Management System Standard (MSS) for AI

• AS ISO/IEC 42001:2023 is a globally recognized standard that provides guidelines for the governance and management of AI technologies.

Introduction

• AI is transforming communities, cultures, and economies worldwide.
• Generative AI may increase global GDP by 14% (US$15.7 trillion) by 2030
• Global AI adoption is steadily growing, with half of organizations reporting substantial benefits, including efficiency, innovation, and profitability gains.
• Concerns regarding gender bias, privacy breaches, and racial profiling accompany the benefits of AI.
• Consumers are increasingly concerned about AI systems' opacity, complexity, potential for bias, loss of accountability, and unpredictability.
• Trust in AI decreased in 25 of 27 countries in 2021.
• Trust is crucial for the success of any organization and can be easily shattered.
• Governments and industry are taking concerted actions to create a credible framework for trustworthy Al.

AI Standards as a Critical Watchdog

• Growing concerns have spurred a worldwide move to regulate AI.
• The EU AI Act aims to protect fundamental rights, democracy, the rule of law, and environmental sustainability from high-risk AI.
• Governments globally are responding to the need for AI standards, including the US National Institute of Standards and Technology (NIST)'s AI Risk Management Framework (RMF).
• Australia consumes more AI products than it produces, closely monitoring responses from other countries like the EU, UK, US, and Canada.
• Standards have a critical role in creating safeguards to manage risks, mitigate harms, and build trust in AI systems.

AS ISO/IEC 42001:2023 - A New Way Forward

• AS ISO/IEC 42001:2023 is a global AI management system to drive the responsible use of AI.
• 42001 addresses the need for consistency and ethical implementation of AI across borders.

About This Report

• The report aims to inform about 42001, its benefits, and its impacts on Australian organizations and the community.
• The report seeks to explain the need for management system standards (MSS), provide background on 42001, the benefits of the standard, and the need for AI standards in Australia.

What is a Management System Standard (MSS)?

• Management system standards (MSS) support organizations in implementing an integrated system for areas like health and safety, environmental issues, governance, risk management, and training.
• MSS helps put an integrated system in place with senior management support, training, governance processes, and risk management.
• There are over 80 ISO MSS covering organizational culture and processes.
• Examples: ISO 9001:2015 (Quality management), ISO/IEC 27001:2022 (Information security), and ISO 14001:2015 (Environmental management).

Benefit of an ISO MSS

• An ISO management system standard can help organizations by:
• Specifying clear steps for achieving company objectives and goals.
• Assisting with risk assessment and system impact assessment.
• Helping establish a healthy organizational culture.
• Delivering better services/products and increasing customer value.

Why do we need an MSS for AI?

• AI presents unique risks due to its self-learning nature, unlike traditional software and IT systems.
• The automated, complex, self-learning, and scalable nature of AI systems poses particular risks beyond regular software.
• A challenge is AI systems could produce different results over time from the same input.
• Al can adapt and use data instead of human-coded logic, emphasizing data quality and protection; Al could make decisions, requiring governance and proper decision-making power.

Key challenges of Al

• Relative transparency of automated decision systems.
• Degree of AI system autonomy.
• Use of data to train machine-learning systems.
• AI systems are inherently socio-technical.
• AI risks and benefits emerge from the interplay of technical aspects and societal factors related to how a system is used, its interactions with other AI systems, who operates it, and the social context.

Importance of Guardrails for AI

• Risks organizations face without appropriate AI safeguards:
• Non-compliance with data protection and privacy regulation.
• Over-reliance on AI for automation - requires human oversight/review.
• Reputational damage from AI misuse in products/services.
• Discrimination/bias from lack of training data oversight/review.
• Inaccurate insights or faulty decisions from training data quality, model design, training approach, or improper model usage.

Background and Development of 42001

• The AI ecosystem is ripe for standardization.
• Digital transformation has fundamentally changed the landscape for IT standardization.
• Non-technical requirements such as ethical considerations and designing trustworthy AI systems are increasingly important.
• Data ecosystem importance alongside hardware, software, and operational technologies is also growing.
• The increasing importance of certification and third-party audits
• Due to AI regulation gaining momentum, the development of 42001 became critical.
• The MSS approach was leveraged, similar to ISO 9001, which specifies requirements for a quality management system.

What is 42001

• A MSS developed specifically for AI.
• Specifies requirements and guides establishing, implementing, maintaining and continually improving an Al management system.
• Given Al's wide reach, the MSS will likely become integral to organizational success, like ISO 9001, ISO 14001 and ISO/IEC 27001.

