AI in the energy sector guidance consultation

2.1 The government identified the following five principles to guide and inform the responsible development of AI in all sectors of the economy.

a. Safety, security and robustness: AI systems should function in a robust, secure and safe way throughout the AI life cycle, and risks should be continually identified, assessed and managed. 

b. Transparency and explainability: AI systems should be appropriately transparent (communication of appropriate information about AI system to relevant people) and explainable (extent to which it is possible for relevant parties to access, interpret and understand AI system decision making processes). 

c. Fairness: AI systems should not undermine legal rights of individuals or organisations, discriminate unfairly against individuals (whether directly or indirectly) or create unfair market outcomes. 

d. Accountability and governance: governance measures should be in place to ensure effective oversight of supply and use of AI systems, with clear lines of accountability established across the AI life cycle. 

e. Contestability and redress: users, impacted third parties, and actors in the AI life cycle should be able to contest an AI decision or outcome that is harmful or creates material risk of harm. 

2.2 Ofgem considers the five principles are a way to protect consumers from AI harm. We also consider that some of these principles are outcomes whereas others are either a subcategory of a particular outcome or a means to achieve those outcomes. We have taken the government’s five principles and interpreted them into four distinct principles for the energy sector. We think that ethical use to AI relates to its application being safe, secure, fair and environmentally sustainable. We interpret the remaining government principles as follows in the context of the energy sector:  

a. robustness helps to deliver safe and secure use of AI 

b. effective transparency and explainability are a subset of fairness and can help achieve safe, secure and robust application of AI 

c. effective governance and accountability of AI systems are a means to achieve safe, secure and fair outcomes 

d. contestability and redress enable impacted parties to contest harmful outcomes  

Safe, secure, fair and sustainable AI 

2.3 As an outcome focused regulator, our ethical approach to the adoption of AI builds on the government’s AI principles and aims to deliver safe, secure, fair and sustainable outcomes for consumers. 

2.4 AI technology is a continuation of the development of broader digital technologies. It builds on the foundations of hardware, software and data to create a novel capability.

2.5 As well as the expectations around the regulatory framework and good practice outlined in this document, existing good practice covering hardware, cloud, software, supply chain, security, data good practices, standards, and associated regulations remain applicable. This includes the full life cycle management for each of the constituent assets.  

2.6 Stakeholders are expected to ensure appropriate oversight and controls are in place that are relevant to both the AI and its business use to make sure reasonable steps are taken to minimise any negative impact on safety, security, fairness and sustainability. We summarise the risk around each of these in turn below, with key considerations of how to address them.  

Safe AI 

2.7 AI could be used in systems ranging from applications where failure will result in negligible impact on safety through to its use in decision making for interlinked critical national infrastructure which could, for example, result in power outages if not effectively managed. Assessments of AI use must therefore consider the context of operation across the energy value chain and encompass appropriate use, and account for potential misuse (both intentional and otherwise).  

Secure AI 

2.8 AI can present novel security risks alongside standard cyber security threats and risks due to the high pace of development and rate of change for this emerging technology. As a result, security should be considered at the outset and throughout the AI system life cycle.  

Fair AI  

2.9 AI can introduce unintentional bias that can result in direct or indirect discrimination. Ensuring fairness in AI outcomes is crucial for fostering consumer trust and increased consumer confidence in AI use across the energy sector. Stakeholders must ensure AI systems serve energy consumers fairly and transparently by implementing robust governance, systems and processes.  

Sustainable AI 

2.10 AI is viewed as playing an important role in the UK’s economic development, national security and improving our energy system efficiency and sustainability. However, the growth of AI and other data intensive technologies is predicted to consume relatively large amounts of electricity due to the need for additional data centres and computing power. 

2.11 Ofgem recognises that AI operates within a complex and often fragmented landscape. Policy regarding its environmental sustainability will continue to evolve. We are committed to encouraging innovation while raising awareness about sustainable practices through collaboration and where necessary the creation of guidance to help the UK to meet its net zero target and other associated targets. 

2.12 To deliver sustainable AI outcomes requires stakeholders to adopt good practice in the areas of policies, governance, and through life risk management including design, development, deployment, operations, monitoring, maintenance and decommissioning after use. 

2.13 Delivering safe, secure, fair and sustainable AI outcomes requires stakeholders to comply with their regulatory obligations and to adopt good practice approaches to governance and policies, risk and AI implementation throughout the AI life cycle including design, development, deployment, operations, monitoring, maintenance and decommissioning after use. 

2.14 Through appropriate governance, robust risk management, and the right capability across the AI life cycle, organisations should be able to manage AI in line with this ethical approach. We outline good practice expectations in relation to each of these in the following three sections.