decision-making from static, external policies to dynamic, autonomous reasoning by individual IoT entities. This is crucial for scalability.
  • Context as the Key: The paper highlights context as the central element for making privacy decisions. By understanding the situation an entity is in, it can better determine what information is appropriate to share.
  • Relationships over Roles: Adapting CI to focus on relationships rather than strict roles is a practical and necessary step for modeling the complex interactions in IoT environments.
  • Argumentation for Robustness: Using argumentation provides a structured way to handle conflicting information, weigh evidence, and arrive at a justified privacy decision, even when faced with uncertainty and incomplete data.
  • Trust as a First-Class Citizen: Explicitly incorporating trust values for other agents is essential for agents to effectively process and evaluate information in a distributed system.

Key Results

The paper demonstrates the approach through a running example involving a surveillance camera, Alice, and Bob. The camera agent successfully infers Alice's potential emergency context by gathering information from other agents (Bob, Alice's home device, and the police). It then uses this context and its norms to make a privacy decision, ultimately deciding to share Alice's footage because the inferred emergency context overrides the default "no-share" norm for work contexts. The paper also presents theoretical results (Theorems 1 and 2) proving that the proposed algorithm can always reach a sound privacy decision given a world view.

Limitations and Future Work

  • Trust Model Semantics: The paper acknowledges that the specific semantics of the getTrustValue function are not detailed, assuming it evolves over interactions.
  • Contextual Relationships: The authors suggest future work to capture relationships between contexts, allowing for inheritance of norms and relations, which would further enhance scalability and adaptability.
  • Norm Learning: Enabling the system to learn norms over time is identified as a crucial direction for improving scalability.
  • Computational Complexity: While not explicitly stated as a limitation, the reliance on argumentation and multi-agent communication implies potential computational overhead that would need to be managed in real-world deployments.

Summary updated

BibTeX
@inproceedings{kokciyan2017context,
  author = {Nadin Kökciyan and Pinar Yolum},
  title = {{Context-Based Reasoning on Privacy in Internet of Things}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2017},
  doi = {10.24963/ijcai.2017/660},
}
Figure