#GESmart: What are the opportunities of generative AI for ground engineering

The GE Smart Geotechnics conference is taking place TOMORROW!

Don't miss out on?this?essential gathering for anyone involved in geotechnical monitoring, data management and visualisation.

The conference will provide?up-to-date insights into projects, technologies and instrumentation, data systems and modelling software, as well as the use of the cloud, machine learning and AI.

Join us on tomorrow – secure your place today!

***Ground Engineering subscribers are entitled to a ￡100 discount***

At the conference, Arup associate Ben Gilson will explore the opportunities of generative AI, and how this emerging technology can be used to positively impact the ground engineering industry.

Generative AI is a branch of artificial intelligence (AI) that represents a shift from analysis to creativity. It involves the creation of new content—be it text, code, images, or summaries—by leveraging advanced Large Language Models (LLMs).

Unlike traditional AI, which follows predefined rules, generative AI interprets prompts and crafts original outputs. This offers significant improvements in how we interact with and apply it within ground engineering.

Generative AI is purported to disrupt traditional ways of working. As an industry our response varies from extreme hype to dismissive cynicism.

As a geotechnical engineer working across transport, energy, water, commercial and residential markets, these are the key opportunities I see gaining momentum in ground engineering.

Productivity increase. Whether searching for information, bug fixing python code, summarising documents and meeting discussions, creating documents; LLMs offer to rapidly accelerate typically the first 50-80% of a process.

Inclusivity by levelling the playing field. This includes new starters formulating technical report language, and helping people write in second or third languages. Studies such as those by consulting firm BCG's on How People Can Create—and Destroy—Value with Generative AI support this, indicating greater time savings for less naturally productive users.

Breaking down barriers to tools and analysis. Lowering the barrier to text-based programming, performing analysis in software programs based on written prompts such as geospatial analysis.

Information retrieval. This is key for our industry as there is a significant risk of knowledge draining away as the great minds of our industry retire. Over 750 papers were published at the XVIII ECSMGE 2024 Conference alone. How can we ensure the knowledge, and data, built up over many decades of successes and, more importantly, failures (when the observational method was a necessity not a choice!) are accessible to future generations?

Exercising “reasonable skill and care” is key to realising these opportunities within the ground engineering industry. This wording in our consultancy agreements provides a framework for evaluating potential use cases and software. Can I stand up in front of a judge and defend decisions informed by LLMs?

Growing sophistication in the application of AI in the industry raises the question whether the level of understanding and quality assurance (QA) is keeping up – the potential for the latest GPT to write a desk study and undertake geotechnical analysis including parameter derivation are real examples.

We work in a world governed by the complexity of the ground, with known and unknown unknowns, which often do not lend themselves to the pattern recognition that underpins AI. The term “Assisted Intelligence”, rather than “Artificial Intelligence”, gaining momentum is helpful, reminding us of the need to retain an appropriate level of control. Until a client accepts an AI agent’s signature on both a contract and a deliverable, we retain the role of QA.

To date I have heard warnings against the use of AI. But I foresee a growing warning against not using AI in our duty to exercise reasonable skill and care. Take the example of my team’s ongoing search for obstructions that may impact an urban tunnel portal and alignment. Manual searches for records within archives and targeted use of browser search engines is routine. But what if a missed obstruction record resulted in claims from a tunnelling contractor, and it was identified that a simple prompt in a LLM such as Microsoft Copilot would have identified the obstruction in an obscure reference? Is it reasonable that I should have used AI to search and summarise information?

So, what next?

Engineering a resilient and low-carbon built environment and effectively stewarding our natural environment requires more effective access to information, past and present. Efficient access to the latest guidance, past case studies and records, standards and papers is vital. Generative AI can connect decision makers with information, improving how we analyse data, and freeing up time to come up with better, more sustainable solutions.

At Arup we are testing our customised AI chatbot for our global Ground Engineering Skills Network to accelerate knowledge sharing and information retrieval. Simple prompts provider our engineers and geologists with access to internal guidance and publicly accessible guidance, as we work with information custodians to navigate licence agreements to integrate commonly used documents.

Custodians of information have a role to play. Siloed information restricts access. Pay walls are necessary to preserve quality and encourage publication. However, we need new, more flexible commercial licensing models that provide access to LLM to facilitate querying industry guidance and published papers together with in-house knowledge. LLMs operating knowledge retrieval algorithms will be increasingly responsible for finding content. Industry bodies will need to decide what is more important, allowing LLMs to query free-to-access guidance or maintain control of downloads.

As I fondly recall time consuming and limited searches on my Encarta encyclopedia CD-ROM disks, I hope I will look back on the recent past when accessing key knowledge is often time consuming and not guaranteed. At the current rate of content generation, we need to take advantage of the opportunity that LLMs offer to retrieve information, whether from obscure papers, or Ciria guides of the second generation of Eurocodes.

Lack of time was raised as a key blocker to providing more sustainable designs during a sustainability debate at the recent GE Piling and Foundations conference. Technology such as generative AI, when used appropriately, demonstrates the potential to free up time to focus on understanding and applying information, rather than searching for it.

Taking up this opportunity requires buy in across the information cycle, from content creators, information custodians, designers and contractors alike, and technology providers.

What are you doing to take up this opportunity in your organisation?

To discover more about the 2024 programme click HERE .