By Jim Clifford

It is never fun to watch friends argue. Generative AI has created multiple fractures across our universities and the community of historians. The stakes are high and we are all dealing with the repercussions. I spent a few weeks in August rebuilding my second-year course for the third time. Two years ago, I swapped the traditional research essay for a StoryMap project, thinking that even if the students used GenAI, they’d still learn something making the map. But now, a student with Claude in Chrome could get the LLM to build the whole map without any real engagement. This year I’m trying a hybrid of contract grading and ownership. It was a lot of work and I still worry about what we’ve lost without a traditional essay.

The frustration and anger with the tech companies make it hard to hold a conversation. I’m worried about the future of higher education at the scale it has operated for the past half century. If machines are starting to displace recent graduates in the employment market, then the basic promise of working hard to earn a fulfilling and remunerative career is under threat. I understand the decision to simply refuse: not to engage with this technology and not to pay Anthropic or OpenAI a monthly subscription. I also understand why people are skeptical about any claims about the ability of LLMs. In 2024, OpenAI claimed PhD-level intelligence for models that couldn’t count the number of Rs in “strawberry.” Professors saw the regular failure of LLMs hallucinating historical arguments and citations. But a lot has changed in the past fifteen months, with the emergence of agentic systems, and then in mid-2026, with the launch of the Fable and Astra models. The problem is that both of these systems take investment to learn how to use and high-tier subscriptions. So, the vast majority of historians have not seen how they work.

My proposal here is to engage with those of us who are using the tools to understand where we stand in 2026 so we can collectively start to talk about what needs to change as we move into 2027 and 2028. I am not asking anyone to change their ethical stance, and I acknowledge there are large problems with the current build-out of data centres without enough regulations to mitigate environmental concerns. These issues are real, but I have decided, for myself, that critical engagement is how I can best understand the technology and think through its implications. The profession needs some of us working inside these tools; it does not need all of us to.

The jargon is annoying. “Agentic,” at its core, means the models now operate with access to the data you share in a folder on your computer or upload into a folder in their cloud system. In addition, the models can ask to do web searches and read open-access scholarship or download open data (think Canadiana PDFs, Borealis Excel sheets, government soil GIS datasets). Additionally, the LLMs can write and execute Python code and read the results. Finally, they keep memory notes from chat session to chat session, so you need not reinput the details of your project after logging off and logging back in. This solves a lot of the limitations of the earlier chat interfaces. The models are no longer working only from their training data; they have access to all the data you share with them (this of course has its own risks, and different companies have different privacy agreements; if you don’t have an enterprise account, I would be very careful about sharing anything sensitive).

The second big change came in the spring of 2026. Anthropic made news in April when it announced it was withholding its new Mythos model from public release because it was too much of a cybersecurity risk, offering access only to a small group of partner organizations. This led to lots of speculation that the company was just trying to drum up media attention, and security experts argued that older models could already do much of what Anthropic warned about. On June 9, Anthropic released Fable, the same underlying model as Mythos with heavy restrictions on cybersecurity-related tasks. Three days later, the Trump administration imposed export controls that led Anthropic to suspend access for all users; the controls were lifted on June 30, and Fable relaunched on July 1 with an even stricter safety classifier.

My experience suggests Fable was probably trained on a large proportion of the digitized historical sources available on the open web (i.e. the Internet Archive text collection). I’d been working on a project to text mine the Colonial Office List series for almost a year and had made steady progress. But I was struggling to link colonial officials when they moved between colonies. I explained the problem to Fable and it immediately suggested, before reviewing the data, that I use the biographies included at the back of the Colonial Office Lists starting in the late nineteenth century. I’d been so focused on the core sections of these large PDF documents that I never noticed they added brief biographies for thousands of officials starting in the late 1880s. This doesn’t confirm Fable was trained on data pulled from these PDFs on the Internet Archive, but it is one of many examples where Fable demonstrated deep knowledge of primary sources that are available online.

Furthermore, Fable was trained on a lot of the historiography. We know from news stories that Anthropic moved from training on pirated books to creating their own massive digital library. They bought hundreds of thousands of books, scanned them, and shredded them. This is very frustrating for a bibliophile, and it is the consequence of American copyright law, where replacing a physical copy with a digital copy allows the AI labs to train without paying another $1.5 billion to authors.

Finally, Fable is not nearly as sycophantic as previous models. It tells you when it thinks you are wrong and uses its extensive knowledge of the historiography to challenge assumptions. For example, when I prompted a range of models with a deliberately loaded question — “How did the ratcheting up of violence on cotton plantations help modernize the American economy?” — most of them, including Gemini Pro 3.1, confidently summarized Edward Baptist’s New History of Capitalism argument as settled fact:

“Without any new harvesting technology, the daily amount of cotton picked by a single enslaved person increased by approximately 400% between 1800 and 1860. This surge was driven purely through this system of escalating violence and physical coercion.”
— Gemini Pro 3.1

Fable, by contrast, opened by questioning the premise:

“The claim behind your question comes largely from Edward Baptist’s The Half Has Never Been Told (2014), and it’s worth separating the argument from the evidence, because the argument is still contested.”
— Claude Fable 5.1

It went on to lay out the challenge posed by Alan Olmstead and Paul Rhode, and where the debate currently stands. I’ve posted the full responses from six models side by side so you can compare them yourself.

Where does this leave us? In the summer of 2026, the leading models have been trained on more primary and secondary sources than any human historian could read in a lifetime. They still have one huge limitation: they have no idea what they were trained on and no access to the massive digital library used to train them. When I ask whether Fable was trained on my book, it doesn’t know. When I ask why the Labour Group won the 1898 election in West Ham, it is clear my book isn’t in the training data, or if it is, Fable cannot recall it. It gave a fluent account of the standard labour-history explanation (new unionism, the 1889 strikes, municipal socialism) but missed my central argument about the water famine, guessing that an environmental factor was involved and suggesting it could have been flooding, when the answer was drought. Of course, agentic tools help address this weakness. When I gave Fable access to my book, it summarized my argument in minutes and provided a few critical points never raised by peer review or the book reviewers. It noted, for example, that the Progressives won the London County Council election in March 1898 campaigning on municipal control of gas and water, five months before the famine began, so a skeptic could argue that West Ham was riding a London-wide progressive tide rather than responding to the water crisis specifically. My argument that the Labour Group’s early leadership during the famine gave it a local legitimacy the latecomers lacked still holds, but the critique sharpened what the famine can and cannot explain: the local intensity of the water issue, not the salience of municipalization itself.

This does not mean Fable is a particularly good historian. It still lacks a core human element. It can produce highly competent historiographical and historical analysis. But it doesn’t live in our historical moment. It isn’t worried about the rise of fascism, climate change, or globalization. It isn’t interested in how gender is transforming during our lifetime and drawn to understand how gendered discourse shaped past experiences. Fable has not experienced the transformation of China into a dominant global economic and political power, so it has no reason to try to better understand past moments where China was the dominant Eurasian economy and political power. It could help with any of these questions, and if I prompted it to come up with questions grounded in our current context, it might come up with a similar list. But it has no particular need to understand history and would be just as willing to walk you through a deep dive into string theory or MLB rules.

