Arches and Artificial Intelligence

This webpage provides an overview of how the Arches Project approaches engagement with AI, the potential of Arches software tools to improve AI performance and accuracy, and examples of practical AI usage in conjunction with Arches. On this page, “AI” is used as a general term referring to artificial intelligence in a variety of forms, including machine learning, deep learning, generative AI and LLMs (large language models).

Arches Project Engagement with AI

The Arches Project governance committees – Steering, Technical Advisory and Community Advisory – regularly consider how AI should be dealt with in the Arches Project, including the establishment of related guidelines and procedures.

The Arches Project follows the following principles in assessing and engaging with AI:

  • Mission first: support for cultural heritage internationally is foremost in all matters.
  • Human-centered: use of AI technologies should support or enhance human efforts, not replace them
  • Governed use: potential AI usage in Arches code is a subject of continuous evaluation by Arches Project Governance, including any updates to policy
  • Trust and transparency: AI-related decisions are made in the open and with an invitation for feedback from the community

At this time, the official Arches Project software codebase does not directly include AI agents. However, given that Arches is open-source software, anyone can make changes to their own instance of the software code, which incorporates AI tools.

The Arches Project acknowledges the inevitable use of AI within the sphere of the community, but also that an overly prescriptive statement on AI use would be ever-changing and quickly outdated. Productive and ethical usage of AI in the Arches Project depends on ensuring human accountability. The Arches Project Contributor Guidelines include recommendations intended to preserve human accountability in AI-assisted code production.

The Potential of Arches to Improve AI Performance

Arches offers the possibility of substantially enhancing AI performance, if its data structuring capabilities are fully utilized.  Arches can support smart data systems by transforming raw data into structured, contextualized, and actionable information. This can assist AI agents in producing more relevant and meaningful results. This is made possible through two essential features of how Arches data can be structured:

  • ontology-based semantic data structure 
  • ontology-based application of controlled vocabularies

Arches can provide semantic, relevant, and machine-actionable cultural heritage data by supporting semantic data modeling (through application of ontologies such as the CIDOC-CRM), applying semantic, concept-based thesauri, and publishing Linked Open Data (as RDF and JSON-LD). Through its ability to build deep relationships between data, when used to its full potential Arches can provide a structured, semantic framework that imparts the necessary context and data standardization that AI systems and data-processing technologies require to function effectively.

Like any system, however, the potential for an individual implementation of Arches to improve AI performance can only be realized to the degree that the above features are utilized to generate smart data. Applying an ontology within the Arches data structure is optional. Because Arches provides flexibility in data modeling, data can be structured within an implementation in a way that leads to suboptimal performance. If the data is made publicly available and open to ingest by AI agents, the results of data analysis or processing by AI are only as reliable and accurate as the data ingested.

Practical AI Usage with Arches

There are several ways in which software developers, implementers, and researchers have utilized AI tools in conjunction with or in support of Arches. Some of these include:

  • Providing initial code review and feedback prior to human review and approval, such as in identifying bugs in Arches code
  • Assistance with the creation of new Arches resource models
  • Assistance with data cleanup and mapping to support data ETL (extract, transform, load) in the process of importing data into Arches

There are additional potentially productive uses of AI in conjunction with Arches that, when undertaken responsibility, align with principles outlined above.  The following are additional examples of how AI can provide benefit with minimal risk:

  • Support to identifying Arches security issues
  • Support generating Arches-related documentation
  • Support to data extraction from Arches implementations
  • Summarizing data and other content managed within Arches implementations

Last Updated on September 8, 2026