15 ppl attended

Invited presentation

• Speaker: Gabriele Gattiglia
• Topic: AI and Coding in Archaeology

Gabriele introduced the basic structure and functioning of large language models (LLMs) and discussed their growing impact on how archaeological research and coding are conducted. He emphasised that AI is not a substitute for expertise, but rather a tool that can accelerate workflows while also amplifying existing methodological challenges.

A central part of the presentation focused on transparency and reproducibility. Since prompts are not reproducible, they pose significant problems for documentation and accountability. Gabriele therefore argued that the prompting process itself should be documented, as it directly shapes both code and interpretation.

Matteo Tomasini compared this to stochastic simulations, pointing out that both require an awareness of probabilistic outcomes and non-deterministic behaviour.
Zack Batist referred to the experiment “What does Canadian History look like to the machine?” (Electric Archaeology, 2024) as an example of how training datasets can reveal the biases and limitations of machine knowledge.

Gabriele reflected on the implications of AI-assisted coding for transparency, reproducibility, and maintainability. While AI tools can assist with documentation or refactoring, they also risk promoting “vibe coding” — coding guided by intuition and AI suggestions rather than explicit reasoning.
Participants discussed the potential of RMarkdown-based workflows to counteract the “black box” effect by integrating code, data, and interpretation in one transparent structure.

At the end of his talk, Gabriele presented the MAIA COST Action, an interdisciplinary initiative exploring AI in archaeology and the humanities. He invited participants to collaborate on research about AI-assisted coding, data collection, and the development of standards for transparency and reproducibility.

Discussion highlights

Documentation and transparency

• Consensus that prompt documentation might help reproducibility.
• AI’s tendency to produce confident but sometimes incorrect results makes open code and explicit workflows more important than ever.
• The group agreed that archaeological coding should include documentation of how AI was used, not only the final code.

Teaching and education

• Petr and Jim asked how AI can be responsibly integrated into archaeological teaching.
• Gabriele stressed that students must learn to be critical, reflective users of AI.
• Sonja compared the current introduction of AI to earlier methodological transitions such as GIS and radiocarbon calibration.
• Participants noted a generational divide in how AI is perceived — younger scholars are generally more accepting, while senior researchers tend to be more cautious.
• Matteo framed this as part of a broader humanistic debate on technology and interpretation.

Software practice and sustainability

• Participants observed that challenges associated with AI (e.g. copy–paste coding, opaque dependencies) are long-standing issues in software development.
• LLMs might serve as an opportunity to learn and reinforce good software practices such as code review, pair programming, and proper commenting.
• Archaeological coding should pursue conceptual and interpretive goals, not merely technical functionality.
• AI-assisted refactoring and documentation were seen as potentially beneficial when compared to no documentation at all.

Broader considerations

• The group discussed whether archaeological research should also explore AI-assisted data collection, and what methodological standards would be needed.
• Zack suggested building a community of practice to define standards for AI-assisted and reproducible programming in archaeology.
• A first step could be to revisit published papers including code to evaluate and develop shared guidelines for transparency, sustainability, and reuse.

Conclusions

• AI represents not only a technical but also an epistemological and pedagogical challenge for archaeology.
• Prompting should be documented as part of the research process.
• Transparency, openness, and reproducibility remain core principles.
• Training and community standards are crucial to ensure responsible and sustainable integration of AI tools.

Next SIG Meeting: Friday, November 7, 2025

• Details and topic will be announced on the mailing list.
• Possible follow-up discussion on documentation standards and AI-assisted coding practices.
• Other topics might include an maintain-a-thon, proposed by Zack.