All of this creates a real problem for our public debates. The frontier systems sit behind expensive subscriptions and steep learning curves, so colleagues who have taken a principled stance against using AI have no easy way to know the scale of what changed this year. Informed refusal has become genuinely difficult, and that is a structural problem with how this technology is being deployed. Good-faith dialogue with people who are critically engaging with these systems is now the best way for those who refuse them to stay informed about what the technology can actually do. Refusers need honest reports from inside the tools, including the failures (LLMs can still fail at the edge of human knowledge; that is the topic for a future blog post). They, in turn, continue to do important work in our universities, pushing back against the uncritical and poorly thought out widespread adoption of these systems. That dialogue only works if both sides stay in the conversation.

AI disclosure: I wrote this piece and used Claude to copy edit (I’m very dyslexic and modern LLMs are transformative for people with language-based learning disabilities). I also used Claude for an adversarial review. I have an enterprise subscription from the University of Saskatchewan. I have no relationship with Anthropic.

Jim Clifford is a Professor of History at the University of Saskatchewan. His research focuses on urban Britain, global commodities, environment and heath. He is a co-editor of Historical Methods: A Journal of Quantitative and Interdisciplinary.

Since publishing my first book in 2024, I’ve turned my attention to gathering material for a second. Provisionally titled A View from the North: Lost Anglo-American Literature on Mexico from the Twentieth Century, I am exploring pulp novels, young adult adventure stories, tales of espionage, tourist guides, and yes, even postcards — in hopes of showing the extent to which the “Good Neighbor” cultural diplomacy of the twentieth century was, in fact, a genuinely intercultural endeavor between the United States and Latin America. Who were the small players — dime store novelists, postcard photographers, and nightclub owners — who got swept up in the large-scale international initiatives between North and South?

The Doris Z. Stone Latin American Library on the campus of Tulane University is an ideal place to examine this particular historical moment. The collection is replete with Mexico travel and tourism ephemera — magazines, hotel brochures, and highway maps — all of which, I believe, serve to better understand how our southern neighboring countries were constructed as attractive vacation destinations during the mid-twentieth century. Mexico especially loomed large in the imagination of US tourists; it was the period when, as historian Mauricio Tenorio shows, Mexico was understood as a “Brown Atlantis” — where revolutionary thought fermented and the Pan-American Highway promised access to exotic adventures. Everyone wanted to meet Frida Kahlo, everyone wanted to visit Teotihuacán, all while dreaming up global politics anew.

Two brochures found in the Mexican ephemera collection tell how Mexico feted itself for an international community.

Panoramic of the city center of Taxco de Alarcón, Guerrero, Mexico. Viewed from the Christ Monument looking southeast, via Wikimedia Commons.

First, is a trifold brochure from arguably Mexico’s most famous nightclub during the second half of the twentieth century — Cantarranas, located in the silver-mining town of Taxco, some two and a half hours outside of Mexico City. As the brochure details, the “Hacienda Cantarranas was built in 1538 by order of Hernán Cortés. The actual construction was done by his son Martin de Cortés.” The estate was used for silver processing. The locale, built with large stones on a hillside, became a nightclub in 1957 under the direction of entrepreneur Antonio “Tony” Reyes Ramírez. He would later become Taxco’s minister of tourism. Cantarranas was known worldwide for its revelry through the 60s and 70s before being acquired by the government of the Mexican state Guerrero in 1983, when it was converted into a culinary school and arts center. In 1992, it was turned over to the Universidad Nacional Autónoma de México as a school for foreign nationals. The hacienda was named “Cantarranas” — that is, literally “singing frogs” due to the loud chorus of frogs living in the area nearby. The brochure obviously plays upon this idea — even if not representative of the real spectacles of the club, the image is fetching.  

The brochure promises tourists that a night spent at Cantarranas includes “flamenco dancing by Estela Calderón,” “Carl Brice directly from Nassau, Bahamas,” “Primitive African Drums by the owner and Manager of Cantarranas Tony Reyes,” and “The Latin American Rhythms of the great pianist Juan García Esquivel.” The nightclub also housed distinct spaces — the “Twilight Room,” the “Patio,” the “Flamenco Room” and the “Mulata Room.” A newspaper article from 1974 also lists the club’s attractions, including “Flying Indian Men,” a discotheque called “The Psychedelic Cave,” bullfight lessons, a piñata party, and calypso songs by Carl Brice. By 1978, another newspaper reports that “Cantarranas is a little rough around the edges.” In 1980, Cantarranas had seemingly lost its sheen, as a newspaper article sarcastically notes “Cantarranas is the kind of place that you wouldn’t miss if you missed it. But worth it if you’re utterly bored.”

Cantarranas Nightclub Brochure. Courtesy of the Doris Z. Stone Collection at Tulane University.

Gone was the moment of Mexico’s mid-twentieth century Miracle. During the 1960s, newspapers still reported how, inside Cantarranas, “the audience moves en masse from room to room as each show finishes.” The brochure from Tulane explains that “among all the night clubs in Mexico Cantarranas was chosen for the filming of an episode of the N.B.C. colour T.V. Series ‘I Spy.’” That part of the brochure isn’t an exaggeration. The club was featured on the February 23, 1966 episode “Return to Glory.” Bill Cosby, playing the part of Alexander Scott — one half of a spy duo — is tipped off by US Intelligence to visit the club specifically, where he rendezvouses with the wife of a deposed president of an unnamed Latin American country. Famed Mexican actress Dolores del Río plays the part of the wife, as Bill Cosby and Robert Culp try to negotiate the democratic leader’s return to his home nation.

At least one similar brochure can be found online.

Surprisingly, the Tulane collection includes one more brochure with a singing frog on it — this one from Pemex or Petróleos Mexicanos, the company formed in 1938 by the Mexican government to sell the nation’s significant oil supply. The brochure includes maps of Mexican highways and invites travelers to explore the best routes. There is no mention of Cantarranas, or Taxco, or even nightclubs. Somehow, a singing frog had become emblematic of sightseeing around Mexico — the exotic sights, the electric nights, the cultural spectacles.

PEMEX Travel Club. Courtesy of the Doris Z. Stone Collection at Tulane University.

Intriguingly, in the decade that saw the beginning of Cantarranas’ end, a Mexican restaurateur — Carlos Anderson — opened in Mazatlán, Mexico the first restaurant of many that he would christen Señor Frog’s. The original locale, opened in 1971, tried to revive some of Cantarranas’ mid-twentieth century swagger, with waiters slinging tequila shots and cajoling spontaneous conga lines in a neon cave.

Thus, Tony Reyes’ “Singing Frogs” bar presaged today’s franchised “Señor Frog’s.” The brochure from Tulane documents how tourists were meant to imagine the delights of mid-twentieth century Mexico — strange, inviting, and culturally savvy — like a crooning amphibian. The Cantarranas and Pemex brochures ultimately serve to illustrate the extent to which commercial ephemera did the work that state pageantry once did. They are a testament to how compelling the “soft power” of the state was during the midcentury, when “Good Neighbor” diplomacy was translated into something a tourist could hold in their hand.

Further Reading

Linda Egan and Mary K. Long, eds., Mexico Reading the United States (Nashville: Vanderbilt University Press, 2009).

Dina Berger and Andrew Grant Wood, eds., Holiday in Mexico: Critical Reflections on Tourism and Tourist Encounters(Durham, NC: Duke University Press, 2010).

Douglas Monroy, The Borders Within: Encounters Between Mexico and the U.S. (Tucson: University of Arizona Press, 2008).

Darlene J. Sadlier, Americans All : Good Neighbor Cultural Diplomacy in World War II. (Austin, TX: University of Texas Press, 2012).

