Generative AI systems such as ChatGPT have become famous for their ability to learn the patterns behind data to create realistic new variations of existing data. Well known applications range from auto-generated translations to entire books co-authored by machine. What is less well known, however, is that recent advances in generative AI could also unlock the immense data resources across the energy industry to transform everything from regulatory compliance and maintenance to innovation and workforce skills.
Today, workers in industries such as the nuclear power generation sector have to manually search millions of documents for records of everything from maintenance checks to design changes and stay abreast of rapidly evolving protocols and regulations. The data is often fragmented and employees typically lack a holistic overview of all the valuable information outside their specialist silos. The ability of generative AI systems to rapidly translate complex data into human-like text has the power to democratise knowledge and skills, drastically improving everything from learning and development to plant design and operational efficiency and safety.
Nuclear power is expected be a key pillar of the energy transition and the industry is growing in response. Globally some 170 new plants are planned or under construction while old plants are being restarted or kept operating for longer. Yet, the industry’s growth could be jeopardised by
a skills shortage exacerbated by an aging workforce and increasing early retirements. This, combined with a safety-conscious, conservative culture is stifling innovation, with the nuclear industry taking over a decade just to move from paper to digital records. The industry has also been slow to adopt new innovations such as AI, with nuclear workers the least likely of any energy sector to use AI in their current job role.
This is leaving substantial technology resources significantly untapped. New advances in Large Language Models (LLMs) for example, which are trained on vast amounts of text to generate their own text, have the potential to help overcome many of these challenges and in turn, help speed the energy transition.
The new paradigm in human-computer interactions
Human-machine interactions have undergone many paradigm shifts over time, from the days of paper punch cards and command-line input by keyboards, buttons and switches, to visual graphical user interfaces, and then voice user interfaces such as Amazon Alexa. LLMs represent a similarly game-changing evolution, enabling machines to turn vast amounts of existing data into refined human-like text in any format or language, and transforming human-machine interactions from one-way commands into two-way conversations.
While generative AI systems such as ChatGPT have traditionally been associated with errors and biases due to the fact they were trained on un-curated, outdated information from the internet, industry-specific LLMs add a new dimension. They have the ability to be trained on industry-specific data such as nuclear records and regulations to auto-generate everything from multilingual training and maintenance audits to document management reviews, and regulatory interpretation.
For example, an LLM could be asked to check records of all plant maintenance between April-June 2023 and instantly produce digestible, on-demand summaries which also reference all the relevant standards or regulations. A nuclear worker could then qualify this by asking the AI to identify defects or delays during the process and it will re-analyse and summarise the records in that context. The AI could also be used to audit the record-keeping process itself by being asked to identify missing documents or other gaps in the audit trail that could pose compliance risks.
Their capacity to rapidly amalgamate and translate millions of records for human consumption means that LLMs can give a much more holistic view of an organisation’s historic performance than any human. This would put enormous information resources at the fingertips of workers and could yield vital new insights into everything from operational safety practices to design defects. For example, workers could ask an LLM to identify and explain the three most common design faults across several plants over the last two decades.
The technology also has major potential to boost learning and development and help address a chronic industry skills shortage. It could rapidly synthesise and summarise a vast amount of knowledge to upskill workers, translating records or regulations into accessible training materials such as how-to guides in multiple languages.
As well as making information accessible to all, AI could also democratise innovation itself by putting huge cognitive resources at the disposal of workers. By deriving new innovations from patterns in existing data, LLMs could even auto-generate suggested operational or design improvements.
Data is the key
The success of AI technologies though, depends on many factors such as the quality and availability of industry data, requiring the digital integration and curation of all documents. Enterprise information management systems already widely used in the nuclear industry can now automate processes like version control and tag extraction, and enforce rigorous document management standards such as sending reminders of overdue document deliverables.
For example, 15% of all active commercial nuclear power stations in the US now use an Idox Engineering Information Management solution. This enables all their documents and meta-data to be stored in a single centralised digital environment with a detailed end-to-end audit trail of all changes from design to decommissioning. Tag-centric searches enable all these critical records to be rapidly retrieved on demand. Large Language Models could therefore easily interrogate these records to answer any query from the most frequent operational challenges experienced in each plant over a certain period, to what caused certain projects to run over time or over budget. Crucially, the system includes automatic revision control and a complete version history of all changes to documents and data, ensuring the content is of the quality required for AI applications.
This is helping to transform industry data from raw records into a refined, reliable and accessible resource that could be unlocked by future LLMs. The platform even allows new nuclear projects to clone successful templates of similar past projects to accelerate development times. These templates could in future be used by LLMs to suggest the optimal template for each type of project or even propose new templates.
In a cautious industry where accurate information is particularly imperative, trustworthy and well curated data holds the key to alleviating concerns over AI and harnessing its full transformative potential. Those willing to embrace change will fast become frontrunners in a burgeoning industry, with AI having the capacity to unblock some of the biggest bottlenecks in nuclear today from the skills shortage to the pace of innovation.
