Leashing AI: the need for checks and balances ISO/IEC 42001:2023
AS ISO/IEC 42001:2023, referred to as 42001, is a new global Artificial Intelligence (AI) management system standard developed specifically for AI systems. It is a management system standard (MSS) that specifies requirements and offers guidance on how to establish, implement, maintain, and continually improve an AI management system. This standard is designed to address the need for consistency and ethical implementation of AI across various borders and contexts.
Purpose and Need for 42001: The primary purpose of 42001 is to guide organizations on how to best manage their AI systems. It provides a framework with built-in checks and balances for responsible AI implementation. The need for 42001 stems from several critical challenges and concerns associated with AI:
• Unique Risks of AI: AI presents unique risks that require careful and systematic management due to its self-learning nature. Unlike traditional software, AI is an "inference machine" trained for decision-making, with or without human involvement. Continuous machine learning AI systems can produce different results over time from the same input, making them distinct from traditional systems with predictable outcomes.
• Governance Ramifications: AI systems' ability to make decisions has serious governance implications, necessitating careful consideration of the decision-making power granted to AI and ensuring that human responsibility and control are maintained.
• Key Challenges: Challenges exist around the transparency of automated decision systems, the degree of AI system autonomy, and the fact that machine-learning systems are trained on data, unlike traditional programming. AI systems are also socio-technical, influenced by societal dynamics and human behavior.
• Trust Deficit: There's a growing concern among consumers regarding the potential risks of AI systems, including perceived opacity, complexity, potential for bias, loss of accountability, and unpredictability. Trust in AI decreased in 25 out of 27 countries in 2021. Standards play a critical role in creating trust and assurance in AI.
• Knowledge Gap and Lack of Codified Principles: Many organizations globally, including Australian ones, have a knowledge gap in managing AI risks. While most companies understand the importance of consumer trust in AI, relatively few have codified these principles into official rules and policies.
Benefits of 42001 for Organizations: 42001 offers numerous benefits to organizations of all sizes and types that provide or use AI systems, helping them develop or use AI responsibly:
• Risk Management: It assists organizations in identifying and mitigating risks in AI technologies, safeguarding against biases, security breaches, and other potential harms. It prompts organizations to conduct impact assessments, understand boundaries, and implement controls and monitoring.
• Ethical AI Implementation: The standard helps embed an ethical approach into an organization's AI management system.
• Improved Decision-Making and Accountability: It promotes transparency and explicability in AI systems, leading to better decision-making processes and clear accountability for AI outcomes.
• Safeguard Against Bias: By prompting organizations to "ask the right questions" regarding data quality, 42001 indirectly helps guard against bias that can infiltrate AI systems through training data. Addressing algorithmic bias is a commercial and social imperative.
• Fostering Public Trust and Transparency: It plays a vital role in fostering greater public confidence in the responsible use of AI by demonstrating trustworthiness through systems and processes. Over 85% of IT professionals believe consumers are more likely to choose companies transparent about their AI models.
• Scalable and Integrable Management System: Its harmonized structure allows for easy integration with existing privacy or cybersecurity management systems.
• Boosting Innovation and Creativity: By acting as a safety net, it helps boost confidence, innovation, and experimentation with AI, enabling organizations to try new things safely and scale internationally.
• Globally Recognized Standard: 42001 provides guidelines for AI governance and management that are recognized globally, which is particularly important for organizations looking to export products or services.
• Systematic Approach: It offers a structured way to address key challenges like ethical use, data privacy, bias, accountability, and transparency in AI technologies.
• Certification: As the first ISO standard defining a certifiable management system for AI, it provides a "trustmark" to the market, signaling an organization's serious commitment to responsible AI use. Certification can offer a competitive advantage, raise public confidence, and facilitate reliance on other organizations' systems, especially in larger entities.
Importance of AI Standards in Australia: AI standards are crucial for Australia to build trust through the responsible use of AI. They provide common guidelines, principles, and technical specifications, addressing the challenges of developing governance frameworks for the scalable and international scope of AI. Consistent international standards enhance interoperability, security, reduce trade barriers, ensure quality, and build public trust in digital products. Standards help bridge the gap between "hard ethics" (legislative policy) and "soft ethics" (community expectations), acting as critical connective tissue in a whole-of-society approach to digital governance. They are analogous to electrical safety standards in ensuring public health and safety with a powerful new technology.
Ultimately, while 42001 is a powerful tool, it emphasizes that human oversight remains critical, as AI systems should not be afforded too much autonomy. People must remain accountable for AI implementation, which is the essence of responsible AI.