We have the pleasure of hosting four institute lectures at DHSI 2025, hosted on the Université de Montréal’s campus in room B-2245, 3200 Jean-Brillant Street.

2025 institute lectures will include talks by two CRIHN members and two international speakers: Ichiro Fujinaga (McGill University), Amanda Lawrence (RMIT University), Marcello Vitali-Rosati (Université de Montréal), and Béatrice Joyeux-Prunel (Université de Genève).

You can see the previous DHSI institute lectures on our legacy website (2024 here and the prior years here).

2025 Institute Lectures

Ichiro Fujinaga (McGill University), “On the Virtues of Lazy Machines”

Monday, May 26, 9:00 am-10:30 am

In an era dominated by resource-intensive artificial intelligence, “lazy machines” offer a sustainable alternative—flexible, collaborative systems that defer computation and minimize upfront effort. This talk explores the virtues of such machines through two music information research case studies. First, in Optical Music Recognition, we incrementally teach machines to recognize music symbols through human-in-the-loop interaction, combining Convolutional Neural Networks with a lazy learning method using the k-Nearest Neighbour classifier. Second, in the LinkedMusic project, we adopt a “lazy searching” strategy, integrating diverse music datasets into an RDF graph without schema alignment. At query time, Large Language Models translate natural language queries into SPARQL, shifting the computational and organizational burden to the moment of need. Embracing this ‘lazy’ approach challenges conventional AI, offering a more adaptable, ecological, and human-aided machine learning model.

Amanda Lawrence (RMIT University), “Wikimedia, LLMs and Information Ecosystem Observability”

Friday, May 30, 12:00 pm-2:00 pm

Although the internet is dominated by massive commercial platforms that are increasingly operated like walled gardens, community led initiatives such as Wikimedia, open access publishing, open source software, open stacks and open source LLMs built by research and community sectors continue to carve out a space in the digital public sphere and digital public infrastructure. Yet it is challenging to understand the relationship between the two given the lack of transparency, data sharing and access from commercial platforms. This relationship is complex and often symbiotic and dynamic as new technologies and business models emerge. Generative AI systems have trained extensively on Wikimedia data but are they now driving traffic away from Wikimedia in search engine result pages (SERPs), and chat results? Will the use of LLMs ‘pollute’ Wikipedia for future training of LLMs? To understand these issues we need new ways of accessing data from digital platforms such as Google Search and GenAI systems. Observability of digital data, platforms and services is critical to our capacity to research and analyse their impact across the information environment and particularly the intersection with digital public goods and digital public infrastructure such as Wikimedia projects. The Australian Internet Observatory and other research infrastructure being developed by researchers in the social sciences and humanities are providing new ways of accessing data from digital platforms, including the use of LLMs as both tool and object of study. This research infrastructure is essential for digital research and can play a key role in supporting the wider digital public infrastructure and a healthy information ecosystem. This paper will explore these tools and how they can be used to provide insights into LLMs and Wikimedia.

Marcello Vitali-Rosati (Université de Montréal), “A Formal Definition of Creativity: LLM, Softmax and Temperature”

Monday, June 2, 9:00 am-10:30 am

“Machines can’t be creative! They can’t create anything new!” This is an idea already mentioned by Turing in his 1950 article (Computer Machinery and intelligence, the “Ada Lovelace objection”) and which recurs notably in the recent (2021) seminal article “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”. But what does “creative” mean? What if the question wasn’t whether or not a machine can be creative, but rather to give one or more formal and unambiguous definitions of creativity? In my presentation, I will analyze this hypothesis by focusing on the notion of temperature as implemented in the activation functions (softmax) of LLMs

Béatrice Joyeux-Prunel (Université de Genève), “What Can Data Do for Art History?”

Friday, June 6, 12:00 pm-2:00 pm

Digital tools have revolutionized the way we access, analyze, and visualize art. But what happens when we apply computational methods to a discipline rooted in visual interpretation, close case studies, historical nuance, and material culture? This talk introduces the field of digital art history, exploring how digitization, databases, and machine learning are reshaping our understanding of artistic production, circulation, and value. Through concrete examples—from mapping exhibition networks and revisiting social art history, to tracing the visual “contagion” of motifs—we’ll examine how data can raise new questions just as much as it can answer established ones. Along the way, the talk reflects on the epistemological and ethical challenges of working with digitized images, biased archives, spatiotemporal visualizations, and AI-generated image apparitions. It also raises a broader concern about the autonomy of “digital art history”—and, by extension, of the digital humanities more generally: that we might lose sight of meaning and critical research questions, simply by spending too much time circling around code and methods.

Ce contenu a été mis à jour le 5 November 2025 à 16h14.