2025: the year of AI

The Christmas and New Year holidays generated a slew of articles about what is in store for us in 2025. Artificial intelligence (AI) got the most mentions, even though we have already been using it for years, either consciously with tools such as ChatGPT or unconsciously as a result of big tech influencing our social media feeds. It has been used within non-destructive testing (NDT) for a number of years, so it is not new. What is new and demands a response is the accelerating improvement in performance of the technology. Some of the figures I read in a newspaper article before Christmas are mind-boggling: the cost of calculations is falling by 99% every four years and since 2018 the efficacy of AI has grown 3000 times and it is getting faster.

An increasing number of papers detail the use of AI in NDT: as well as those presented at the BINDT conference in September 2024, there was a plenary paper on ‘Next-generation NDT: how artificial intelligence will define the future of inspection and structural health monitoring’, by R Gr Maev (www.bindt.org/events-and-awards/ndt-2024/plenary-paper-next-generation-ndt-how-artificial-intelligence-will-define-the-future-of-inspection-and-structural-health-monitoring). The December 2024 issue of Insight had a special feature on machine learning with five different applications and on 9 January 2025 the BINDT website reported on the use of a new machine learning algorithm for the reconstruction and analysis of computed tomography (CT) inspection data (www.bindt.org/News/january-2025/new-ornl-algorithm-enables-faster-safer-inspection-of-nuclear-materials-1).

The benefits of using AI in data analysis are easy to see. AI algorithms reduce the subjectivity of an operator and mitigate the risk of human error, providing enhanced accuracy and reliability. Their ability to analyse large amounts of data quickly can significantly reduce inspection times. They can also perform analysis of data from many different sensors and different NDT methods to provide an overall assessment of a component, something that would be prohibitive for a human to do.

My take on this is that the use of AI in NDT is only going to increase in 2025 and the NDT profession as a whole needs to be preparing for it now. Reading about the use of AI in a different sphere, I came across the terms ‘humans in the loop’ and ‘humans on the loop’. There are still some challenges around the use of AI in NDT. Models can operate as a black box, producing results output from data input without clear understanding of how the decisions were made. This can raise concerns in safety-critical situations. One way to address this is for AI to augment the data analysis, with the operator providing the final sentencing, keeping the human in the loop. In order to realise the full benefits of speed and efficiency, there will be pressure to relocate the human from ‘in the loop’ to ‘on the loop’, where they merely monitor operations and check that certain conditions exist.

So, going forward, ‘humans in the loop’ will be those who write the code and algorithms for the applications. These people seem to be currently based in academia and research organisations. ‘Humans on the loop’ will require data science skills to complement their practical NDT knowledge and maintain their analysis skills. We need to start the upskilling required as soon as possible so that we have a cohort of employees who can effectively use AI tools and understand the results. Given the rates of change predicted, we do not have time to prevaricate.

In the meantime, explainable AI techniques are being developed so that the humans on the loop can understand the AI model’s decision-making process. The NDT profession should be considering using explainable AI techniques to ensure confidence in asset management decisions based on AI inspection results.

Please note that the views expressed in this column are the author’s own personal ramblings for the purpose of encouraging discussion within NDT News. They do not represent the views of Amentum or BINDT.

Letters can be mailed to The Editor, NDT News, Midsummer House, Riverside Way, Bedford Road, Northampton NN1 5NX, UK. Email: ndtnews@bindt.org or email Bernard McGrath direct at bernard.mcgrath1@global.amentum.com