Embedded Files
LEARNING OBJECTIVES
LEARNING OBJECTIVES
By the end of this unit, you should be able to:
Articulate why technical mechanisms are essential for effective AI policy and governance.
Describe how technical tools and metrics, such as compute governance and model evaluations, can be integrated in policy to achieve policy objectives
Explain the role Canada might be able to play in relation to technical AI governance, considering the development of most frontier models takes place outside Canada
LESSON CONTENT
LESSON CONTENT
Slides: Technical AI Policy
RESOURCES
RESOURCES
AI Governance needs Technical Work by BlueDot
Read section “Types of Technical Work in AI Governance” and onwards.
This piece provides an overview of the role that technical work plays in the regulation of AI. It gives examples and insights into hardware, software, and information security examples of technical AI safety work.
The Role of Compute in AI Governance by GovAI
This article demonstrates the role compute-related policy plays in regulating AI. Although most frontier model developers are currently concentrated in the US, Canada can still leverage its technical expertise and research talents to assist in developing or implementing hardware-based safeguards. This is especially relevant in the case that Canada goes on to play a mediating role in international AI governance.
Extra (fun) resource: Youtube Video on the global AI chip supply chain and related geopolitics (Length: ~16 mins)
Feel free to watch this (if you're interested) to get a better understanding of the supply chain dynamics of AI chips as well as other intricacies within the industry.
Understanding the landscape of AI Safety Evals by AI Policy Perspectives
This piece provides insight into the distinction between capability and safety evals, while highlighting many evals currently being used or developed in the field. It also shows what kinds of risks can be addressed using evals. Use this resource as a way to get familiar with the kinds of evals currently available and think on how they might interact with your policy objectives.
(Optional) Vector Institute Benchmarks
This is a recently released, Canadian-made benchmark made to help make LLMs safe and trustworthy.
(Optional) EpochAI Benchmarking Hub
This platform allows you to play around and understand comparative scores on capabilities benchmarks across all the frontier models currently available, including Mistral and DeepSeek. Playing around with it can give you some idea of the benchmark landscape that often dominates the technical discussion around model capability in online spaces like X, Substack, and AI forums.
Additional links:
Examples of technically-aware approaches to AI Governance: Safety Cases for Frontier AI, IDs for AI Agents, Infrastructure for AI Agents
REFLECTION QUESTIONS
REFLECTION QUESTIONS
The readings provide several examples of technical analysis and tools that can inform AI policy (e.g., compute monitoring, benchmarking, analysis of policy reports). If you were advising a policymaker, how would you recommend they integrate these technical perspectives into Canadian policy and regulatory frameworks to ensure they are effective?
Even though things like the GPQA, Frontier Math, SWE-Bench and other such capability benchmarks do not factor directly into policy, what role can they play for you as a policymaker/advocate worried about the impacts of frontier level AI systems?
What role might technical governance play in relation to the use of frontier models in Canadian businesses, enterprises, and even government?
SUPPORTING MATERIALS
SUPPORTING MATERIALS
The following materials offer some more insight into various kinds of technical work that plays a major role in AI policy, and could soon become relevant to Canadian AI policy as well:
National AI Registries - Hadfield et al.
Taxonomy of AI-Bio Safety Evaluations - Frontier Model Forum
LLM Safety and Bias Benchmark List - EvidentlyAI
Lesson Plan Template - Unit 4
Lesson Plan Template - Unit 4
Page updated
Report abuse