The Nuclear Decommissioning Authority (NDA) is working a quartet of 'proof of concept' AI projects to improve the management of risk in the nuclear industry.
Carl Dalby, risk specialist for the non-departmental public body, outlined the work to the UKAuthority AI and Data4Good conference yesterday.
Qualifying his remarks by explaining he regards AI as ‘augmented intelligence’ rather than ‘artificial intelligence’, he said it provides the potential for big advances in managing data that can vastly improve risk management.
The first project involves using AI to recognise patterns and exposure to risk over the next six to 12 months; and the second covers financial risk forecasting, using the technology to shape the probabilistic model that helps identify risk in managing budgets.
“For proof of concept one and two, particularly around the data analytics part, if we can create increased certainty, we can create efficiencies around mitigation investment, which amounts to a return on investment for our fund management,” Dalby said.
The third project is the development of a ‘risk assistant’, using AI to deliver a real time analysis of meetings to help better understand risk and optimise resources. It will also support the effort to ensure the right people with relevant expertise are available at the right time for meetings.
“There is also an enrichment of current data,” Dalby said. “Historical data comes in various levels of quality. I have in mind for proof of concept three that for the Teams sessions we’re in, the information on the call is transcribed, so in time we will be able to build this unstructured, tacit knowledge base from meetings that have been going on over many years.
“We don’t necessarily have the technology now to exploit that data right now, but I’m sure we’ll have it within five years to exploit large amounts of meetings information and how we can pull out the nuggets of insights and value in real time.”
He said the risk assistant – which the NDA team hopes to have ready for a ‘show and tell’ by the end of the year – will also have a ‘jargon buster’ function and a ‘virtual shelf’ for further reading on topics raised in meetings.
Data and software
The prime data source for the three projects is all the risk log information from the past 10 years from all members of the NDA group, and the software is Microsoft based – described by Dalby as “AI algorithms out of the box” – using existing licences.
The fourth project, named Risk Live, involves using AI to connect group strategic risks (GSRs) information with global events in real time to deliver a ‘living risk’ picture. This can feed into alerts and urgent decision making.
Dalby said the risk log matrices that are currently used provide a snapshot but not a living view of how the world is changing.
“The ambition with Risk Live is that we will be able to present a living picture of how our risks are being impacted by external global events,” he said.
“We are using vast amounts of unstructured data from sources such as the Economist Intelligence Unit, and using AI algorithms on how we mix and match and building patterns of data on impacts of our current analysis of risk.”
This will provide a new alert system monitoring factors such as the supply chain. “I’m very keen to show that live risk position live risk position for supply and demand into the nuclear industry.”
NDA is working with the National Decommissioning Centre and the University of Aberdeen on the project, using external data sources through secure connections and its Microsoft software, and Dalby said it hopes to have a full demonstrator ready by Christmas.
Among the expected outcomes from the collection of projects, he pointed to better risk pattern analysis, the automation of risk dependency mapping, the possible use of natural language processing applications, the potential to identify data discrepancies, connections to smart security systems and real time horizon scanning.