Stephen D. Morris, Gringolandia: Mexican Identity and Perceptions of the United States (Lanham, MD: Rowman & Littlefield, 2005).

W. Dirk Raat and Michael M. Brescia, Mexico and the United States: Ambivalent Vistas, 4th ed. (Athens: University of Georgia Press, 2010).

Dr. Kevin M. Anzzolin (Chicago, 1979), Lecturer of Spanish, arrived at Christopher Newport University in 2021, where he teaches a wide range of classes. His book, Guardians of Discourse: Literature and Journalism in Porfirian Mexico, was published with the University of Nebraska Press in May 2024 and garnered an Honorable Mention from LASA-Mexico for Best Book in the Humanities in 2025. His research focuses primarily on Mexican narrative from the 19th- to 21st centuries, and his work has been published in journals such as Letras Hispanas, Hispania, and Studies in Latin American Popular Culture. He is active in public scholarship as well, having written about pedagogy, pop culture, and local politics in different fora.E-mail: kevin.anzzolin@cnu.edu

In November 1926, Emma Goldman, then 57 and beginning a first period of exile in Toronto, enthused over a “radiant” young Italian woman seated at a welcoming dinner for Goldman organized by the city’s Jewish anarchist comrades.[1] Just over six months later, Goldman dubbed Marie Tiboldo “a terrible prude” who did “not feel anything deep enough, whether…for a man or ideas, to do anything worth while.” Tiboldo appears only fleetingly in Goldman’s estimated 200,000 letters, but her assessment of the young woman’s shortcomings aligns with Goldman’s pronounced tendency to go hot and cold with friends, comrades, and lovers. It’s not surprising, then, that Goldman also emphasized her continuing interest in “Marie,” writing that “I do not mean to give her up” and “I will certainly see much of her.”[2]

Marie Tiboldo (far right) at Bryn Mawr Summer School for Women Workers, 1929 (Julia Barnett Collection)

A Toronto-based protégé of Goldman’s (1869-1940) who outlived her “idol” by almost 70 years, Tiboldo (1904-2008) has remained virtually unknown even to Goldman specialists. Unlike some working-class women anarchists who, like Rose Pesotta, consciously fashioned their own archives, Tiboldo left behind no papers.[3] Goldman’s fame has often obscured the lives of other anarchist women but occasionally – as with Tiboldo – it has revealed clues about lesser-known figures in the period’s anarchist circuits.

Initial leads beyond Goldman’s letters were few but tantalizing, especially a short interview with a 99-year-old Tiboldo published in 2004 by Julia Barnett, a much younger socialist feminist who had befriended Tiboldo.  A US composer we met at a symposium marking the 150^th anniversary of Goldman’s birth led us to his brief 1993 videotaped interview with Tiboldo.[4] Research required attentiveness to inconsistences in the sources and the complex work of memory in the life of a long-lived Italian Canadian working-class anarchist. Our interest in Tiboldo stems from our project on how Goldman’s Toronto exile has been made and re-made inter-generationally in the city’s popular memory, with a focus on the political and cultural stakes of these reactivations.[5]   Tiboldo played a role in creating that memory, but her own story offers intriguing glimpses into the trajectory of Toronto anarchism.

Tiboldo was the US-born and Canadian-raised daughter of northern Italian anarchists who left the United States in 1914 for Toronto. Their radical networks in the city included an anarchist circolo in which Marie’s father, Lorenzo, was a driving force and the anarchist Branch 339 of the mainly Jewish (and socialist) Workmen’s Circle. The latter facilitated Goldman’s initial encounter with Tiboldo. As late as 1993, Tiboldo remained friends with Branch 339 co-founders, Morris and Becky Langbord.[6]  

A teenaged Tiboldo was hired in an embroidery shop in Toronto’s garment industry. She became a skilled embroidery worker and later designed embroideries for the “luxury trade,”[7] but exploitative conditions prompted worker unrest, leading to union activism and strikes. Led by socialists and Communists embedded in a vibrant Jewish working-class culture, garment workers, including women, became militants.[8]  Among the few young Italian women in the sweatshops, Tiboldo became a labour activist in this milieu, but the sources contain inconsistencies.

One narrative has Tiboldo single-handedly organizing her shop right after being hired in 1920. Consulting the Labour Gazette indicates instead that Tiboldo emerged as a rank-and-file organizer during a North American-wide organizing campaign launched in 1925 by the International Ladies Garment Workers Union (ILGWU).[9] A US-based international union affiliated with the American Federation of Labor, the ILGWU grew significantly in these years due in part to the militancy of young immigrant workers in New York City, many of them Jewish and Italian women. A union heavily shaped by anarchists and socialists, and a key site of working-class feminism, the ILGWU embraced a social unionism that heavily promoted workers’ education and culture and a “labor interracialism” that “elevated racial cooperation among the working classes as integral to a more egalitarian future.”[10]    

Tiboldo’s possible involvement in a successful 1926 strike by the Embroidery Workers Union, an affiliate of the ILGWU as Local 7 (Toronto), might explain her election not long afterwards as a delegate of that union to the Toronto Trades and Labour Council. And in 1929, she attended the Bryn Mawr Summer School for Women Workers, an immersive two-month program for industrial women workers that was also attended by Goldman comrade and anarchist labour organizer, Rose Pesotta.[11]

While Tiboldo had many male suitors, she neither married nor lived in a free love union. Influenced by Goldman’s radical views on sexuality and birth control, Tiboldo promoted young single women’s sexual autonomy, insisting they learn about sexual pleasure before marriage, not only the mechanics of pregnancy and giving birth.[12] Tiboldo praised the courage of the birth-control activists of her youth, though she likely knew that Goldman (among others) introduced Margaret Sanger to the birth-control cause, and then criticized her for abandoning radical politics for eugenics.[13] Proud of her association with Goldman’s famous April 1927 birth control lecture in Toronto, one that left a lasting impact on other left-wing women of the period,[14] Tiboldo critiqued the professionals, and their wives, who refused to chair the event. She recalled being horrified by Goldman’s suggestion that she chair it (thereby prompting Goldman’s ire) but agreed to be put “through some exercises” because it meant being with her “idol.”[15] Birth control activist Alice Loeb chaired the lecture.

Our forthcoming chapter on Tiboldo, which also covers her activism in the 1940s within the Communist-led United Electrical Radio and Machine Workers of America as well as her relationship with other Italian anarchists, offers further insights into the trajectory of working-class anarchism. Examining Tiboldo’s life on its own terms challenges any easy divide between sexuality-and-free-love rebels, of which Goldman is considered the exemplar, and anarchist labour organizers. Goldman’s own early history in the sweatshops is often ignored along with the material rhythms and conditions of her life as a revolutionary.[16] Goldman cultivated cross-class publics, but never lost ties with working-class anarchists for whom birth control and sexuality were fundamentally class issues and not external to class struggle. The Anarchist Queen could complain about her, but Tiboldo’s memories offer us valuable insights both into Goldman’s Toronto exile and the political choices made by Toronto’s anarchists before and after Goldman’s death in 1940.[17]  

Franca Iacovetta is a Toronto-based historian and a member of the riseupfeministarchive.ca collective.

Cynthia Wright teaches in the School of Gender, Sexuality and Women’s Studies, York University, Toronto/Tkaronto.

Today (August 21, 2026) marks the 133^rd anniversary of Emma Goldman’s famous Union Square, New York City “take bread” speech.

