Utilised effectively, powerful AI systems have the potential to transform the energy and infrastructure sector. New models can boost business efficiencies, enhance grid stability and speed up the transition to green technologies. However, many businesses in the sector are still grappling with how to unleash the huge opportunities of AI in a cost-effective, risk-managed way.

Of course, AI already proliferates complex systems in the energy sector and beyond. A report by Microsoft and LinkedIn found that over 75% of global ‘knowledge workers’ are already using AI systems. However, this ‘use’ captures everything from ChatGPT to virtual assistants. Truly disruptive changes to core business operations through AI technologies are much harder to come by, and less prolific than the AI hype might have led you to believe.

The ‘how’ of implementing disruptive AI models is the role of data scientists, operational management and machine learning experts, but lawyers can also play a vital part in helping businesses overcome the strategic and legal challenges of deploying powerful AI systems. The risks cannot be understated, particularly as the European Union (EU) has just imposed far-reaching legislation to control firms using AI systems, and specifically those environments defined as ‘high-risk’, which include energy and utilities. In light of this, solid AI and data governance, regulatory awareness and savvy contract negotiations will all be crucial to reaping the full rewards of groundbreaking AI applications.

Energy and infrastructure
The industry is set to benefit from the deployment of powerful AI systems. First, energy creation, distribution and consumption are data-rich activities, with both structured and unstructured data that is ripe for exploration and analysis. Secondly, the sector propagates complex processes across the value chain, from power generation to energy trading. AI will augment and improve technical systems with its powers of prediction and automation, adding intelligence to data which can then inform decision making.

Broadly, it is possible to split AI into two separate use cases. Those which fundamentally transform operations (sometimes called the ‘moonshot’ uses), and systems performing back-office functions. The latter is more straightforward in relative terms, as it does not require huge amounts of technical expertise and consists of automating standard tasks, including virtual admin tools, chat bots for customers or co-pilots for software development. These tools aren’t specific to the energy sector and infrastructure, and will generally be provided by ‘off the shelf’ third party software providers. However, in the context of regulatory and practical uncertainties, businesses need to ensure that risk and liability are fairly allocated between contracting parties.

Moonshot applications present a greater challenge for firms, along with the potential for much greater rewards. One significant opportunity is in predictive asset maintenance. Firms operating complex asset networks incorporating solar and battery energy storage systems (BESS), for example, spend vast sums of money on maintaining these assets. Predictive maintenance models can be trained on integrated data sources, incorporating information from previous corrosion tests, damage reports, visual inspections and current asset sensors. Whilst some UK firms are already developing these capabilities in-house, there are incumbent technology companies providing off-the-shelf solutions for the energy sector.

For example, Above Surveying is a UK firm who uses AI robotics to capture aerial topography, construction monitoring imagery and component visual data to help developers and asset owners track the progress of ongoing projects or assess long-term asset conditions.  

As well as monitoring assets, AI systems can teach themselves to predict complex patterns and ingest far more data points than conventional analytics systems. Flextricity, a UK energy-tech firm, has started harnessing AI and advanced modelling to help its clients optimise returns from BESS and gas peaking plants. This is achieved through algorithms which predict market demand and enable swift adjustments to assets. AI systems, which ingest far more data points than conventional analytics systems, have the ability to forecast these complex scenarios with unprecedented accuracy.

The risks [of implementing disruptive AI models] cannot be understated, particularly as the EU has just imposed far-reaching legislation to control firms using AI systems, and specifically those environments defined as ‘high-risk’, which include energy and utilities.

Smart contracts
One AI application which could yet see a surge in the energy sector is ‘smart contracts’. These are legally binding contracts in which some or all of the contractual obligations are defined and/or performed automatically by a computer programme.

Digitised operations include peer-to-peer energy trading, automated response to demand, and the coordination of electric vehicle (EV) charging. The energy sector is data-rich and this data could form the basis for smart contract use.

For example, smart contract enabled meters could release energy once payment has been received. Another application is coordinating charging stations and EVs of all types (scooters, cars) where drivers pay charging station owners through a smart contract. Consumers could also track the provenance of the electricity they are consuming through smart contracts, having access to information about residential solar panels, stations, wind, offshore and more.

More complex transactions such as trading of energy commodities and peer-to-peer energy trading might also be facilitated through smart contract use, although the relevant regulations would of course apply. One of the major concerns with smart contract use for anything more complex than the basic ‘payment received, product provided’ examples above is the automated nature of the process. Nuance can be challenging to build into the programme, and logical conditioning could tie parties into contract execution without having a human sense check built in.

AI, whilst seemingly a very different type of technology, could bridge this gap. A trained AI standing between the ‘if X, then Y’ case could ensure the ultimate aims of the parties are met. A very well-trained AI could go as far as taking the place of a human expert at the point in contracts where expert determination is called for. This type of AI-enhanced smart contract is still some way off – developers would need to be convinced the market for them exists, and data for training the AIs may be hard to come by. If these obstacles could be overcome, smart contract use could significantly cut down on the administrative burden and improve the transparency of the energy sector.

Key challenges
Data quality is one of the key challenges for data scientists and operations managers looking to develop in-house AI systems. First, data must be collected and cleaned into the right granularity, before being exported to secure data warehouses where it can be used to train machine learning systems. Data breaches and cyber-attacks are a real concern, and all data-rich firms must have robust policies in place to mitigate against these threats. Attacks can potentially expose sensitive data as well as compromise intellectual property that subsists in AI models and/or the data sets they are trained on.

The UK regulatory position on AI also remains unclear. However, it is anticipated that the new Labour government approach will be to decentralise regulation to industry-specific bodies.

There is more certainty within the EU with the advent of the EU AI Act. This came into force in August 2024 and, whilst an EU law, will also apply to any UK firms who ‘provide’ or deploy AI systems with an output in the EU. Systems deployed in essential infrastructure (water, gas and electricity) are categorised as ‘high risk’ under Article 6 of the Act, meaning strict risk management and procedural obligations will apply. Penalties for breaching the regulations are severe – ranging from €7.5mn (or 1.5% of global annual turnover) to €35mn (or 7% of annual turnover), depending on the type of infringement. Clearly, robust internal AI governance policies are fundamentally important to all firms building or utilising third party AI systems.  

Contracting with AI system providers
In the context of ambiguous regulations and the threat of cyber-attacks, the appropriate allocation of risk and liability in software agreements is important. This can apply when procuring an ‘off the shelf’ AI product or training bespoke AI systems for a particular customer. First, you need to know if AI technology is actually being used – sometimes software providers might not be forthcoming with that information. Secondly, is the other party accepting an appropriate amount of risk?

For example, if the model is trained on public image or text databases, business must ensure they are protected against third party intellectual property rights (IPR) infringement claims. Where a supplier’s system could inadvertently introduce bias or faults into consumer-facing programs, businesses must ensure they are not landed with the responsibility for complying with AI regulatory guidance or best practices.

It's clear that the benefits of AI systems in the sector are huge. However, firms need to manage the inherent risks – effective AI and data governance, understanding of regulations and well negotiated contracts are all crucial components of an effective AI strategy.

• Further reading: ‘How AI enhances operations in the energy transition’. Artificial intelligence offers a raft of opportunities for optimising renewable energy operations both upstream in solar and wind farm operations, and downstream for flexible power distribution and consumption.
• Find out how the intelligent use of data offers material benefit to every aspect of energy and utilities, according to James Forrest, Global Industry Leader for Energy & Utilities at Capgemini.