Keynote Abstract: Machine Learning in Conflict Studies: Reflections on Ethics, Collaboration, and Ongoing Challenges

Kristine EckACL 2021

TL;DR

This keynote abstract argues that while machine learning (ML) offers revolutionary potential for analyzing vast conflict-related data, scholars must prioritize ethical considerations and robust collaboration. It highlights the challenges of applying ML to messy, non-systematic conflict data and emphasizes the need for interdisciplinary approaches to ensure analytical utility and avoid perpetuating biases.

What Problem Does This Paper Address?

The paper addresses the growing application of machine learning (ML) in conflict studies, a field traditionally reliant on qualitative methods. While ML promises to unlock insights from massive datasets, its application to conflict data presents unique challenges. The core problem is how to ethically and effectively leverage ML for conflict analysis, considering the inherent complexities of conflict data (e.g., non-systematic generation, idiomatic language, diverse actors) and the potential for ML models to inadvertently reproduce or obscure political biases. The authors also highlight the need to bridge the gap between ML capabilities and the theoretical interests of conflict researchers, ensuring that ML-produced data is analytically useful.

Methods

This paper is a keynote abstract, not a research paper detailing specific methods. However, it discusses the application of machine learning and natural language processing (NLP) techniques to conflict data. It mentions the use of ML for "conflict event extraction" and analyzing other data types like social media and government policy documents. The abstract implicitly suggests that standard ML and NLP approaches may need adaptation for the unique characteristics of conflict-related text. It also advocates for collaborative, interdisciplinary methods to anchor ML projects in the theoretical interests of political violence scholars.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Ethical Imperative: Scholars have an ethical duty to "get it right" when using ML for conflict analysis, which extends beyond model accuracy to include data quality and the avoidance of bias.
  • Data Complexity: Conflict data is often messy, non-systematic, lexically idiosyncratic, and generated by diverse actors, making it ill-adapted to standard NLP and ML tools.
  • Beyond Event Extraction: While conflict event extraction is a common ML application, scholars are increasingly turning to ML for analyzing other aspects of the conflict environment, such as hate speech, public opinion, and policy documents.
  • Interdisciplinary Collaboration is Crucial: Addressing the challenges of ML in conflict studies requires collaborative projects that integrate expertise from both ML/NLP and conflict research.
  • Bridging Theory and Practice: ML classification projects must be grounded in the theoretical interests of political violence scholars to ensure their analytical utility. There's a need for better ontologies that reflect these interests.

Key Results

Not discussed in the paper.

Limitations and Future Work

The abstract acknowledges that many challenges remain in applying ML to conflict studies. It emphasizes the need for:

  • Conceptual Development: Creating better ontologies for classification that align with the interests of conflict researchers.
  • Technical Development: Adapting ML and NLP techniques to better handle the complexities of conflict data.
  • Interdisciplinary Collaboration: Fostering partnerships between ML experts and conflict scholars to ensure the ethical and effective use of these technologies.
  • Anchoring in Theory: Ensuring that ML-driven analysis is guided by the theoretical frameworks of political violence research.

Summary updated

BibTeX
@inproceedings{eck2021keynote,
  author = {Kristine Eck},
  title = {{Keynote Abstract: Machine Learning in Conflict Studies: Reflections on Ethics, Collaboration, and Ongoing Challenges}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2021},
}
Figure