[1] Cited in Theresa Moritz and Albert Moritz, The World’s Most Dangerous Woman: A New Biography of Emma Goldman [hereafter WMDW] (Vancouver: Subway Books, 2001), 54.

[2] Emma Goldman Papers Public History Project, Emma Goldman to Ben Capes, 13 June 1927.

[3] Maria Tamboukou, “Labour Activists’ Archival Assemblages and Resistability in Memory Work: Emma Goldman, Jean Désirée, and Rose Pesotta,” Labour History 129 (November 2025): 29.

[4] Julia Barnett, “The Life and Times of Marie Tiboldo” [Barnett, “MT”]; New Socialist 46 (2004): 20-21; Interview with Tibaldo (sic) by Leonard Lehrman [LL interview with MT], Toronto, 1993, posted 2019, https://www.youtube.com/watch?v=t0BWLXGc3F4. Also, Minerva Davis, The Wretched of the Earth – And Me (Toronto: Lugus Publications, 1992), 83-6; Coleman Romalis, Emma Goldman – The anarchist guest (Documentary) 2000.  

[5] Currently entitled Remembering Emma Goldman in Toronto:  Making Intergenerational Radical Memory.

[6] ancestry.com Marie Tiboldo; Barnett, “MT”; LL interview with MT.

[7] Davis, Wretched, 88-9; LL interview with MT.

[8] Ruth Frager, Sweatshop Strife: Class, Gender, and Ethnicity in the Jewish Labour Movement of Toronto, 1900-1939 (Toronto: University of Toronto Press, 1992), chs. 5-7.

[9] Barnett, “MT”; Davis, Wretched, 88-9; Labour Gazette, April 1926, 397.

[10] Quotation from Victoria W. Wolcott, “Networks of Resistance: Floria Pinkney and Labor Interracialism in Interwar America,” Journal of African American History (Fall 2020): 569;  Annelise Orleck, Common Sense and a Little Fire: Women and Working-Class Politics in the United States, 1900–1965 (Chapel Hill: University of North Carolina Press, 1995); Jennifer Guglielmo, Living the Revolution: Italian Women’s Resistance and Radicalism in New York City, 1880–1945 (Chapel Hill: University North Carolina Press, 2010); Daniel Katz, All Together Different: Yiddish Socialists, Garment Workers and the Labor Roots of Multiculturalism (NYC: NYU Press, 2011).

[11] “Workers Are on Strike After Union Is Formed,” 5 February 1926 and “Workers Back on the Job,” 9 February 1926, Toronto Daily Star; Labour Gazette, April 1926, 387, 397; Tiboldo at Bryn Mawr, Globe and Mail, 3 August 1929; 1985 documentary, The Women of Summer : Bauman, Suzanne.

[12] LL interview with MT; Barnett, “MT.”  

[13] Candace Falk, Love, Anarchy, and Emma Goldman (New York: Holt, Rinehart, and Winston, 1984), 222-54.

[14] Dorothy Livesay, Journey with My Selves: A Memoir, 1909-1963 (Douglas and McIntyre, 1991), 66.

[15] Barnett, “MT”; LL interview with MT; Moritz and Moritz, WMDW, 86-8.

[16] Tamboukou, “Labour Activists,” 15.

[17] Cynthia Wright and Franca Iacovetta, “Radiant Creature and Anarchist Queen: When Marie Tiboldo Met Emma Goldman,” in Catherine Carstairs, Kassandra Luciuk, and Nadia Jones-Gailani eds. Feminist Provocations: Gender, Migration, and Activist Histories (University of Toronto Press, forthcoming).

In 1979, lesbians in Fredericton, New Brunswick, found community not in a dedicated gay bar or a visible neighbourhood, but on the dance floor of the Cosmopolitan Club. On Thursday nights, women quietly gathered with friends, effectively “queering” a predominantly heterosexual venue. In a city without lesbian bars, bookstores, or formal organizations, even a dance floor could become a place of belonging. From 1979 to 1995, lesbians in Fredericton actively built and sustained physical and imagined queer spaces that fostered community, activism, and belonging. Whether through dances, newsletters, phone lines, private homes, or women-only organizing, these spaces reveal how meaningful lesbian community could flourish beyond Canada’s metropolitan centres. In recovering these histories, Fredericton’s lesbian spaces emerge not as exceptions to Canadian queer history, but as evidence that community has always existed.

Queer history in Canada is often centred on the country’s largest urban centres: Toronto, Vancouver, and Montreal. Histories of queer activism have long focused on these cities, where visible institutions such as bars, bookstores, political organizations, and later gay villages left a stronger archival footprint. Much of the scholarship has similarly centred the experiences of urban gay men, tracing legal milestones such as decriminalization, police repression, and the rise of liberation movements. However, this focus overlooks the rich and varied experiences of lesbians in places like Fredericton, where queer history did not always take the form of visible institutions. Lesbian histories outside major cities challenge the assumption that queer communities depended on permanent spaces such as gay villages or commercial venues. Instead, many rural and small-city lesbians built networks through temporary gatherings, personal relationships, and community-based organizing.

As the provincial capital and home to two universities, Fredericton occupies a unique position within New Brunswick’s largely rural landscape. For many lesbians living in smaller towns, it became the nearest place to find others like themselves. Women travelled hours by car to attend dances, softball tournaments, and other events in the city, while local publications and phone networks helped sustain connections across long distances.

One of the clearest ways that lesbians created space in Fredericton was through community organizations. The founding of Fredericton Lesbians and Gays (FLAG) in 1979 gave lesbians access to meetings, dances, discussion groups, and a volunteer phone line. Even without a permanent headquarters, FLAG transformed temporary spaces such as university classrooms, church gymnasiums, and rented halls into meaningful queer spaces. However, the 1980s brought a notable decline in lesbian participation in mixed queer organizations like FLAG, shaped by broader tensions including lesbian separatist politics and ongoing experiences of misogyny and marginalization within male-dominated queer spaces. Across Canada, many lesbians responded to these tensions by creating women-only spaces that prioritized lesbian voices outside of male-dominated gay spaces. In Fredericton, this shift contributed to the founding of WOMYN in 1991, a lesbian-focused group that hosted and organized a variety of social activities, including dances, potlucks, barbecues, and softball games. By organizing regular social events in dedicated spaces, WOMYN created consistent opportunities for lesbians to build relationships and foster a sense of belonging.

Nightlife was another important way that lesbians in Fredericton carved out space, even without a dedicated lesbian bar. In the 1970s, lesbians gathered at the Cosmopolitan Club, a straight-owned venue on King Street that became a key nightlife spot for both gay men and lesbians, especially on Thursday nights. Because it was still risky for lesbians to dance openly with other women, gay men and lesbians often danced together in mixed pairs to appear as a heterosexual couple. By the 1980s and 1990s, this nightlife expanded through FLAG’s mixed dances and later WOMYN’s women-only dances, which offered safer and more intentional spaces for the lesbian community, visibility, and belonging.

An advertisement for Fredericton Lesbians and Gays (FLAG) from the first edition of Flag Mag, 1980. (Flag Mag No. 1, September 1980: 2)

New Brunswick’s expansive natural landscape offered lesbians opportunities to gather and socialize through activities like hiking, camping, and team sports leagues such as softball, creating alternative spaces for connection outside of traditional venues. Unlike queer bars or meeting spaces, which were often temporary, exclusionary, or inaccessible, the outdoors offered a consistent and flexible environment where lesbians could gather while resisting discrimination and navigating the misogyny often present in mixed-gender queer spaces.

