The Law of AI and Policy Intent

AI policy establishes a unified framework for the government to engage with AI confidently, safely, and responsibly in order to realise the benefits of AI. Laws give effect to policy intent. Attentive policy scrutiny and prompt attention to legislative design can illuminate a path to reach this result. The policy process must proactively consider legal issues. This enables an informed assessment of whether legislative change is the appropriate response to the opportunities presented by Artificial Intelligence (AI) during the Infosphere Revolution.

The state has a spectrum of levers at its disposal along a continuum ranging from soft forms of influence, such as information and public outreach, education, and nudges to shape choices towards desired outcomes, to subsidies, funding, levies, and the more formal prescriptions of power enshrined in Acts of Parliament. Policy may result in changes to the law, including modifications to Acts of Parliament, regulations, and other legislative instruments.

In developing an AI policy, the state needs to assess the following: (1) the impact of any changes to legislation or regulation, (2) when non-legislative alternatives may be a more suitable approach to the opportunity, (3) how to ensure that Cabinet Paper recommendations create a smooth pathway between policy approval and drafting instructions, (4) how policy stewardship obligations apply to maintaining a cohesive set of laws while considering changes, as well as the operational implications of any modifications for the government, (5) transitional provisions, and (6) the outreach and education that might be needed as a result of shifting to a new constellation of legal beliefs, tools and practices.

A necessary part of this process is to support agencies in meeting stewardship obligations by identifying emerging gaps in the law or areas where the current law is becoming outdated, and to assist departments in identifying longer-term reform needs.

Policy must also include provisions to facilitate policy labs and sandboxes that test and evaluate the impact of regulations, technologies, and strategies in practice, providing support for risk management and the transition from theory to reality in controlled environments before migrating to a broader set of beneficiaries.

Policy can bolster public trust in the government’s use of AI by providing clear transparency and risk assurance. Technological change is now so rapid that AI policy must be forward-thinking and adaptable to the government’s use of AI, and it must be designed with an open architecture to evolve and develop over time.

This openness is fundamental, as policy must iterate on global initiatives. Local AI projects must be scalable globally. Policy must make the country an attractive destination for AI infrastructure investment.

An AI Opportunities Policy Prescription can enhance the state’s appeal for AI investment. This will expand the impact of the international AI agenda on the national policy agenda. Policy-driven local AI initiatives must also shape the global agenda.

Policies set the framework for addressing opportunities and achieving specific aims. It does not necessarily contain enforceable rules, but it may identify the need for new laws.

To unlock the innovative use of AI, the state needs a modern and effective regulatory system to address AI’s distinct risks, preventative measures, and risk-based guardrails throughout the AI lifecycle. One area for critical analysis is the scrutiny of interpretative repertoires at the nexus of power/knowledge and how they are embedded in the nearly ubiquitous AI-driven systems that masquerade as neutral and universal.

Technological redlining involves the process by which algorithms create and normalise structural and systematic isolation, reinforcing oppressive social and economic relations. This has resulted in a charter outlining six ethical concerns that now define the conceptual space for ethical AI. Three are epistemic: inconclusive, inscrutable, and misguided evidence.  Two are normative: unfair outcomes and transformative effects. The sixth is traceability, which encompasses both epistemic and normative concerns.

The epistemic factors underscore the relevance of the quality and weight of the data to the justifiability of the conclusions that algorithms reach. The normative concerns pertain to the ethical impact of algorithm-driven decisions, including opacity, unfair outcomes, and unintended consequences.

These six parameters shift the dialogue away from mere deliberations about “bias” and survey a horizon of future visions of AI as creole technology. Policy must therefore allow for evolving AI technologies and locally derived forms of AI assemblages to be combined in original ways, forming hybrids.

Inside of this hybrid frame, AI policy can cater for AI-driven analytics that can lead to better decision making in supply chains, the creative industries, the climate transition, disaster risk reduction, predictive healthcare, e-commerce, improve educational outcomes, enhanced security and efficiency in transportation, reduce losses in water and energy provision, clean energy transitions, access to financial services for the unbanked, and crime reduction.

Humans have learned how to make sense of the world by using films of information layered in hierarchies. Moreover, the world evolves according to an infinite set of possibilities, and the only way to train AIs to deal with this is through abstraction.

AIs are still unable to understand the physical world, have persistent memory, reason, and plan hierarchically in thin layers. Policy must be sensitive to these issues, not because they constitute limitations, but because they provide clues to hazy AI policy horizons.