Habemus papam! Minutes ago, the Vatican announced that U.S. Cardinal Robert Prevost would be the next pope. Artificial intelligence (AI) made its own prediction earlier this week—but Prevost was not on the shortlist. In a paper posted on the preprint site arXiv, a machine learning algorithm that analyzed the ideological positions of the 133 high-ranking Catholic Church officials who made this decision predicted that Italian Cardinal Pietro Parolin would be the next head of the Catholic Church. Though the program was off this time, experts say its approach could eventually be useful in predicting other types of electoral contests.
AI algorithms that analyze text, such as social media posts and candidates’ speeches, can be very accurate at predicting political election outcomes. But papal conclaves present a unique challenge. The election process, which takes place over multiple rounds until one person receives two-thirds of the vote, has remained the same for centuries. There are no polls or primary elections to analyze, and the papabili are sworn to secrecy about their votes.
“Election processes generate a lot of data, but conclaves generate very few data,” says Eugenio Valdano, an epidemiologist at INSERM, the French biomedical research agency. “You know who goes in and you know who is elected pope.” He adds that the dynamics of an election in which just a few dozen people choose a leader from among their own ranks are very different from those of a political election in which millions of people vote for one of two or three candidates.
So with the help of Michele Re Fiorentin, a physicist at the Polytechnic University of Turin, and University of Madrid mathematician Alberto Antonioni, Valdano set out to develop a way to predict papal elections. When it was announced in February that Pope Francis was ill, this group was already studying the emergence of political and ideological factions within the church, using an algorithm the researchers had trained on five centuries of meticulous “genealogical” records of bishops and the successors they appointed. The researchers’ logic is that a bishop’s or pope’s decision to appoint a new bishop or to elevate one to a cardinal may be partly determined by shared ideology. And when the time comes to appoint a new pope, the thinking goes, each elector is more likely to vote for a colleague who shares his own stances.
To model how these dynamics could play out in the current conclave, the researchers chose four broad topics likely to be important to this year’s papabili: attitudes toward same-sex couples, international migration and poverty, the Catholic Church’s ongoing dialogues with other religions, and synodality—the degree of autonomy and authority enjoyed by local church leaders relative to the pope. Using data from a website that compiles cardinals’ public statements, the researchers trained an AI model to determine how progressive or conservative each elector’s stance on each issue is. The model then categorized the 135 eligible electors (two cardinals didn’t attend the conclave)—and Pope Francis—by their ideological similarity to other candidates.
Next, the scientists simulated the conclave election process among their virtual cardinals. After eight or nine rounds of voting, the electors typically converged on one candidate: Parolin, currently the Vatican’s secretary of state. As the Vatican’s top diplomat and second-in-command to Pope Francis, Parolin is widely considered a front-runner and has the best odds on online betting sites.
The model’s next likeliest candidate, South African Archbishop Stephen Brislin, came as more of a surprise given his lack of media attention. Brislin is considered a moderate and has spoken out in favor of immigrants and called for the church to address poverty. Valdano says it’s possible voters could converge on him as the election dwindles to a few compromise candidates.
Brislin is followed by another bookie favorite: Filipino Cardinal Luis Antonio Tagle. Tagle is also a moderate and his election could support the large Catholic community in the Philippines.
The model’s results changed if the researchers adjusted parameters such as the most important issue in the election. If the conclave prioritizes international migration and poverty, the most likely winner by far would be the progressive Archbishop Matteo Zuppi of Italy, who is known for speaking out against anti-immigrant policies.
The authors’ analysis had put Prevost’s views near the center on all four ideological topics, meaning he was likely elected as a compromise candidate. Electing a U.S. pope could also help shore up support from Catholic communities there.
Re Fiorentin says the model probably missed Prevost as a likely pope because it didn’t consider political and geographical factors that played a role in the election. Lacking that information, he says, “is a major shortcoming of our model.”
In the future, he adds, the model could incorporate geographic information about the candidates. “However, we think that other important data about geopolitical influence, lobbying, etc., are much more difficult to obtain and to use.”
“I think this is a very interesting contribution in our field,” says data scientist Rohitash Chandra of the University of New South Wales Sydney. Chandra cautions, however, that the data were probably insufficient to train the model: Candidates’ actual beliefs don’t always match their public statements, so including information such as social media posts might improve the model’s predictions. Even so, he says, the approach of categorizing candidates’ positions on specific topics could be applied to other types of election predictions such as local elections.
Luigi Curini, a political scientist at the University of Milan, says he likes the model election the authors ran, adding that other research into conclaves’ proceedings supports the idea that ideological affinity and the voting behavior of others drive electors’ decisions. He points out, however, that other factors can also influence voting, such as nightly conversations among the electors, who are sequestered in Vatican housing and live together for the full conclave.
The authors acknowledge these limitations, especially the lack of historical data that could shed light on actual conclave dynamics. “To us it has been a fun and stimulating exercise, and we share hoping that fellow nerds will find it interesting.”
More: https://www.science.org/content/article/ai-predicted-next-pope-did-it-get-it-right