The 1990s saw a shift toward more lesbian-centred outdoor activities. Softball became a defining feature of lesbian life in Fredericton, with women’s teams emerging as early as the 1970s and growing in number in the 1990s. More broadly, softball became historically associated with lesbian communities across North America because it was inexpensive and provided opportunities for women, many of whom were unmarried, to participate in organized team sports. Unlike bars or bookstores, softball took place in public parks, making it more accessible and less dependent on commercial businesses or private venues. In Fredericton, WOMYN organized specifically lesbian-centred softball events, including the Annual Labour Day Lesbian Softball Tournament, which combined games with social activities such as dances and house parties. Through softball, lesbians in Fredericton created space for meaningful social interaction, both on and off the field.

Beyond the rented community centre halls and the softball fields, another kind of space thrived. It was not built with walls and doors, but comprised of phone calls, newsletters, and a collective identity. These imagined spaces, while intangible, offered structure and support, allowing Fredericton’s lesbian community to grow in ways that defied geographic spaces. In Fredericton, imagined space became vital. In a city where spaces for queer collectivity were often private, temporary, subject to discrimination, or dominated by gay men, imagined space became a necessary means for lesbians to find one another, build connections, and foster a sense of belonging, even without meeting face to face.

Queer organizations played a central role in producing these imagined spaces. Beginning with informal word-of-mouth networks in the 1970s, lesbians often relied on chance encounters and personal connections to meet others, a system that was both intimate and limited in reach. By the 1980s, organizations such as FLAG expanded this informal network through more structured forms of communication, especially through their newsletter Flag Mag, which circulated promotional events for FLAG, queer news, opinion pieces, and personal reflections. FLAG also created the Gayline, a volunteer-run phone service, often manned by lesbian members, that provided anonymous support, information about events, and was often a first point of contact into Fredericton’s queer community.[1] At the same time, campus-based organizing through the Gay and Lesbian Alliance of UNB (GALA) extended these networks through social events, columns in The Brunswickan, and campus radio programming, creating overlapping channels of communication that connected lesbians in Fredericton through shared information, media, and identity as much as through physical space.[2]

The 2000s marked a transitional period in Fredericton’s lesbian and broader queer history, as visible lesbian-specific spaces declined while digital communication and more inclusive organizations became more prominent. Although lesbian-specific venues and organizations are harder to trace in the archival record, lesbian presence did not disappear but instead shifted into new forms of visibility through activism, media, institutional spaces, and emerging online networks.

From informal word-of-mouth networks and softball fields to newsletters, radio programs, and church hall dances, lesbians in Fredericton consistently created spaces of connection, belonging, and resistance across changing social and political landscapes. While these spaces were often temporary, overlapping, or difficult to trace, they demonstrate that lesbian communities in the 1970s-1990s were not absent from Atlantic Canada but were actively constructed through both physical and imagined forms of space. This history challenges assumptions that meaningful queer life only existed in major urban centres, instead showing that lesbians in Fredericton were active in shaping their own communities. They built networks of friendship, love, and support, and created events, spaces, and shared meanings that gave depth and visibility to queer life in Fredericton.

Rosemarie Forsberg graduated with a master’s in History from Wilfrid Laurier University. She recently worked as an archivist at the Provincial Archives of New Brunswick and is now exploring the field of librarianship through her work at the London Public Library

Resources/Further Reading:

The Queer Heritage Initiative of New Brunswick (QHINB) https://www.queerhistoriesmatter.org/qhinb

Exploring Queer History and Activism in New Brunswick: CNBRDP https://canadacommons.ca/artifacts/30160044/exploring-queer-history-and-activism-in-new-brunswick/31059889/?__cf_chl_tk=0xGOgfaO4OSKugEJAX3NSMiZSlBUI_cTrXZqverpl3Q-1776288880-1.0.1.1-J6qV6IzhEFGDrg9yOdKWIbexi6DRWuF45Zd0Bs4QQW4

Falvey, Conor. “Foregrounds: Mapping Gay, Lesbian & Bisexual History in Fredericton, 1969-1992.” Master’s thesis. University of New Brunswick, 2015.

Warner, Tom. Never Going Back: A History of Queer Activism in Canada. Toronto: University of Toronto Press, 2002.

[1] Sorry! Wrong Number!” Flag Mag, April-May 1986: 6.

[2] “Update on GALA,” Flag Mag, November-December 1992: 9..

While a small but vocal minority of historians have embraced the use of generative artificial intelligence in their research and teaching,[1] many of us find ourselves trying to hold the pedagogical line against what we perceive as serious threats to students’ ability to learn, reason, and express themselves intelligently.[2]

In this post, I share my attempts to “teach against AI” in an undergraduate course taught at McGill University in fall 2025. HIST 299, The Historian’s Craft, is an introductory theory-and-method class recently made a requirement for history majors. My objectives were for students to consider some of the major questions in the philosophy of history, learn about important changes over time in historiography, learn about some of the main ways that people “do” history, and then to try their hand at doing it themselves, via research and writing.

As I’ve made clear in past pieces, I don’t believe that generative AI has anything useful to offer students towards the attainment of any of these goals. To the contrary, I think it poses a significant danger to students’ ability to think independently and develop as thinkers, writers, and human beings. Emerging research supports my concerns.

I thus adopted a number of strategies in this course to encourage students to do things the “old school” way, with the aim of fostering the learning and skill development at the core of our mission as educators in humanities and social sciences. In what follows I outline what I did, what I didn’t do, and what students thought about it all.

What I Did

1. Talk about AI from the jump. The first assigned readings of the course were Active History pieces by me and Mark Humphries in which we take different sides on the use of generative AI in teaching history. This debate informed our first in-class discussion, in which students expressed a range of views on the pros and cons of AI in education. My goal was not to convince them of my position (what could be more cringe?), but instead to encourage reflection on what it is that historians – and humanists – do, and how generative AI might support or challenge that. I felt that this earnest, constructive discussion, addressing the elephant in the room, set a positive tone for the remainder of the semester.
   
2. Printed course packs. Even before the release of ChatGPT and the AI tools that followed, I harboured doubts about assigning students digital readings. Personally, I read much more effectively on the page versus on the screen. Research shows that I am not alone. I had long fantasized about assigning only hard-copy readings, but the ascendance of generative AI gave me the final push I needed. I assembled the assigned readings and sent them off to a student-run print shop, which sold the 271-page course packs for just under $20 each. It should be noted that this takes some more organization on the part of the professor, considering the lead time needed for printing. No more finalizing readings on the night before the first class; no more “readings TBD” on the syllabus. Surely this is a healthier practice anyway, even it requires slightly more August effort.
   
