Artificial intelligence is poised to change the world. Its potential benefits touch on everything from speeding up the development of effective pharmaceuticals to the early detection of extreme weather events. But the potential risks surrounding AI, including the amplification of misinformation and the erosion of privacy, are alarming.
Rebecca Finlay, BA’88, is the CEO of Partnership on AI (PAI), which works to “ensure developments in artificial intelligence advance positive outcomes for people and society.” The global nonprofit is supported by the Ford, Knight and MacArthur foundations. It was created with the backing of six tech giants – Amazon, Apple, Facebook (now Meta), Google/DeepMind, Microsoft and IBM.
Finlay says that AI itself is not inherently good or bad.
“It’s a product of the choices we make as individuals, as companies, and as governments about the way in which we want to use it and the way in which we want to build it and deploy it and build safeguards around the way in which it’s used.” PAI exists, in part, to explore what sorts of safeguards we should be considering.
“The one thing that we always need to think about when we think about AI systems is the data that the systems are trained on, [which] are reflected in how well the systems work,” says Finlay. “If someone is developing a system and not attending to the data and the biases that sit within that data, that’s where it can move to potential harms.”
The Toronto-based Finlay became the head of PAI in 2021. Before that, she was the vice-president of engagement and public policy at the Canadian Institute for Advanced Research (CIFAR), where she launched a program that focused on AI’s impact on society.
“If someone is developing a system and not attending to the data and the biases that sit within that data, that’s where it can move to potential harms.”
Rebecca Finlay, CEO of Partnership on AI
Partnership on AI involves 140 partner organizations (CIFAR is one of them) in 17 countries. These partners are diverse – some come from academe (the MIT Media Lab and the Berkeley Center for Law and Technology, to name two), some are NGOs and media organizations (the BBC, UNICEF, the Wikimedia Foundation), and some come from industry (OpenAI, Sony). PAI partners bring a range of interests and concerns to the table, as well as different sorts of expertise.
“PAI is unique in our ability to bring together stakeholders from across sectors to explore how AI can be developed and used safely and responsibly,” says Finlay. “We develop resources and guidelines, we publish frameworks and recommendations, all with the intent of driving better industry practice, informing better public policy and advancing citizen understanding.” She compares the burgeoning AI industry to the automobile sector, where manufacturers comply with rules and standards that emphasize safety.
In an interview on the MIT Sloan Management Review podcast Me, Myself, and AI last year, Finlay offered another description of PAI’s mission. “We do think it’s really valuable to be helping those organizations that do want to be good and responsible actors to know what ‘good’ looks like.”
While her studies at McGill focused on history and political science, Finlay says courses with professors like Charles Taylor and James Tully helped prepare her for the work she does today.
Taylor and Tully were among a “really remarkable group of thinkers who were trying to better understand and explain and interrogate the social systems and structures, the politics, the economics and social norms that drive society forward,” Finlay says. “And so, I’ve always been interested in the role of technology in society and how it both shapes human interactions and communications and our media.”
One recent PAI initiative is the Synthetic Media Framework, a project aimed at establishing guidelines for the development and sharing of audiovisual content that has been created or modified by AI. The framework emphasizes the importance of transparency and informed consent, while calling for the use of content labels, watermarking, and other measures to make it clear when content has been AI-generated.
Finlay says PAI’s partners played an instrumental role in developing the framework. A number of them, including Adobe and CBC/Radio-Canada, contributed case studies that delved into real-life situations related to the issues covered by the framework.
“We received institutional support from across the partnership – including tech companies, social media platforms, and media organizations,” says Finlay. “While this guidance is not a silver bullet in solving the issue of deepfakes or misinformation, the support for the framework indicates the organizations’ commitment to disclosure, transparency, and consent – and the case studies have brought to light how these work in practice.”
An emerging AI challenge, Finlay says, is “agentic AI” – AI-powered assistants that can, for example, plan your next trip and make all the arrangements.
Such assistance “becomes much more potentially risky when you get into things like making doctor appointments for you, where you might be providing access to personal health care information,” Finlay says. “It opens up a whole set of possibilities — wouldn’t we all like to have that kind of assistance in our pockets? — and a whole new set of questions.”
