Congruence Engine investigations

Logo

Documentation, data and code

Uncovering Positionality in data-driven/AI cultural heritage pipelines

Short summary

This investigation aims to explore and develop a framework to uncover and assess positionality as it emerges in a series of data/AI-driven pipelines within the social machine of the Congruence Engine.

Positionality is the culmination of a person’s personal experiences, cultural background, social context, and political environment. It serves as a lens through which they interpret and understand the world. Positionality, and how it embeds itself in processes and systems produced and performed by individuals, is the root for bias in our data and algorithm, and,thus, plays a significant role in the entire data lifecycle in digital cultural heritage pipelines, from the formulation of standards and ontologies, the collection and interpretation of data, to the development of data-focused AI tools, such as ML models. However, positionality is often overlooked in the context of data collection, analysis, and algorithm development. Without an awareness of positionality, the ongoing discussion about bias in data-driven approaches, especially AI and ML pipelines, remains constrained, as merely augmenting the dataset doesn’t de-bias data and models.

In this investigation I will advocate and explore the framework for a paradigm shift towards positionality-aware data/AI -driven digital cultural heritage pipelines. By adopting a ‘walking alongside’ participatory research technique, in this investigation I will ‘shadow’ two ongoing CE investigations and use them as case-studies in order to unveil and analyze the subtle ways in which positionality infiltrates the pipeline of these investigations, often going unnoticed. By embracing positionality-aware data/AI- driven workflows, I seek to place the human and social element of the social machine (human-in-the-loop) at the forefront of our data-driven processes, while transcending the limitations of traditional bias mitigation strategies and fostering a more inclusive, responsible and equitable landscape for data/AI-driven digital cultural heritage.

By proposing and designing a positionality-aware framework, I seek to explore the following areas:

  1. Challenge the traditional model of assessing bias found in individual areas of the pipeline, that is in the data or the algorithm. This approach reproduces an outdated and rather limiting way of doing digital scholarship, presupposing that the data is independent from the algorithm or the process /systems or, more importantly, the human agent(s), and the other way around. Especially with AI pipelines, we are able to understand how deeply interrelated all the parameters/factors are. Thus a positionality aware framework offers a modular, comprehensive way to investigate and assess such interrelations.

  2. Place the human and social element of the social machine (human-in-the-loop) at the forefront of our data-driven processes, by adopting a three-fold approach on < people, data, systems>. I seek to introduce a more humane perspective within the pipeline concept and foster a more inclusive, responsible and equitable landscape for data/AI-driven digital cultural heritage.

  3. Suggest that it is way more important to acknowledge and reveal positionality than mitigate bias. While current bias mitigation strategies are focusing on locating and correcting obvious or visible gaps or errors, a positionality aware framework helps us unveil and analyze the subtle ways in which positionality infiltrates the pipeline, often going unnoticed.

  4. Explore the modular, complex & interconnected nature of the organisational concept of the ‘pipeline’ in digital cultural heritage scholarship.

By embracing a positionality-aware AI framework we aim to transcend the limitations of traditional bias mitigation strategies and foster a more transparent and equitable landscape for AI in digital cultural heritage. Such a framework, focused on continuous assessment and evaluation of the fit between the positionality embedded throughout the AI pipelines and the scenarios within which it is deployed, promotes the values of responsibility, transparency and accountability in the AI ecosystem.

People

Anna-Maria Sichani Conceptualization, Methodology, Data curation, Formal analysis, Visualisation, Writing

Outputs

Read here the full report.

Access here the positionality-aware framework cards.

Licence

This work is licensed for permissive re-use under a Creative Commons Attribution 4.0 License - CC BY 4.0.