House of Lords Communications and Digital Select Committee inquiry: Large language models
The IEEE Standards Association (IEEE SA) welcomes the opportunity to provide its response to the House of Lords Communications and Digital Committee questions.
IEEE Standards Association (SA) is a globally recognized standards-setting body within IEEE, the largest organization of technology professionals in the world. We develop consensus standards through an open process that engages industry and brings together a broad stakeholder community.
As stated in the inquiry request, IEEE SA would like to provide the following responses to the set of questions the Communications and Digital Committee has posed.
IEEE SA released a Communique, 9 June 2023 on Generative Artificial Intelligence Applications, which notes that the deployment of large language models and other generative artificial intelligence (AI) applications has precipitated a worldwide conversation about the benefits and possible harms of AI systems, as well as the need for appropriate standards and measures that should be taken to ensure scientific integrity and the safety of the public.
Emerging and new technology comes with unknown obstacles and unintended risks requiring accountable design and lifecycle planning to ensure ethical and responsible innovation. Forecasting the impact that Large Language Models (LLMs) may have on institutions, societal cohesion, and the mental health of our children, should not be understated when considering the importance of advances in AI research and development.
Noting the scale and scope of AI and LLMs and indeed the whole technology that is seen as the Fourth Industrial Revolution, the adoption of a set of Principles to guide research and development should be a priority.
IEEE supports the principle that design, development and implementation of AI should be undertaken with consideration for the societal consequences and safe operation of systems with respect to: Human Rights, Well-Being, Data Agency, Effectiveness, Transparency, Accountability, Competence and Awareness of Misuse.
It will be vitally important to adopt a balanced regulatory risk framework and a set of standards to ensure that no social group or community is unduly impacted and that includes ethical alignment with a clear set of principles that addresses the emergence and proliferation of Open Access LLMs.
The UK’s framework and principled approach to AI is reasonably comprehensive in so far as it considers most critical ethical facets of accountability, bias, transparency, and fairness. However there appears to be a gap in an integrated approach for consistency, coverage, best practice cross-pollination and overall strategic UK wide perspective and impact.
Delegation of responsibility appreciation of the extensive challenges of the AI technologies in complexity and novelty can set an unreal expectation unless there are provisions to both apply ethical principles and to encourage the development of standards, codes of best practice and sandboxes as a means of addressing the ecosystem to support the private and public organizations in interpreting, scoping, scaling, and implementing LLMs within the UK.
We suggest that he risks of unintended consequences should be balanced against the potential rewards/gains for specific affected groups, through continuous monitoring, evaluation, assessment, and compliance. While global participation is encouraged to offer the broadest regional and cultural perspectives required to offer greatest operational benefit, legislation and regulation is developed and applied country by country. The recognition of global standards for the implementation of LLMs does create a globally uniform foundation for each country to best approach how it will apply its regulatory and enforcement framework to emerging technology and will minimize regulatory divergence.
IEEE SA has several standards that may be of interest as the House of Lords, Communications and Digital Committee looks at the opportunities and risks posed by LLMs.
They include:
IEEE P7009™ Standard for Fail-Safe Design of Autonomous and Semi-Autonomous Systems that establishes a practical, technical baseline of specific methodologies and tools for the development, implementation, and use of effective fail-safe mechanisms in autonomous and semi-autonomous systems. [still under development]
IEEE 7000-2021™ Standard Model Process for Addressing Ethical Concerns during System Design which establishes a set of processes by which engineers and technologists can include consideration of ethical values throughout the stages of concept exploration and development, which encompass system initiation, analysis, and design. This standard provides engineers and technologists with an implementable process aligning innovation management processes, system design approaches, and software engineering methods to help address ethical concerns or risks during system design.
IEEE 7001-2021™ Standard for Transparency of Autonomous Systems which establishes measurable, testable levels of transparency, so that autonomous systems can be objectively assessed, and levels of compliance determined.
IEEE 7002-2022™ Standard for Data Privacy Process which contains requirements for a systems/software engineering process for privacy-oriented considerations regarding products, services, and systems utilizing employee, customer, or other external user’s personal data.
IEEE 7005-2021™ Standard for Transparent Employer Data Governance which contains specific methodologies to help employers in accessing, collecting, storing, utilizing, sharing, and destroying employee data, including specific metrics and conformance criteria regarding the types of uses from trusted global partners and how third parties and employers can meet them.
IEEE 7007-2021™ Ontological Standard for Ethically Driven Robotics and Automation Systems which contains a set of ontologies with different abstraction levels that contain concepts, definitions, axioms, and use cases that assist in the development of ethically driven methodologies for the design of robots and automation systems.
IEEE 7010-2020™ Recommended Practice for Assessing the Impact of Autonomous and Intelligent Systems on Human Well-Being which provides specific and contextual well-being metrics that facilitate the use of a Well-Being Impact Assessment (WIA) process in order to proactively increase and help safeguard human well-being throughout the lifecycle of autonomous and intelligent systems (A/IS).
We would look forward to further discussions with the House of Lords Communications and Digital Committee on Large Language Models (LLMs).
5 September 2023
3