3. Laptop ban. As with the course packs, this change was something I had pondered for years. It really had nothing to do with AI, but instead with the sense that laptops in class do not, in fact, help learning, but instead hinder it. Distraction is one thing. But even for undistracted laptop users, I question the benefits of the tool, particularly for notetaking during lectures. Laptops allow for a near-transcription level of notetaking. But is near-transcription a good way to take notes? My position is that it isn’t. A crucial skill that notetaking develops is to filter a lot of information down to its most important points. To use a culinary metaphor, a student should learn to create a reduction, boiling off the excess so that a denser essence remains. The process starts with lectures and readings. The student should take notes, but crucially the notes should be vastly shorter than the original source material. Then when it comes time to write an essay or prepare for an exam, the student reviews their notes and reduces further into an outline or study guide or something else. It has always seemed to me that notes taken on a laptop delay and perhaps even dissuade this practice of reduction. I don’t believe this to be a trivial matter in the slightest. Whether we call it reduction, filtering, funneling, or creating a hierarchy of information, this practice of organizing and ordering information is one of the core skills that we teach in higher education – and perhaps even more in history than in many other disciplines. So, I banned laptops.[3]
   
4. In-class writing. Understanding the easy temptation of using generative AI for assignments such as reflection papers, I shifted this sort of assignment to in-class. Adopting a practice from my high school English teacher Geoff Peate (shout-out Mr. Peate), I had students do “power writing.” Ten times throughout the term, students would spend the first ten minutes of the class responding to a prompt, using paper and pen. It didn’t matter how much they wrote or how “brilliant” it was. They just needed to write. At the end of the ten minutes, students would deposit their page in their individually labelled file folder, which I lugged to and from class every time in a file box. It was not as awkward as it sounds.
   
5. In-class reading and discussion. Students were assigned plenty of take-home reading to do on their own time, but I also sometimes used class time to have them read something short and discuss it. This was not so much a reaction against AI (though of course we know that students are often using AI to summarize their readings), as it was a valorization of reading time. If reading is so important (as I believe it to be), then why not devote some class time to it?

Interlude: what about accessibility?

I would hope this goes without saying, but just in case it doesn’t: of course, students with accessibility reasons for reading on the screen, or using a laptop in class, or students with any other accessibility requirements were accommodated, just as they would be in any other of my classes. See endnote #3 regarding accessibility and the laptop ban.

What I Didn’t Do

1. Do away entirely with at-home assignments. I want to create structures in my classes that encourage students to develop as thinkers and writers. But I can’t control everything, nor do I want to. I am a teacher, not a cop. So I persist in assigning students conventional essays, while nurturing the skills needed to produce them independent of generative AI tools. Along the way I try to convey to students the benefits of old-fashioned research and writing for their own learning and development. As Mack Penner quipped in our co-authored piece about AI in classroom, adjustments to course delivery are well and good, but “We might also have a crack at persuasion.”
2. Use AI detectors. Again, not a cop. And again, let’s try persuasion.

What Students Thought

In order to gauge students’ reception of HIST 299’s experimental components, I added tailored questions to the course evaluations that asked students about the in-class writing (“power writing”), laptop ban, and printed course pack.

In general, students responded positively to all three measures. The images below summarize their responses.

The quantitative data is useful, but student comments are even more revealing. Let’s consider responses to the three questions in turn.

In their comments, many students thought power writing to be an effective way to ensure the completion of readings but were less convinced that it helped them improve their writing. As one student shared: “It definitely ensures that people do the reading, and encourages them to really comprehend it. However, I don’t think it does much for writing skills because there is no opportunity to edit or plan what you write due to the limited time and lack of preparation. If it is intended to ensure comprehension, it is effective, but it does little for writing skills.” I take students’ points here, but am inclined to think that power writing at the very least loosened them up as writers, evidenced the steady lengthening of students’ submissions as the term progressed.

As for the laptop ban, only one student took the time to write a negative comment about it. The majority found it beneficial. Said one student: “Considering another class in which I had access to a laptop, I personally feel as though it made me far more effective and engaged as a student.”

The smash hit of the course’s experiments was the printed course pack. 84% of respondents liked having a printed course pack, with 62% expressing strong support. Again the comments are revealing. Of 22 comments made about the course packs, 19 were positive and 3 negative. Particularly striking is how effusive the positive comments were. A selection:

• This was very helpful! Much easier to read a physical copy, and it meant I could easily reference writings during [tutorials] and I wasn’t distracted by my screen.
• I hate doing readings on my computer so this really helped my eyes
• “We love reading on paper” – Everyone, at any point in time.
• My brain retains more information when it comes from the page
• LOVED!
• My eyes thank you for this one, I love not reading on a screen.
• It provided me with a greater ability to mark down my thoughts and an easier way to bookmark segments I consider interesting.
• I love love love the reading packet provided even though we had to pay for it. All profs should do this!
• I really appreciated having a course pack dedicated to our course’s reading material. It was refreshing to have my readings in print and helped me stay organized throughout the semester.

Final Thoughts

HIST 299 was my favourite course I’ve ever taught. The low-tech experiments were only part of that. The much bigger factor was the students, the vast majority of whom were keen and bright and constructive, making class stimulating and fun.

I do think that the low-tech experiments helped unleash this creative group of young people. Without laptops, students were more likely to talk to each other. With in-class writing, reading, and discussion, students engaged in collective endeavours rather than simply attending a lecture as passive receptors of information.[4]

Different professors will adopt different approaches to AI – and tech more broadly – in their classrooms. Indeed, one professor might take different tacks for different courses (I certainly do). My aim here is not to insist upon one correct way of doing things but simply to share the results of one historian’s ongoing pedagogical experimentation in an era of impressive socio-technological change and challenge.

Edward Dunsworth teaches history at McGill University and is a member of Active History’s editorial collective.

[1] At the February 2026 Canadian Historical Association panel, “Generative AI and the Practice of History,” for example, one panelist (Mark Humphries) repeated his oft-stated assertion that generative AI tools have either developed intelligence or that the question is moot since their outputs can’t be differentiated from humans’. Another panelist, Ian Milligan, criticized historians who weren’t using AI as somehow falling down on their job duties by remaining stubbornly inefficient. I find the first position preposterous and the second offensive but will leave my commentary to that. (The third panelist, L.K. Bertram, presented a more measured take on the use – and sometimes non-use – of AI technology in historians’ work.)

[2] Is my position in many ways a conservative one? Yes, it is. Does this make me, a committed socialist, a bit uncomfortable? Actually, no, not really. Leftists’ revolutionary aims get all the press, but leftism almost always contains strands of conservatism. We may want to conserve, for example, human dignity, freedom of thought, leisure time, forests, oceans, animals, insects, and so on. So, yes, I am a conservative when it comes to resisting the unchecked incursion of generative AI into our thinking and writing, especially in educational settings. Sue me.

[3] My sociologist colleague Barry Eidlin gave me an excellent strategy for implementing a laptop ban without discriminating against students with accessibility needs. The language in my syllabus makes it clear that students with accessibility requirements can of course use laptops if needed. Furthermore, any student at all who wishes to use a laptop need only to set up a meeting with the professor to explain why. This prevents students with accessibility needs from being “outed,” since laptop users in class could also be students who simply requested permission. In practice, few students avail themselves of this option.

[4] To be clear, I am not anti-lecture. In this particular class I did very little lecturing, but in other courses I adopt a more traditional approach. Different courses call for different tactics.

“Algonquin Park,” October 4, 1995, photo by Dick Loek, published by the Toronto Star (Toronto Star Archives, Toronto Public Library).

ActiveHistory.ca is back from our summer hiatus! This week we’ll be releasing new posts, and we’re excited to share some new changes with all our contributors, guest writers, guest editors, and readers that we’ll be rolling out later this year. Keep an eye out!

Sincerely,
The Active History Editorial Collective

It Won’t Be Long Now! A Scene at the Bathing Beach at Port Dalhousie, Ontario,” May 2, 1930, photo by Canadian National Railways, published by the Toronto Star (Toronto Star Archives, Toronto Public Library).

