5th AMLD kicks off by exploring the link between democracy and AI
Victor Kristof, doctorant en apprentissage automatique à l’EPFL a animé la session matinale de la première journée des AMLD 2021 © C4DT 2021
The 5th edition of the Applied Machine Learning Days (AMLD) kicked off on 25 January 2021 with a day of online talks exploring democracy and artificial intelligence. Due to the pandemic, this year’s edition will take place on a monthly basis throughout 2021. In all, 11 days of talks on 13 topics (or “tracks”) and 25 workshops will examine various issues related to AI, and in particular how machine learning is being used in various industries and fields of study.
By way of introduction, professor Robert West from EPFL’s Data Science Laboratory discussed how “coronavirus is a source of both challenges and opportunities for applied machine learning,” a field that “could help solve problems created by COVID-19.” The AMLD organizers – Prof. West along with Marcel Salathé and Martin Jaggi – announced the creation of an academic committee, whose members include the three organizers together with professors Lenka Zdeborova, Carmela Troncoso and Mackenzie Mathis.
As Victor Kirstof – an EPFL PhD student in machine learning and a coordinator of the day's events – pointed out, the relationship between AI and democracy is often misunderstood, as reflected in public perceptions of the Cambridge Analytica scandal and the influence of “fake news.” The event organizers therefore wanted to show the more beneficial side of this technology, demonstrating how machine learning and AI can help bolster democracy and how synergies could be leveraged between computational and political science.
Using AI to measure public opinion and predict voting results
The morning started off with a talk by Lucas Leemann, Assistant Professor of Political Science at the University of Zürich and co-founder of online polling organization LeeWas. His work shows how machine learning can be used to improve public opinion surveys.
Alexander Immer, a former EPFL Master's student, presented an algorithm that he helped develop during his time at EPFL and which is now part of the Predikon.ch website. Predikon employs probabilistic modeling, large-scale data analysis and machine learning to generate predictions about the outcomes of Swiss referendums, based on partial regional results and historical data.
Steven Eichenberger, a research and teaching fellow in Swiss politics and public policy at the University of Geneva and a coordinator of the day's events, pointed out that “understanding voting is a key concern for political scientists” and that voting predictions can “affect voter behavior.” According to Eichenberger, “applying machine learning to the study of voter choices remains rare in the field of political science,” although data scientists can improve their predictive models by reviewing the political science literature.
The use of machine learning by governments and civil society
The afternoon program provided examples of how machine-learning programs are being used by academics, civil-society organizations and governments to tackle a range of problems. For example, machine learning has been deployed to identify corrupt practices in Brazil's National Congress, and an algorithm (known as Bot Dog) was created to track down online hate speech with help from Internet users.
The final event of the day was a much-awaited presentation by Rayid Ghani, professor at Carnegie Mellon University and the chief scientist for Barack Obama's 2012 presidential campaign. Among the many applications he’s working on, Ghani spoke about how machine learning can help identify children who are most at risk from lead poisoning and police officers who are liable to use excessive force during arrests.
The next “track” is AI and Food & Nutrition, and will be discussed on 1 March 2021. A full list of the topics covered and the dates of all scheduled talks, workshops and AMLD 2021 events is available at www.appliedmldays.org and on Twitter at #AMLD2021 and @appliedmldays.