AI4AS is the Artificial Intelligence for Ancient Scripts mini-conference, co-located with DH2026, the 36th annual conference of the Alliance of Digital Humanities Organizations (ADHO) taking place 27–31 July 2026 in Daejeon, South Korea.
AI4AS2026 focuses on the advantages and risks of using AI to engage with ancient scripts and will be held entirely online.
Overview
Generative AI has created unprecedented opportunities for interacting with ancient sources. The ability to query, summarise, or translate specialised material using pre-trained or general-purpose models has rapidly expanded engagement: students, researchers, and non-specialists can now access materials that once required years of linguistic training. Yet the widespread availability of AI tools contrasts sharply with the limited number of specialists who understand how these systems are constructed, how their training data shape their behaviours, and where their fundamental limitations lie. This gap creates conditions in which errors or distortions readily propagate, often unnoticed.
For ancient low-resourced languages, these risks are magnified. The corpora on which these languages survive are inherently small data: broken tablets, damaged inscriptions, and disparate manuscripts. Their scripts (cuneiform, hieroglyphic, hieratic, demotic, sinographic, kana/hanja systems) are structurally complex, polyvalent, and highly context-dependent. Their semantic fields are embedded in cultural, religious, and historical frameworks that models cannot infer from predominantly modern Western training data. As a result, AI outputs frequently flatten nuance, erase ambiguity, and impose linguistic or cultural norms unsuited to the material.
AI4AS directly addresses DH2026's theme of Engagement by scrutinising the tension between expanding accessibility and maintaining interpretive rigor. Engagement-driven uses of AI, such as translation for public audiences, summarisation for classroom use, or automated paraphrasing for introductory materials, may unintentionally reinforce oversimplification or revive outdated scholarly assumptions. When LLMs impose modern Mandarin grammar onto Old Chinese or collapse the polysemy of Ancient Egyptian and Akkadian lexemes, they do more than make mistakes: they generate interpretive narratives that appear coherent, authoritative, and accessible, while masking the underlying instability of the evidence. Such distortions directly affect pedagogy and research, shaping how ancient cultures are understood, represented and ultimately taught and learnt.
The conference considers these risks not as arguments against AI, but as invitations to think critically about engaged scholarship. It asks:
- How can scholars promote accessibility without compromising methodological principles?
- How can emerging technologies support, rather than erode, the interpretive traditions of philology, epigraphy, and historical linguistics?
- Can multilingual scholarly communities collaborate to set discipline-specific boundaries and best practices?
- What forms of digital literacy are necessary for researchers, educators, and students to engage responsibly with AI?
This event gathers experts across diverse ancient-language traditions to highlight structural vulnerabilities shared across fields, while remaining attentive to cultural and linguistic specificity, and to build a shared understanding of responsible AI adoption within Ancient World Studies. Through a series of presentations, collaborative discussions, and brainstorming sessions, this mini-conference aims to:
- Develop a clearer understanding of the methodological vulnerabilities of low-resourced ancient-language corpora in AI contexts.
- Foster field-specific awareness of how LLMs may distort primary and secondary sources.
- Identify strategies for balancing accessibility with rigor in teaching, research dissemination, and public outreach.
- Produce a collaboratively authored set of responsible use guidelines to support transparent, critical, and disciplined engagement with AI in ancient languages.