My name is Stephanie Cairns, and I鈥檓 a master鈥檚 student at McGill studying applied math and machine learning. I am particularly interested in the intersection of artificial intelligence (AI) and public policy. To further explore this interest, I completed a remote internship with the Organization for Economic Co-operation and Development (OECD), working within the Directorate for Science, Technology, and Innovation, and more specifically, the Science and Technology Policy Division.
听
Founded in 1961, the OECD is a prominent intergovernmental organization (IGO) headquartered in Paris and made up of 37 member countries. The organization provides a forum for member and partner countries to exchange data and research, shape policies, and establish global standards. Each year, the OECD publishes 300 to 500 books, as well as innumerable reports, working papers, and statistics. I was personally drawn to the OECD on account of their stellar work in the field of AI policy, particularly the creation of the OECD AI Principles, which were adopted by 42 countries.
听
I have long sought a career path that would combine my technical interests and knowledge with my love of writing and politics. Technology policy, which merges the technical with the political and social, is thus of great interest to me. Throughout my master鈥檚, I have sought opportunities to delve into the world of tech policy, working as a research assistant on McGill鈥檚 Tech Informed Policy initiative and even completing a course on the politics of AI at the Universit茅 de Montr茅al. This internship with the OECD allowed me to expand and deepen my understanding of the field and of IGOs more widely, while also giving me the opportunity to apply my machine learning skills to a fascinating, policy-related research project.
听
During my internship, I worked primarily on two projects. The first involved applying text mining tools to funding databases (such as NSERC鈥檚) in order to map and measure various governments鈥� support for AI-related research and development (R&D). While the text mining work had been completed before my arrival, I was able to contribute, through writing and substantial editing, to a 100-page working paper presenting the team鈥檚 results. I also contributed to the paper by performing analyses and creating data visualizations. One example of this involved organizing topics that had been previously identified through a topic modelling exercise into common themes and then performing and plotting the results of a correspondence analysis on these overarching themes. The highlight of the AI project, by far, was the opportunity to write a soon-to-be-published blog post for the OECD AI Wonk blog. Communicating with a non-expert audience is both fun and surprisingly challenging, and the experience made me very grateful to have taken McGill鈥檚 鈥楥ommunicating Science to the Public鈥� course last winter.
听
The second project built upon the first but instead of searching for AI-related documents the goal was to identify research associated with one (or several) of the UN鈥檚 17 Sustainable Development Goals. To aid the OECD in setting up this project, I researched and put together a 25-page literature and methods review describing various approaches undertaken by other organizations and researchers, as well as additional methods that may be of interest to the OECD. I presented my findings at a team meeting, in which I also made recommendations for how to improve upon other groups鈥� approaches, which mainly consisted of relatively simple keyword-based methods. I then developed and tested a machine learning approach to identify research abstracts related to SDG 13 (climate action). This involved building a training set by applying search queries developed by another organization to a large funding database and then manually labeling a subset of the selected and non-selected abstracts. I then trained and tested a Longformer model, a language model capable of accepting long textual inputs, on my dataset, achieving promising results, which I presented in a final report.
听
Remote internships are always a challenge, especially ones involving large time differences. It can be difficult to make personal and professional connections with colleagues, and it鈥檚 natural to mourn the 鈥渨hat if鈥� of spending three months in one of your favourite cities. While there鈥檚 no cure for the latter, I attempted to fix the former by inviting colleagues for virtual coffee dates. Additionally, due to his very demanding schedule and the aforementioned time difference, effectively communicating with my supervisor also proved challenging at times. This led me to take greater initiative than I had in previous internships, designing projects myself and pitching them to him in order to best utilize both of our time. Because of the independence and initiative required to conceive, propose, and complete them, the SDG reports I prepared and the machine learning prototype I tested were major highlights of my time at the OECD.
听
While I did not receive academic credit for this internship, I believe it has played a key role in my education, allowing me to combine my rather disparate skills in an exciting, policy-oriented setting. Moreover, my OECD experience will hopefully serve as a doorway into future tech policy work, be it in Canada or overseas. I want to thank the McGill alumni and the Arts Undergraduate Society for funding the Faculty of Arts Internship Award I received. IGOs are famous for their unpaid (or barely paid) internships 鈥� I am thus doubly grateful to have received funding from McGill to undertake mine.