This keynote explores the impact that transformer-based machine learning brings to the interpretive work of historians. As historians increasingly encounter vast amounts of digitized and born-digital sources, the challenge has shifted to developing strategies for making sense of large, complex collections with the nuance that historical inquiry demands. The discussion begins with an earlier phase of my research, which aimed to engage in digitally mediated multiscale exploration (“digital (re)reading”) through graph queries and data reuse. Using structured and relatively unambiguous sources, such as biographical data modeled in Neo4j, this phase underscored the potential of digital historical research to uncover latent structures and reveal surprising connections in a manner that preserves the historian’s interpretive agency.

Building on this foundation, I then turn to the present, where my team and I are focused on leveraging large language models (LLMs) and vision-language models (VLMs) to assist with “algorithmic reading” across heterogeneous and semantically complex corpora. This next phase of inquiry explores the affordances of LLMs and VLMs for conducting semantic, stylistic, sentiment, and multimodal analysis, moving decisively beyond the limitations of keyword-based search and frequentist approaches. Whereas the earlier digital macroscopes allowed users to zoom in and out of structured datasets, transformers enable engagement with more affectively and rhetorically rich sources, such as memorials, petitions, contracts, philosophical treatises, ritual guidelines, and poetry.

Finally, I introduce the modular artificial intelligence (AI) framework developed in the DeepPast project, which promotes the use of pluggable, task-specific components running on low-power hardware rather than a hyperscale, monolithic, general-purpose system. The DeepPast architecture supports varying interpretive modes in a flexible environment where the historian purposefully engages in conversation with an AI assistant and research partner—one capable of offering critique, reframing questions, and proposing alternative perspectives. The lecture concludes with a set of guiding principles designed not only to keep the human in the loop but also to produce AI-assisted historical research marked by greater interpretive sophistication.

The institutions that we have long relied on to sustain knowledge, higher education, and even data are under attack. As digital humanists and citizens of the world, we have an important choice to make: do we keep reproducing the extractive and colonial systems in which we work, or should we build something else? Risam will argue for an approach to digital humanities that is grounded in justice, tying access to accountability, repair to care, and scholarship to solidarity. The choices we make about digital accessibility, inclusive platforms, and the role of equity and diversity in our work will determine which types of knowledge persist in a world unmade and whose voices and stories survive for the future.