Assessing risk when introducing AI-driven changes to the railway
Project number:
T1363
Enabling the effective introduction of AI solutions into rail
'The AI Change Toolkit is a fundamental step forward concerning governance in a safety critical industry.'
The challenge
Artificial intelligence (AI) solutions have the potential to transform the rail industry for the better. However, it is essential to assess their impact on the operational railway.
The five principles published in the government white paper indicate what good AI looks like. However, understanding how to apply and test them in practice remains challenging. Changes made to the GB rail network must be assessed using the Common Safety Method (CSM) to provide safety assurance. Introducing AI solutions can be difficult to assess due to the novelty and the fast pace of this technology.
Deploying AI solutions with the potential to affect rail operations without the right tools increases the chance of issues. These could have negative effects on customers and staff, and unintended safety and performance consequences.
What we did
RSSB produced an initial AI Change Toolkit. This takes the user through guided steps to help generate and resolve concerns when introducing an AI solution into rail.
We reviewed similar resources from different sectors and across the globe to understand good practice. We also engaged extensively with industry to understand how early adopters are assessing and assuring AI solutions that have the potential to impact railway operations.
The toolkit has three elements. The first set of open questions helps users to define the change that they are aiming to make with the use of AI. Then, based on the CSM significance criteria, a second set of questions helps to assess the change. These can then be reviewed and addressed via pointers to specific guidance for the user to consider.
Benefits delivered
The proof of concept AI Change Toolkit was welcomed by the industry and received significant positive feedback. The ask is now to develop it further into a fully functioning tool.
Having a common method to consider and evaluate the implications of AI solutions will save time and resources for the organisations that use it. The concerns generated and actions taken by engaging with the toolkit should directly support the engineering change process.
We are currently refining the first two toolkit elements using the feedback collated from industry. We are also populating the guidance element of the toolkit. This will be targeted depending on the user’s responses. It will provide an easy mechanism for users to find help with the successful development and deployment of their AI solutions.