From the inception of the Vector Institute, its community has shared a strong commitment to developing AI that is both safe and trustworthy. It is at the core of our existence. Now, powerful and easily accessible generative AI models have revealed to the general public the short-term and existential risks of AI. As the pace of AI development quickens, policy makers and AI experts around the world are seeking to develop shared guardrails to help guide future development of this technology.
As a central hub of Canada’s AI ecosystem, Vector has distilled six AI trust and safety principles for Canadian and international organizations that reflect the ongoing dialogue, shared values, and best practices of the businesses, governments, and globally-renowned research communities with whom we work.
Following these principles from research through to implementation and/or commercialization will help to ensure that AI systems are safe and trustworthy. Vector will be updating its own Code of Conduct to ensure that these principles continue to shape the institute’s work.
Vector believes these trust and safety principles can provide guidance for other organizations developing their own code of conduct and AI policies. They are a starting point for organizations at the beginning of their AI journey, while more established organizations can build on them to best suit their needs.
By embracing these principles, organizations developing their own AI policies around the world will signal their commitment to harnessing the power of AI as a means to democratic ends.
A strong commitment to these principles across Canada’s AI ecosystem would extend the country’s global leadership in the responsible development and deployment of AI.
The Vector Institute’s unique ability to convene stakeholders across AI domains, institutions, and industry sectors gives the institute a novel perspective on the trends and challenges of implementing AI.
These principles are built on international themes gathered from across multiple sectors to reflect the values of AI practitioners in Vector’s ecosystem, across Canada, and around the world:
We are committed to developing AI that drives inclusive growth, sustainable development, and the well-being of society. The responsible development and deployment of AI systems must consider equitable access to them along with their impact on the workforce, education, market competition, environment, and other spheres of society. This commitment entails an explicit refusal to develop harmful AI such as lethal autonomous weapons systems and manipulative methods to drive engagement, including political coercion.
We are committed to building appropriate safeguards into AI systems to ensure they uphold human rights, the rule of law, equity, diversity, and inclusion, and contribute to a fair and just society. AI systems should comply with laws and regulations and align with multi-jurisdictional requirements that support international interoperability for AI systems.
We recognize the fundamental importance of privacy and security, and we are committed to ensuring that AI systems reflect these values appropriately for their intended uses.
We recognize that maintaining safe and trustworthy AI systems requires the continual assessment and management of their risks. This means implementing responsibility across the value chain throughout an AI system’s lifecycle.
We recognize that citizens and consumers must be able to understand AI-based outcomes and challenge them. This requires the responsible transparency and disclosure of information about AI systems – and support for AI literacy – for all stakeholders.
We recognize that organizations should be accountable throughout the life cycles of AI systems they deploy or operate in accordance with these principles, and that government legislation and regulatory frameworks are necessary.
The Vector Institute’s First Principles for AI build upon the approach to ethical AI developed by the OECD. Along with trust and safety principles, clear definitions are also necessary for the responsible deployment of AI systems. As a starting point, the Vector Institute recognizes the Organization for Economic Co-operation and Development (OECD) definition of an AI system. As of 2024, the OECD defines an AI system as follows:
“An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.”
Vector also acknowledges that widely-accepted definitions of AI systems may be revised over time. We have seen how the rapid development of AI models can change both expert insight and public opinion on the risks of AI. Through Vector’s Managing AI Risk project we collaborated with many organizations and regulators to assess several types of AI risk. These discussions informed the language around risks and impact in the principles.
The dynamic nature of this challenge necessitates that companies and organizations should be prepared to revise their principles as they respond to the changing nature of AI technology.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。