Resourcing Tomorrow: From AI hype to real geological signal

Data structuring is no longer optional

Across juniors, majors, and government agencies, one message came through consistently. Everyone knows they have a data problem, but very few know how to solve it.

From exploration through to production and blasting, data is both the main barrier and the main enabler. Inconsistent formats, misaligned projections, and fragmented datasets are preventing teams from applying analytics and AI across the value chain in a meaningful way.

This aligns closely with what we see in practice. Modern exploration generates dense, multi-layer datasets from airborne magnetics, hyperspectral imagery, geochemistry, and geophysics. Without a standardized data foundation, the signal gets buried under redundancy and noise. Our work on dimensionality reduction has shown that datasets with 60+ layers can often be reduced by more than 80% while retaining nearly all meaningful variance.

The takeaway is simple: AI only becomes useful once the data is clean, aligned, and structured.

AI needs to earn its place on every project

There is clear fatigue with “AI window dressing.” Companies are no longer interested in running pilots just to say they tried AI. They want to see tangible value, project by project.

That means fewer black boxes and more workflows that geologists can interrogate, test, and improve. This is why geoscientist-in-the-loop approaches are resonating so strongly. AI models should support hypothesis testing, not replace it.

In our prospectivity work, this means framing analysis around explicit geological assumptions, generating mineral probability maps alongside uncertainty metrics, and refining models as new data comes in. This feedback loop is what allows AI systems to actually improve over time, rather than producing static outputs that quickly lose relevance.

Central Asia is moving fast

One of the most striking aspects of the event was the level of engagement from Central Asia. Governments and companies from across the region are actively looking for partners to help unlock large volumes of underexplored resources.

These regions are often data-rich but workflow-poor, which makes them ideal candidates for modern data integration and analytics platforms. For anyone looking at exploration, development, or investment opportunities, this is a region to watch closely.

‍

‍

Prospectivity works when it respects geology

There is strong demand for AI-driven prospectivity, especially when it goes beyond ranking maps and moves toward defining drillable targets with associated uncertainty. What matters most is trust.

That trust comes from workflows that respect geological reasoning, preserve metadata, and allow practitioners to understand why a model is producing a given result. When AI is used to reduce noise, integrate datasets, and surface patterns that geoscientists can validate in the field, it becomes a powerful accelerator rather than a distraction.

What respecting geology looks like in practice

Speaking of respecting geology, at MinersAI we’re developing an iterative geologist-in-the-loop mineral system analyst. This will enable the user to step through guided workflows for minerals and mineral-systems of interest, consistently applying robust methodologies to your data in order to find mineralization systems of interest. What’s really exciting in our approach is that you can use it throughout various stages of exploration, from greenfield to brownfield, and iteratively as more data comes in and you continue to test geological hypotheses.

As part of the mineral systems analysis modules, we’ve also built a ‘bare earth composite’ search algorithm for satellite data that can sift through masses of satellite data to find the best pixels in any given spot. This allows the geoscientist to skip past vegetation that may grow in the wet season, or snow that might blanket the ground in the winter, and analyze what is actually on the ground. For more background, see for example Demattê, J.A.M., Safanelli, J.L., Poppiel, R.R. et al. Bare Earth’s Surface Spectra as a Proxy for Soil Resource Monitoring. Sci Rep 10, 4461 (2020).

On the data processing front, we have implemented a robust human-in-the-loop workflow that leverages LLMs and allows us to process large quantities of diverse geoscience data into a consistent set of formats adhering to a data model. The human is responsible for making key decisions during the workflow, and our code captures everything that is done to the data, ensuring we understand and can track all data transformations. This is an essential module to transform the vast quantities of data we’re currently dealing with from different countries and companies.

‍

The shift is clear. The industry is no longer asking whether AI belongs in mining. It’s asking who can apply it rigorously, transparently, and in a way that delivers real geological insight. That’s the problem space we’re focused on at MinersAI, and it’s exactly the kind of conversation we’re keen to continue.