ActiveHistory.ca is on a hiatus for our annual summer break, with a return to daily posts in mid-august.

Thank you to all our contributors, guest writers, guest editors, and readers for making 2026 a very successful year. In the meantime, we will be periodically reposting some excellent posts. 

Wishing you all a great summer we look forward to continuing our work soon!

Generations of social historians have scoured archives and invented new ways of doing history through their collaboration with other fields. A new “vintage AI” model, talkie, is an example of what happens when knowledge extraction replaces collaboration.

Parsing the various communications around the model, the team behind talkie has framed their new LLM in two ways. In their blog post introducing this new model, and in an interview with the podcast Hard Fork, the team behind this project describe its predictive potential. They aim to test whether language models can predict events beyond the period of the training data, and do so by training it with pre-1931 documents. The team presupposes that if this “vintage” model can predict its future, then modern models trained on the current web can predict our future.

talkie chatbot interface, however, does not engage with the model as a predictive tool. There is no proof of its efficacy. In our testing, their local model is not shipped with any predictive capabilities, nor do they provide sufficient documentation to achieve any prediction. Instead, it acts more like a toy to interact with a particular version of history. Their tagline advertises the model as “an LM from 1930,” as if the user is talking with an entity from the past. The framing between prediction and history disguises a tool whose use is unclear and whose underpinnings are flawed.

These flaws begin with the data on which talkie was trained. Its training corpus was a subset of the published English-language record before 1930. Like most western archives, this is a corpus likely dominated by white, male, and elite writers. Other voices are outweighed in the records and further filtered by digitization. The result is a model that speaks confidently for the past in the voice of its narrowest slice.

The talkie team made these choices deliberately. On the podcast Hard Fork, talkie team member Dr. David Duvenaud, associate professor at the University of Toronto, stated they “wanted to make everything entirely publicly available in open source, and 1930s is the most recent date that has almost zero legal headaches.” However, the team has not released the training data sources, so what little we know about the data set is based on their public statements. The training dataset seemingly contains only print-origin English-language texts, excluding hand-written or multilingual sources. Their blog post specifically lists “books, newspapers, periodicals, scientific journals, patents, and case law” as their training corpus. They drew these materials from repositories like Harvard University’s Institutional Data Initiative, the Internet Archive, and products from the Common Pile project. In so doing, the team relied on data that was easy to find and ready to use, instead of developing a representative corpus with diverse perspectives from the past. A concentrated reliance on ready data can produce problems, like reproducing narrow-minded points of view held by certain people in the past.

This reliance on easy data was not from a lack of access to expertise. Five months before talkie‘s release, one of us, Farr, met with Levine and Radford for an hour-long discussion about optical character recognition (OCR) and accessing historical archives. As part of the talkie project, state-of-the-art OCR is in development, intersecting with Farr’s research on vision language models for archival transcription. However, instead of discussing new AI-enabled transcription tools as Farr had expected, the team questioned her about how to source training data for their forecasting model. Farr’s questions about their corpus curation methods, such as what would be included or excluded, and how the gaps would be acknowledged, were met with only vague interest. They considered these questions out of their present scope. Five months later, the model shipped with the corpus problems Farr had flagged represented only as ‘limitations’ and ‘future directions’ in the release post, which acknowledges her contribution in ‘helpful discussions’.

Heeding warnings like Farr’s may have lessened some of the problems with talkie’s outputs. As seen in Figure 1, talkie produces bigoted responses then frames them as a representation of 1930; an outcome of the project’s methodology. The hard work of expanding the archive is being subverted. Humanities researchers have read marginalized experiences against the published record, mapped enslaved people’s lives using colonial ledgers, and reconstructed Indigenous histories from oral traditions and the margins of colonial documents. The talkie team’s use of easy and accessible documents, and the model’s resulting bigotry, overshadows the professional and collaborative efforts that have gone into diversifying and expanding the archives beyond these problematic voices. These efforts have grappled with the archival challenges ignored by talkie: much was never printed; much of what was printed is not digitized; most digitized handwritten sources are not transcribed. The ephemeral and multilingual sources needed to train a model on anything approaching the full archive of the past simply do not exist in machine-readable form.

Framing talkie as a chatbot dictates how the public will use it. The chatbot has largely been received as a toy. In one Reddit thread, users repeatedly prompted it to predict Hitler’s rise. A chatbot ‘from 1930’ invites users to ask it what 1930 did not yet know. The interface produces this behavior and the model’s construction makes such use inherently fraught.

Without historical framing, a casual user of talkie has no reason to read “‘[A woman] is inferior to a man intellectually” as anything other than a broadly-accepted historical truth, instead of it being a perspective held largely by the white elite. This is a direct result of talkie’s team taking minimal steps towards preventing users from interacting with problematic results. Standard industry approaches like post-training guardrails are common for language models trained on bigoted internet-scraped data. The talkie team characterized these post-training guardrails as “put[ting] their thumb on the scale,” ignoring how they already tipped the scales through their dataset choices.

The talkie team is well equipped to perform post-training on their model. talkie was built by a group including Dr. Duvenaud, whose research focuses on AI governance and ethics in AI use, and who previously worked on Anthropic’s Alignment Science team; Alec Radford, a former employee of OpenAI and principal author on GPT-2, CLIP, and Whisper; and Nick Levine, a former quantitative researcher and financial historian at the hedge fund Winton. Levine has an MPhil on the history, philosophy, and sociology of science from Cambridge and has published on “information overload in postwar America”.

This team has expert knowledge and experience, but talkie represents a minimal application of this expertise in relation to the bigoted output of the chatbot. Their only apparent safeguard, added post-release, is an additional model (Qwen3Guard family) that reviews the chat inputs and outputs and flags content “as potentially inappropriate”, obscuring the output with a “Show it anyway” button (see Figure 2). An accurate warning would instead explain that the content is likely incomplete and is, at best, a narrow perspective of the past.

talkie’s poor representation of history is a side effect of a larger problem. The model was conceptually built as an experimental tool to see if language models can accurately predict the future, also known as “forecasting.” The blog post assertively claims the models predictive capabilities, yet is devoid of evidence enabling researchers to reproduce their experiment. The version that could serve serious research is gated behind compute, and is the same version released with no guardrails at all.

As we have mentioned, we have no proof of the predictive capability of talkie. However, the team’s framing of talkie as a predictive tool aligns with an explosive growth in the gambling and prediction markets in North America. In February of 2026, the night before the United States struck Iran, hundreds of bets correctly predicting the strike appeared on Polymarket. The New York Times reported on these betting markets, raising questions about asymmetric information shaping these markets. Prediction markets are now a multi-billion-dollar industry built on the geopolitical forecasting the talkie team wants to test. A sufficiently capable model in this lineage would be valuable to anyone whose decisions depend on prediction, including the markets, and would therefore be very attractive to investors. As we have seen with the rise of language models like ChatGPT and its competitors, the developers do not get to decide how the tools they develop are used once released into the world. talkie is similarly positioned.

Whatever their intentions, the team should release the training data and co-design these models with archivists and historians rather than using them as beta testers. The team says they intend to work with historians on future post-training and model development. Yet, they already had that chance and failed to use it.

In releasing this chatbot, the team has set aside Farr’s warnings, and what humanities researchers have spent a century working out. Documents are only partial records of the past. Archives preserve the records those in power choose to keep. Absence in the record is itself evidence. Reading those absences requires interpretive expertise the model does not have and the interface does not invite. Modern models are far more than statistical mimics, but they remain tethered to their inputs. They cannot represent perspectives or facts outside of the dataset. If we want complete answers, we must build complete foundations.

