R³AI: Research Regroupements to support the major advancement of AI knowledge

The aims of this Regroupement will be to formalize the functions of conscious processing (generalization to new contexts – robustness, knowledge representation and causal reasoning, metacognition or reflection on thoughts) in order to advance AI and stimulate the development of new explanatory theories for cognitive neuroscience. These theories can then be experimentally tested and used in neurocognitive research. These new R³AI algorithms will be deployed and tested in IVADO’s various application domains, including AI for science, healthcare systems and supply chains.

The aim of the Regroupement will be to develop cutting-edge research on machine learning algorithms, deep learning and decision making under uncertainty. This Regroupement will also be invited to respond to the problems of other Regroupements and to collaborate closely with them, with the aim of accelerating the development of more robust, better reasoning and more transparent algorithms.

The aims of this Regroupement are to design a new generation of ML algorithms for NLP whose architecture is more structured, in terms of knowledge elements that can be recombined. Applying these new systems and working in collaboration with minorities to mitigate the prejudice and discrimination associated with modern NLP and towards systems that can be guided by humans to understand stories, with a particular focus on the stories and learning of indigenous cultures. Research in this Regroupement should be able to address the following issues: (1) facilitating the integration of natural language instructions, social or moral norms, or verbalized knowledge ontologies into R³AI systems, (2) developing systems that allow AIs to be subject to moral principles (in the spirit of Asimov’s Laws of Robotics) and (3) examining indigenous narratives about land and biodiversity.

This Regroupement has four main goals that need to be worked on in collaboration with minority or marginalized groups: (1) informing an accountable process while ensuring greater representation of marginalized groups and leading to more equitable and safer design, implementation, and outcomes of AI, (2) understanding the new business models that R³AI could help support, identifying the human factors that contribute to the use of AI, defining transparent and reliable governance models that avoid concentration of power, for the responsible and environmentally sustainable use of AI, and the data needed to calibrate AI models, and determine how AI can create better and more meaningful work, (3) study the drivers of AI adoption, uncover implicit human responses to AI systems, and correlate these responses with design choices made by developers, (4) clearly define trustworthy and safe AI, and produce methods, tools, and technologies that enable the development of trustworthy intelligent applications, (5) propose governance norms and models for democratic societies that are resilient to the impacts, harms, and risks of increasingly powerful AI.

The aim of this Regroupement is to develop cutting-edge research in this field and explore pressing ethical questions, while ensuring that their findings can be integrated into the research of other Regroupements. Among other things, this Regroupement will work on the various parameters that impact the robustness of algorithms and induce biases (e.g. the data collected), as well as on explainability. Working methods based on co-construction and including minority groups and indigenous communities will also be examined by this Regroupement. Research on ethical aspects of AI and the social implications of R³AI is included. Finally, research interests will also need to reflect the priorities of indigenous communities so that they can be included in R³AI.

The goal of this Regroupement will be to transform the discovery of molecules and materials by replacing brute force screening with AI-based modeling and experimental design methods, allowing a more targeted but diversified search in both model space and the space of possible experiments. R³AI models will incorporate physico-chemical knowledge and be integrated into classical drug discovery pipelines with active learning loops to (1) support search in molecular space, (2) discover causal models involving drug targets, genes and proteins in cells (for target identification), and molecular interactions, (3) enable prediction of molecular structure and properties, and (4) optimize synthesis strategies. A similar methodology will be used to efficiently explore the phase space of new materials for batteries, energy storage or carbon capture.

The goals of this Regroupement will focus on (1) developing algorithms capable of learning from ever-increasing data sets (drones, transportation companies, etc.) and providing analyses to develop effective climate change and biodiversity policies, (2) understanding biodiversity loss in the context of climate change, (3) designing the active learning pipelines needed to guide targeted drone surveys over areas undergoing vegetation change to minimize epistemic uncertainty about, for example, the future fate of carbon and nitrogen stocks and fluxes in Canada and beyond.

The two primary goals of this Regroupement are: (1) to develop and validate a value-based integrative framework to guide the development and deployment of R³AI in health care and to develop guidelines for policymakers and AI developers to support R³AI in health care systems, and (2) to study cases of AI development and deployment in a variety of contexts to empirically explore a) the perspectives and assumptions that guide developers’ work, (b) the expectations and claims that potential users, including minority groups, have about AI for specific innovations, and c) the expectations and strategies that users mobilize in a real-world context to adopt, implement, and assess the value of an innovation.

This Regroupement will integrate modern optimization and machine learning methods to improve the efficiency and resilience of supply chains and mobility systems, while reducing their environmental footprint. The goal will be to exploit machine learning: 1) in the design of optimization models that will make these systems more reactive to changing conditions; 2) to accelerate the optimization of extra large systems that account for economic, societal, environmental and/or human context; 3) to quantify uncertainty and make supply chains more resistant to disruptions caused by natural disasters, pandemics, wars and other global events.

This Regroupement will integrate modern optimization and machine learning methods to improve the efficiency and resilience of supply chains and mobility systems, while reducing their environmental footprint. The goal will be to exploit machine learning: 1) in the design of optimization models that will make these systems more reactive to changing conditions; 2) to accelerate the optimization of extra large systems that account for economic, societal, environmental and/or human context; 3) to quantify uncertainty and make supply chains more resistant to disruptions caused by natural disasters, pandemics, wars and other global events.