Author: Richard Bendall-Jones, Product Manager and Risk Engineer

The field of artificial intelligence (AI) has seen rapid growth and development, and one of its most promising branches is deep learning. There has been a lot of press and social media coverage about ChatGPT and the benefits that this technology may bring.

Deep learning is a type of machine learning that uses algorithms that can be used to forecast a range of potential future outcomes. It’s tempting to say that deep learning merely has ‘great potential’ to offer significant benefits. In truth, it is already being implemented, in a practical sense, on projects in the infrastructure sector to inform decision making and compelling action to manage risk.

What is Deep Learning?
Deep learning is a type of machine learning that uses neural networks, or artificial neural networks, to process and analyse data. Mimicking a biological ‘brain’, these neural networks are designed to learn from and recognise patterns in data, allowing them to make predictions and suggest decisions based on that information. In an organisational or project context, this could mean taking historical data and using the deep learning models to forecast potential future outcomes, which would then drive a discussion about what to do about it, ultimately resulting in action being taken, and the outcome feeding back into the dataset.

How we interact with Deep Learning
You’re probably already using deep learning as a part of your life! We interact with deep learning technology in many ways, such as through voice-activated virtual assistants like Siri and Alexa. This means that deep learning is already with us, and bringing benefits to how we live our lives, even though in an infrastructure risk management context it is a relatively new kid on the block.

Applying Deep Learning to Risk Management
A fundamental element of any effective risk management strategy is to enhance and support an organisation’s decision-making process. Deep learning can play a significant role in this process by providing organisations with a more efficient and accurate way to understand the range of potential outcomes, based on empirical historical data, in comparison to the subjective approaches often found in qualitative and quantitative risk management methods. By providing forecasts and other insights, based on historical information (such as cost plans or schedules), project teams and organisations can make decisions that are freer of bias, with the aim of more quickly getting to the root of problems, or uncovering opportunities.

From a project risk perspective, the vast wealth of project data in organisations can lead to deep learning approaches. By learning from previous project performance in a wide variety of contexts, teams and organisations can use this information to find likely sources of prolongation and cost uplift, and seek to mitigate them earlier than would have been identified using traditional horizon scanning techniques. These approaches are already being adopted by a number of companies in the built environment, as a supplement or an enhancement of traditional risk identification and quantification approaches.

The Importance of Quality Data
Data is the foundation of deep learning algorithms. Therefore, it is essential to have high-quality data to be able to produce insights that inspire confidence, and, ultimately, value-added decision making. So that these insights can be effective, deep learning algorithms must be trained on large and varied datasets that accurately represent the environment that they are trying to model, whether that is an organisational or a project context. Consequently, organisations must ensure that the data they use to train these algorithms is accurate, complete, and up-to-date, and that it includes a broad and true representation of all relevant factors and scenarios.

The Challenges of Implementing a Deep Learning Approach
While deep learning has the potential to revolutionise project risk management, it is important to recognise that there are also significant challenges to implementing this new approach. The most significant of these challenges is the change of mindset required within the environment in which the technology is being deployed. Where traditional, human-centric approaches to project risk management have been applied previously, it can take effort to ‘let go of the reins’ of the risk identification and
quantification process, believe the outputs of a deep learning model, and focus solely on the outputs provided and the action they foster. Similarly, innovative approaches can be misconstrued as being a panacea to solve all problems – in fact, deep learning approaches work best with structured data sets looking to solve well-defined problems.

Another challenge for some project delivery organisations is the volume of data that deep learning algorithms require to be effective. Organisations must have the infrastructure in place to store and manage this data safely and legally, and they must also have the resources to process and analyse the data in real-time. As a result, approaches to deep learning, or other artificial intelligence-led approaches will benefit from having a parallel data strategy to ensure that they can get the best out of their
investment.

The Benefits of Deep Learning
Despite the challenges, the benefits of using deep learning in risk management are significant, for example improved accuracy and precision of forecasting, as well as identifying potential sources of risk. The depth and sophistication of deep learning models extends beyond the limits of human cognition, after all, AI does not have an attention span like we do – if you hear about AI technologies claiming to be
‘superhuman’ – this is why!

Deep Learning: Influencing the Future
As deep learning continues to evolve and mature, deep learning algorithms are likely to become more sophisticated and capable of handling even larger and more complex datasets. This will allow organisations to gain deeper insights into the risks they face, see further into the future, and therefore to help organisations and teams to make more informed decisions about how to effectively manage those risks. By adding these approaches to existing toolsets, deep learning can offer a valuable data-centric ‘second opinion’ to challenge stakeholders as to any biases (conscious or unconscious) they may be harbouring.

And what does it mean for project risk professionals in the infrastructure sector? We may also start to see the requirement for knowledge of deep learning approaches, and how to integrate them with existing processes (or indeed replace them), in future job roles or specifications. In a not too distant future, this may be similar to how the risk profession currently considers risk framework competence, or quantitative risk assessment expertise. Deep learning is a rapidly-growing field with enormous potential to revolutionise the way organisations approach risk management. While it contains a lot of potential, we are starting to see the first practical examples of deep learning approaches helping organisations and their project teams to tackle risk proactively, taking effort out of quantification workshops and into actively mitigating risk.