Grant: Discovery Grants Program - Individual
University of Waterloo
$41,000.00
Responsible and Equitable AI: Tools for Model Owners and Data Subjects
EF 2024-25
Résumé
Recent breakthroughs in deep learning research have catalyzed an intense commercial interest in deploying entirely new categories of products. The result can be seen in the wide availability of expressive new chatbots and high-fidelity text-to-image synthesis. The massive reach of these emerging technologies, coupled with their extremely data-intensive training process, raises new concerns about their societal impacts. When a single model is deployed to a global audience, there is a need to ensure that the system provides the same level of utility to all users, and does not codify existing social inequities. Meanwhile, despite the promise of increased economic productivity, everyday people often feel a reduced sense of agency as AI tools are deployed in workplaces and the public sphere. These power asymmetries motivate my research, which is concerned with ensuring that emerging AI methods convey equitable and just societal outcomes.
Machine learning algorithms are the workhorse of modern AI. They turn data into expressive computer programs ("models") that represent the world through perceptual inputs, and can be used to derive new predictions, decisions, or even synthetic data points. This data dependency drives AI capabilities, but also introduces new risks, such as the replication of existing societal biases, or memorization of sensitive personal information. AI systems can also fail unexpectedly when deployed in new settings that deviate too much from the training data. To improve the reliability and fairness of emerging AI systems, this Discovery Program seeks to build new tools to assist two broadly defined groups of AI stakeholders. For "model owners", the private firms and public institutions who are actively deploying AI systems, new understandings into the complex failure modes of deep neural networks will be generated, along with efficient strategies to correct these shortcomings. For "data subjects", the everyday consumers and workers whose data is used to train AI systems, methods of collective data action will be proposed that help guide the behaviour of privately-held AI models towards shared community values, even without direct model access.
Canada has been a consistent leader in deep learning research, as well as proactive efforts to ensure its responsible adoption. However, rapidly advancing AI capabilities require a new line of technical inquiry into the limitations of current methods, and how their reliability and fairness can be improved. Accordingly, this Discovery Program is expected to have significant impact for several distinct groups of stakeholders within the Canadian AI ecosystem, including policy-makers, industry practitioners, and ordinary citizens who are affected by AI's deployment. Trainees in the program will develop an expertise in deep learning and its societal implications, and will be capable of contributing to the growing AI industry in Canada and abroad through responsible AI development.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2024-05116
Name nom
Creager, Elliot
Department département
Electrical and Computer Engineering
Organization id
33
Institution établissement
University of Waterloo
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
41000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
1200
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Advancement of knowledge
Area of application fr
Avancement des connaissances
Research subject code
2800
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Artificial intelligence (computer vision, use2603)
Research subject fr
Intelligence artificielle (vision artificielle, utiliser 2603)
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Responsible and Equitable AI: Tools for Model Owners and Data Subjects
Keywords
algorithmic fairness; out-of-distribution generalization; distribution shift; robust deep learning; machine learning
Application summary
Recent breakthroughs in deep learning research have catalyzed an intense commercial interest in deploying entirely new categories of products. The result can be seen in the wide availability of expressive new chatbots and high-fidelity text-to-image synthesis. The massive reach of these emerging technologies, coupled with their extremely data-intensive training process, raises new concerns about their societal impacts. When a single model is deployed to a global audience, there is a need to ensure that the system provides the same level of utility to all users, and does not codify existing social inequities. Meanwhile, despite the promise of increased economic productivity, everyday people often feel a reduced sense of agency as AI tools are deployed in workplaces and the public sphere. These power asymmetries motivate my research, which is concerned with ensuring that emerging AI methods convey equitable and just societal outcomes.
Machine learning algorithms are the workhorse of modern AI. They turn data into expressive computer programs ("models") that represent the world through perceptual inputs, and can be used to derive new predictions, decisions, or even synthetic data points. This data dependency drives AI capabilities, but also introduces new risks, such as the replication of existing societal biases, or memorization of sensitive personal information. AI systems can also fail unexpectedly when deployed in new settings that deviate too much from the training data. To improve the reliability and fairness of emerging AI systems, this Discovery Program seeks to build new tools to assist two broadly defined groups of AI stakeholders. For "model owners", the private firms and public institutions who are actively deploying AI systems, new understandings into the complex failure modes of deep neural networks will be generated, along with efficient strategies to correct these shortcomings. For "data subjects", the everyday consumers and workers whose data is used to train AI systems, methods of collective data action will be proposed that help guide the behaviour of privately-held AI models towards shared community values, even without direct model access.
Canada has been a consistent leader in deep learning research, as well as proactive efforts to ensure its responsible adoption. However, rapidly advancing AI capabilities require a new line of technical inquiry into the limitations of current methods, and how their reliability and fairness can be improved. Accordingly, this Discovery Program is expected to have significant impact for several distinct groups of stakeholders within the Canadian AI ecosystem, including policy-makers, industry practitioners, and ordinary citizens who are affected by AI's deployment. Trainees in the program will develop an expertise in deep learning and its societal implications, and will be capable of contributing to the growing AI industry in Canada and abroad through responsible AI development.