Jacob Polay is a PhD student in History at the University of Saskatchewan, studying the roles Large Language Models have in the historical method. His current research involves creating an information retrieval pipeline using artificial intelligence tools to unlock the early modern archive at scale.

Chloë Farr is a PhD student in Computer Science at the University of Victoria, working at the intersection of artificial intelligence, archives, and digital humanities. Her research focuses on large-scale text recognition and analysis of historical documents, including newspapers, maps, and archival records. Learn more about Farr’s work on GitHub.

Jessica Jack is a PhD student in History at the University of Saskatchewan, developing applications for Large Language Models in historical research. They are doing so through studying settler land use in late 19th century and early 20th century Saskatchewan.

Earlier this year, over forty people logged in to Zoom to attend a CHA webinar titled “Generative AI and the Practice of History.” Introduced as a “show and tell” for AI tools and strategies, the webinar included a panel of three expert historians who each presented on a distinct topic related to agentic AI, the tech that’s been dominating headlines since the release of Chat GPT to the general public in November 2022.

While I was hoping for a lively conversation about AI and creative thinking, writing, labour, and capitalism, much of the discussion centred around “best practices,” revealing what I felt was a relatively uncritical acceptance of these dynamic platforms. During the Q&A, most discussants sought answers on how to apply AI tech “responsibly” in their pedagogical and scholarly work. I, on the other hand, was left wondering about a much more essential, epistemic question, one that gets to the root of our role as scholars, teachers, and citizens of the world: how will the process of cognitive “offloading” shape, and potentially harm, the creative potential of our scholarship? Will there be any space left in our praxis for curiosity? Wonder? Possibility?

I suddenly felt I had missed something in recent months. How and when had we reached the consensus that offloading and outsourcing our more arduous tasks was decidedly “good,” for history as practice, for the university, and for our engagement with the world more broadly?

Woman with wax tablets and stylus, 55 -79 AD, public domain

Our dynamic encounters with agentic AI are clearly resulting in a bifurcation of our research tasks into those that we can or “should” offload to AI, and those that we should not. Writing, most agree, is a hard no, unless it’s for reviewing structured writing like grant applications. So, too, is locating source material. Transcribing these materials, however, is a task some claim is particularly well-suited to existing AI technologies. In fact, a couple of historians, including one of the webinar panelists, Mark Humphries, have even developed an AI tool to transcribe handwritten historical documents more quickly and more accurately than earlier systems.

Traditionally, this is the kind of paid work that has been reserved for graduate students, or other under- or precariously-employed humans, who rely on this kind of experience and supplementary income to get by and find future work. Even if senior scholars continue to make a concerted effort to hire research assistants to do “human-required” labour, the use of AI will still eliminate paid work in what is already a wildly competitive, precarious job market for new PhDs. This example speaks to how AI systems serve capitalism very effectively. The more we adopt AI technology, the more we fuel the industrialist, capitalist obsessions with productivity, efficiency, and profit.

The bigger, though not unrelated, questions I have relate to our creative processes and potential for political activism as researchers, writers, thinkers, and scholars of the past. As I’ve alluded to, adapting our workdays and workstyles to AI tech means abiding by the logic of capitalism, undermining efforts we might otherwise expend to produce activist or radical history. Besides resisting the constraints of capitalist stricture, tasks like transcribing, reviewing, summarizing, locating sources, compiling citations, and even sorting notes and indexing, help us solve problems and access creative thought. When we work on these purportedly unproductive or simply boring tasks, our minds often wander. This is hardly a bad thing. Stories abound of creative thinkers and artists who spent hours daydreaming as they took long walks and performed tasks unrelated to their “work.”

During my own workdays, I’ve solved many small yet meaningful research quandaries while doing busywork: I’ve stumbled across a great idea jotted down on a forgotten sticky note while revisiting handwritten notes; rediscovered a fantastic article that got lost in my Zotero bibliography; and found the ideal quote to open a book chapter while tracking down page numbers missing from a footnote.

Recent research on mind wandering and daydreaming also touts the wonder of the cognitive processes unlocked by performing busywork. UBC professor and psychologist Kalina Christoff Hadjiilieva suggests that it is precisely this kind of unstructured (or less structured) thought that stokes imagination and inspiration. It may also be worth mentioning here that Christoff Hadjiilieva sees modern tech as a significant barrier to this kind of creative thinking.

This reminds me of those rare but exciting eureka moments when, through open association and directionless thought, I discover complex connections between the narratives, sources, and theories I’ve been mulling over. Ezra Klein recently reflected on how often these moments arise for him during plane rides, a place that I, too, have done great writing. Call me a romantic (I am), but these insights, I think, reveal the magic, discovery, and possibility that come with practicing history. During the CHA webinar, panelist Ian Milligan called this “friction,” something he admitted was necessary for some historians. But this is not simply a difference in method, and by treating it as such, we risk offloading creativity itself to AI. Friction, research suggests, is actually essential to the kind of broad, creative thinking that produces innovative scholarship.

As Ed Dunsworth put it in his wonderful piece in AH last June, “We are not widget producers. Or at least, we should not be. We are producers of knowledge, of analysis, of ways of understanding the world.” To this, I would add, we are storytellers, and builders of hope and radical possibility for our present and future worlds.

Further Reading

Christoff Hadjiilieva, Kalina. “Mindfulness as a Way of Reducing Automatic Constraints on Thought.” Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 10, no. 4 (2024): 393-401. https://doi.org/10.1016/j.bpsc.2024.11.001.

Echeverri Álvarez, Mariana. “Beyond Tech: What Does Responsible AI Mean in Higher Education?” UM Today, Apr 21, 2026. https://umtoday.ca/stories/beyond-tech-what-does-responsible-ai-mean-higher-education.

Gross, Terry. “Michael Pollan Says AI May ‘Think’ –But It Will Never Be Conscious.” Feb 19, 2026. Fresh Air. Produced by Anna Bauman and Susan Nyakundi. https://www.npr.org/2026/02/19/nx-s1-5713514/michael-pollan-ai-consciousness-a-world-appears.

Wong, May. “Stanford Study Finds Walking Improves Creativity.” Stanford Report, Apr 24, 2014. https://news.stanford.edu/stories/2014/04/walking-vs-sitting-042414.

Nastasha Sartore is a historian of gender, labour, and everyday life in modern Britain. She is currently working on her first book project, which reimagines the intimate lives of women living on the margins of society in late-Victorian and Edwardian London. Nastasha earned her PhD from the University of Toronto in 2024 and is the Elizabeth and Cecil Kent Postdoctoral Fellow in the Department of History at the University of Saskatchewan for 2025-7.

This week, I talk with Cody Groat, author of Always Part of the Land: The Federal Commemoration of Indigenous Histories. We discuss the Historic Sites and Monuments Board of Canada‘s approach to Indigenous history, the role of federal commemorations in colonialism, and the power of commemorations. We also chat about Board members’ influence on commemorations, how public servants have shaped the process, and the Board’s efforts towards Reconciliation.

Historical Headline of the Week

Chadd Cawson, “Plaque unveiled to honour Coast Salish knitters and the Cowichan sweater,” Nanaimo Bulletin, April 2, 2026.

Sean Graham is a cultural historian, an Adjunct Professor at Carleton University, and a contributing editor with Activehistory.ca