“Hi ground, how are you feeling?” How AI is transforming geotechnics
NGI’s Andreas-Nizar Granitzer received the Bright Spark Lecture award at an international conference in Hong Kong in May 2026, a distinction reserved for particularly promising early-career researchers. The recognition follows research showing that AI is already delivering measurable results in geotechnical practice, and that the industry is ready for the step change.
Andreas-Nizar Granitzer presenting at the 5ISMLG conference in Hong Kong in May 2026, where he received the Bright Spark Lecture award for his work on data-centric geotechnics. ( Photo: NGI)
Ground conditions determine whether a structure remains functional or not. That is one of the practical foundations of geotechnics, the discipline concerned with understanding soil and rock, and with building safely in and on them. The field is now changing fast.
Granitzer is a research and project engineer in NGI’s onshore foundations section. At the 5ISMLG conference (Machine Learning & Big Data in Geoscience) in Hong Kong in May 2026, he presented an empirical review of what is actually working in data-centric geotechnics and what is still holding back wider adoption across the industry.
“The technology is evolving rapidly, and much of it performs well in research settings. The challenge is making it work in the day-to-day practice of engineers and generating tangible client value,” says Granitzer.
Broad interest across generations
One of the more striking findings comes from the GEDIAG project, an international survey with nearly 1,000 participants spread across geo-disciplines. The study, executed with Austrian colleagues from the field of geotechnics and social sciences and published in the journal Geomechanics and Tunneling, examined attitudes toward AI and new technology among geotechnical engineers across age groups and countries.
The results challenged a common assumption: older professionals are not necessarily sceptical, while younger ones are not necessarily enthusiastic. Interest in AI and data-centric methods is broad and cut across generations.
“The survey shows that AI is largely seen as a means to reduce repetitive and labor-intensive tasks, not as a threat to job security,” says Granitzer.
That is a significant signal for a field where manual data processing and the interpretation of large volumes of monitoring data could significantly benefit from reliable, AI-driven solutions. The survey also shows that the industry largely still perceives AI as a passive tool rather than an active collaborator in dynamic environments, reflecting a lower maturity level than other disciplines moving toward autonomous systems with human oversight.
From core samples to sensor data
In his research, Granitzer points to several projects where data-centric methods have produced measurable results.
One example is the use of computer vision — the ability of computers to interpret images — to classify soil core samples from ground investigation campaigns and augment geotechnical experts. Human assessments can vary depending on who examines the sample. Image analysis, by contrast, produces consistent and reproducible results.
Another comes from the EU project The HuT – lead by NGI’s Luca Piciullo and supported by Andreas Mathisen – where NGI researchers have developed an agentic system that allows engineers to query large near real-time monitoring databases from real-world sites using natural language. Rather than navigating complex tables, an engineer can simply ask site-specific sensing systems a direct question and receive a traceable answer, including visual descriptions. For example: “Hi excavation pit, how are you doing today?”
“The goal is to make information accessible to the people who actually need to make decisions in the field – including our clients,” Granitzer explains.
The deep excavation at Campus Ullevål in Oslo, where NGI researchers used surrogate models to assess retaining wall stability in quick clay. ( Photo: NGI)
Quick clay and surrogate models at Campus Ullevål
A third example comes from NGI’s own Campus Ullevål in Oslo, where deep excavations were carried out in quick clay, a sensitive marine clay that can suddenly lose strength when disturbed.
To assess the stability of the retaining walls, the team used surrogate models: probabilistic machine learning models trained to deliver fast and reliable answers in situations that would otherwise require time-consuming calculations. The approach enables running hundreds of analyses that traditional methods would allow only a few.
“The approach made it possible to run a large number of analyses in a short time and to handle parameter uncertainty in the input data in a transparent way,” says Granitzer.
Three barriers holding the field back
Despite promising results, the uptake of data-centric methods remains limited across the industry. Granitzer identifies three key barriers: sparse data, insufficient integration into existing workflows, and a lack of interdisciplinary knowledge.
“Geotechnical data is often sparse in nature, sampled by different organizations, and to varying standards. That makes it difficult to feed directly into data-centric solutions,” Granitzer explains.
And even when data quality and volume is adequate, there is no guarantee the systems will align with how engineers actually work. Granitzer’s central argument is clear: technology alone is not the answer.
“What matters is starting with the practical problem and letting it guide the choice of technological solution, not the other way around,” he says.
The field is advancing quickly, and Granitzer’s contribution in Hong Kong is one of several signs that data-centric geotechnics is moving from a research topic to a practical engineering discipline. Which methods will actually take hold in everyday practice remains to be seen, but the groundwork is being laid now.
“Instead of reinventing the wheel in silos, we need to pool inter-disciplinary knowledge,” Granitzer concludes.