Who is the standard for?

• The broad standard is applicable to any organization that provides/uses AI products/services, helping develop/use AI systems responsibly.
• It does not replace existing frameworks but serves as a complementary document, with Australia´s Al Ethics Principles.
• Four fast facts:
  • Based on a common "high-level structure".
  • Promotes continual self-evaluation and accountability Iteratively.
  • Organizations can be certified by a third party.

What is the purpose of 42001?

• 42001 guides organizations to manage AI systems and fosters trustworthy, responsible AI practices.
• 85% of IT professionals agree consumers choose companies adopting transparency in AI model design.

Why do we need 42001?

• The vast majority (82%) of companies believed they were taking best practices for responsibly using AI in their business are.
• Less than a quarter (24%) had any measures in place to ensure that was what they were actually doing.
• Most companies understand the vital need to ensure consumer trust in the way organizations use Al.
• Few have formalized these principles.
• 42001 helps organizations provide a framework to address challenges associated with AI implementation, promoting responsible and transparent AI use

How can 42001 help Australian Organisations?

• This management system standard helps address the "potential risks in an unregulated AI system".
• It requires impact assessments, system controls, monitoring, and “brakes" to ensure compliant decisions.
• The technology brings new opportunities and responsibilities.
• It's all about how powerful can be applied to Al.

Asking the Right Questions

• Prompts Australian organizations to "ask the right questions".
• Standard set out different things to consider, such as required resources and policies.
• It helps organizations miss less out on important steps.
• It is a useful framework
• Controls implementation that is appropriate in line with values and objectives.

As A Safeguard Against Bias

• 42001 also “helps, indirectly, ensure data quality and guard against bias, because it prompts questions.”
• Human bias can infiltrate Al through data.

Benefits of 42001 for Australian Organisations

• The Al Management Standard is a systematic approach that can boost innovation and scale internationally.
• It provides a framework for risk management that can integrate flexible management systems that are globally recognised.
• Helps organizations embed an ethical approach to their AI management system,
• Helps promote transparency and explicability in AI systems
  • Enables decision-making processes
  • Accountability for Al outcomes.

Benefits of 42001

• Provides certification, signalling commitment to the responsible use of AI.
• Improves quality, security, traceability, transparency, and reliability of AI.
• Meets customer, staff, and stakeholder expectations around responsible AI use.
• Boosts efficiency and risk management.

Benefits of Certification

• Sends a signal that an organization takes Al ethics seriously.
• Helps improve quality, security, traceability, transparency and reliability of Al technologies.
• Helps improve efficiency and risk management.

Certification Through 42001 as a Mark of Trustworthiness

• As the first ISO standard that defines a certifiable management system for Al, it provides organizations with a mark of trust that can promote commercial scalability across all critical infrastructure sectors.

Why Certification Matters

• Governments worldwide want to ensure through Al Trust.
• To regulate this:
  • The EU
  • The UK
  • China
• Compliance to regulatory measures is reliant on accreditation and digital governance standards,

Certification Acts as a Trustmark in Australia

• Organizations will use certification to show they are doing AI responsibly.
• Companies and departments expect conformance from the standards.
• Certification in Australia has a key purpose to make sure:
• Organizations themself implements Al properly and responsibly.
• Organizations they deal with use AI in a responsible manner.

Benefits of Certification

• Affects organizations and especially those who would export standards and practices globally.
• Standards benefit all key factors like aged care, health, agriculture and mining.

Potential Benefits of Certification with 42001

• It promotes accountability to the ethical use of Al.
• Provides a competitive advantage.
• Raise public confidence by auditing through an independent body.

The Al Standard is a Roadmap

• Al standards provide an approach that manages its adaptive response.
• Enable organizations to perform evaluation, benchmarking and audits.
• Standards promotes design, usage and helps to protect unauthorized access.

Bridging the gap

• The whole of society must adopt its practices.
• Connect "Hard Ethics" from the government with "Soft Ethics" arising in civil ways.
• Standard also helps users to have transparency and understanding.
• Unless Al users can align to the principles everyone are guessing: standards close up that gap.

Assessing Bias in AI systems

• It’s systems is built on information from Al system training.
• Data can provide human decisions and indicate social inequalities.
• Specifies all measure biases for vulnerabilities and help to address these challenges and metrics effectively.

Conclusion

• 42001 is a tool to limit risks of Al and enhance accountability.
• Must ensure people are still in control
• Must use standards and interventions to manage Al at a high standard.