AI tools are entering geotechnical offices through classification, surrogate modelling, and automated report drafting. The risk is treating outputs as ground truth without the usual model governance.
Productive use cases we deploy with clients:
• Clustering discontinuity scan data before manual stereonet review
• Surrogate models for parameter sweeps ahead of full FLAC3D runs
• Anomaly detection on monitoring time series to flag review windows
• Drafting QA checklists from prior project templates
What AI should not do: pick constitutive models, set support spacing, or sign stability assessments without a licensed engineer in the loop.
Treat ML as an accelerator on well-documented workflows — the same reproducibility standards you expect from numerical models still apply.