Embedded Files
Frédéric Kaplan
Frédéric Kaplan
Associate Professor, Digital Humanities Laboratory, EPFL (École polytechnique fédérale de Lausanne)
Associate Professor, Digital Humanities Laboratory, EPFL (École polytechnique fédérale de Lausanne)
Frédéric Kaplan is Professor of Digital Humanities at EPFL, where he directs the Digital Humanities Laboratory, and President of the Time Machine Organisation. For more than fifteen years, he has pioneered research on transforming large-scale historical archives, maps, images and newspapers into structured spatiotemporal representations, forming what he calls the Big Data of the Past. His current work investigates how this historical infrastructure can serve as a training ground for predictive artificial intelligence, opening what he describes as the age of computational futures. His research argues that large-scale computational representations of the past will become a core infrastructure for artificial intelligence in the twenty-first century.
Time Machines at World Scale: From the Big Data of the Past to Computational Futures
Time Machines at World Scale: From the Big Data of the Past to Computational Futures
Over fifteen years, the digitisation of archives, maps, images, newspapers and administrative records has turned the past into data at an unprecedented scale. Using methodologies developed within the Time Machine project, a network of hundreds of institutions has demonstrated that the scattered traces of history can be progressively reassembled into a structured patiotemporal representation, forming what we call the Big Data of the Past.
Today, around eighty Local Time Machines already reconstruct and computationally represent the past of individual cities. We can reach world scale through a fully distributed system of representation and computing, where each city or region keeps and runs its own Time Machine while interoperability emerges from shared protocols, much as the web connected independent servers into a single information space. The mission of the Time Machine Project is to make this global distributed infrastructure a reality.
Once a sufficient fraction of the world's past has been organised this way, the system undergoes a phase transition. The past becomes a testable laboratory for prediction: because its outcomes are already known, we can learn to forecast from incomplete evidence and measure how well we did, then turn the same models toward the open future. We thus enter the age of computational futures, in which Time Machines, as a detailed spatiotemporal memory from which predictive models of the future can be trained, become crucial strategic assets. This talk is about the world that emerges once enough historical data is gathered and that threshold is crossed, and why Time Machines may become one of the defining infrastructures of the twenty-first century.
Geoffrey Rockwell
Geoffrey Rockwell
Professor of Philosophy and Digital Humanities, University of Alberta, Canada
Professor of Philosophy and Digital Humanities, University of Alberta, Canada
Geoffrey Martin Rockwell is a Professor of Philosophy and Digital Humanities at the University of Alberta, Canada. He is a Canada CIFAR AI Chair and also a Fellow of the Alberta Machine Intelligence Institute. He received a B.A. in philosophy from Haverford College, an M.A. and Ph.D. in Philosophy from the University of Toronto and worked at the University of Toronto as a Senior Instructional Technology Specialist. He has published and presented papers in the areas of artificial intelligence and ethics, philosophical dialogue, textual visualization and analysis, humanities computing, instructional technology, computer games and multimedia. He was the project leader for the CFI (Canada Foundation for Innovation) funded project TAPoR, a Text Analysis Portal for Research, which has developed a text tool discovery portal at tapor.ca. He has published a book Defining Dialogue: From Socrates to the Internet with Humanity Books and a book titled Hermeneutica (with Stéfan Sinclair) with MIT Press.
Care and Repair: Digital Humanities Approaches to Artificial Intelligence
Care and Repair: Digital Humanities Approaches to Artificial Intelligence
There is significant skepticism about the value and safety of Artificial Intelligence (AI) among humanists. There are concerns about the safety of using frontier models given the number of recent hacking events. There are concerns about the epistemological effects of offloading cognitive tasks to AIs, especially research and writing tasks. There are also significant opportunities for enhancing Digital Humanities (DH) research using AI. In this presentation I will first talk outline some of the dangers, and then I will talk about ways to take care when using using AI in DH projects. In this context I will talk about some projects that I am involved in. Above all, I hope to convince you that humanists in general and Digital Humanists in particular should be engaged in this liminal moment. We should bring our practices of care and repair to the challenges ahead.
Miguel Escobar Varela
Miguel Escobar Varela
Associate Professor, English, Linguistics and Theatre Studies, University of Singapore
Associate Professor, English, Linguistics and Theatre Studies, University of Singapore
Miguel Escobar Varela is an Associate Professor at the National University of Singapore and Co-director of the Centre for Computational Social Science and Humanities. He also serves as an Associate Editor of Computational Humanities Research and co-convenes the minor in Human-AI Systems. He studies the changing landscape of Southeast Asian cultural heritage by combining fieldwork with computational methods (such as NLP, computer vision and network analysis). His aim is to understand how the production and reception of cultural forms have changed over time, in areas such as the performing arts and print media.
How to ngoprek AI models for cultural specificity
How to ngoprek AI models for cultural specificity
AI models don’t always capture the cultural nuances that we care about in the humanities, particularly when we work with under-represented artistic and intellectual traditions. But pre-trained, open-weight models offer us a tantalizing avenue for culturally specific, automated workflows. Through supervised fine-tuning (SFT), we can recalibrate the weights of a model to achieve high accuracy within well-defined tasks. The finetuned models that we get at the end of this process are more transparent and reproducible than proprietary alternatives, and they can be deployed in infrastructure of our own choosing. But the road to this destination is anything but clear: SFT is as much an art as a science. While not seeking to minimize the technical hurdles, this talk will emphasize the ways in which SFT rewards curiosity, deep expertise, and the willingness to tinker (or what Indonesians call ngoprek). A key aspect of SFT is the curation of datasets that represent the variety, ambiguities and nuances of the phenomena we aim to study. This process is often slow, and follows no set path. In other words, it is not so different from many endeavours in the humanities (digital or not). In this talk, I will walk through how my team and I have navigated these journeys while working on tasks such as automated text recognition (ATR) and video segmentation for cultural traditions in the Malay-Indonesian world.
Page updated
Google Sites
Report abuse