Industries / Mining and materials / Mining and quarrying
Exploration
Exploration is the search for rock that holds enough of a mineral, in one place, to be worth mining.
5
Milestones achieved, from 1835 to 1997.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1835
A national geological survey
Britain set up the Ordnance Geological Survey, now the British Geological Survey, after Henry De la Beche's geological survey of Devon showed its practical value. The British Geological Survey says it is the world's oldest national geological survey.
Credit
Henry Thomas De la Beche, backed by Thomas Colby of the Ordnance Survey
1945
Magnetic maps from the air
The fluxgate magnetometer, an instrument that measures the strength of the Earth's magnetic field, was built during the Second World War to find submarines from aircraft. After the war it was taken up for mineral exploration, so aircraft could map buried magnetic rock over large areas. By mid-1945 the US Geological Survey had flown a pioneering aeromagnetic survey at Oswegatchie, New York.
Credit
Victor Vacquier and colleagues at Gulf Research and Development Company (instrument); Herbert Hawkes of the USGS and Arthur Buddington of Princeton University (Oswegatchie survey)
1972
Satellite images for geology
NASA and the USGS launched ERTS-1, later renamed Landsat 1. In its first years the USGS used the images to locate oil and mineral deposits, monitor strip mining and find unknown faults and fractures.
Credit
NASA and the US Geological Survey
1989
Rules for reporting ore
After poor reporting during the Poseidon nickel boom and bust in late-1960s Western Australia, an industry committee formed in 1971 began writing rules. In February 1989 it released the JORC Code, which sets how companies report exploration results, mineral resources and ore reserves, each signed off by a named Competent Person.
LessonWhen investors cannot check rock in the ground for themselves, a named, qualified person has to stand behind each public number.
Credit
Joint Ore Reserves Committee, set up by the Australian Mining Industry Council and the Australasian Institute of Mining and Metallurgy
1997
A salted gold find
Bre-X Minerals said its Busang project in Indonesia was one of the largest gold deposits in the world. In 1997 evidence showed its samples had been salted with gold from other sources, and the Toronto Stock Exchange delisted the company in May 1997, wiping out $3 billion of investment, by the Canadian Institute of Mining's account. The exchange and the Ontario Securities Commission formed the Mining Standards Task Force, and Canada's rule NI 43-101, in force from 1 February 2001, requires mining disclosure to rest on the work of a "qualified person".
LessonA drill result is only as good as the checks on how its samples were handled.
Credit
TSE-OSC Mining Standards Task Force; Canadian Securities Administrators
Watch for
AI explorer's copper mine
KoBold Metals, a company that uses artificial intelligence to search for copper, cobalt, nickel and lithium, says it has started developing its Mingomba copper project in Zambia. It targets 300,000 t of copper a year from the early 2030s, at an investment of $2.3 billion to $2.5 billion.
Achieved whenthe mine produces copper, or an independent party publishes a resource estimate for it under a reporting code such as JORC or NI 43-101.
Claimed by
KoBold Metals, through its Africa chief executive Mfikeyi Makayi speaking to Reuters, March 2026
AI reads old reports
The British Geological Survey, working with IBM and the STFC Hartree Centre, built an AI system to pull critical-mineral information out of about 40,000 pages of UK exploration reports from the 1970s and 1980s, some of them handwritten. Its answers cite their source pages.
Achieved whenan independent group scores the extracted data against a hand-checked answer key, which the authors say does not yet exist.
Claimed by
British Geological Survey, IBM and STFC Hartree Centre, June 2026 (conference abstract)
Open
01
Blind tests for prospect maps
A public test for prospectivity maps, which rank ground by the chance it holds a deposit. Each method builds its map only from data available before a cutoff year, and is then scored on the deposits found after that year.
The briefCredit · unclaimed02
An answer key for reports
A hand-checked set of drill results (hole name, depth from and to, and grade) taken from a few hundred public exploration reports, and a score for how well AI models pull out the same values.
The briefCredit · unclaimed
Open 01 · the brief
Blind tests for prospect maps
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test for prospectivity maps, which rank ground by the chance it holds a deposit. Each method builds its map only from data available before a cutoff year, and is then scored on the deposits found after that year.
Why it is still open
Tests of these maps can rest on very few known deposits. One study of gold prospects in eastern Kazakhstan had only 22 known positive sites, and its authors write that heterogeneous data sets "limit reproducibility".
Built with what exists
Public geology and geophysics grids from national surveys such as the USGS and Geoscience Australia; deposit lists with discovery dates, such as the USGS Mineral Resources Data System; Python with scikit-learn and XGBoost; QGIS.
First step
Pick one region with good public data, list its deposits with their discovery years, and freeze a copy of the data as it stood before a chosen cutoff year.
Done looks like
A published leaderboard where at least three methods are scored on the same set of post-cutoff discoveries, reporting the share of later discoveries that fall inside the top 10% of ground each map ranks highest.
Open 02 · the brief
An answer key for reports
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A hand-checked set of drill results (hole name, depth from and to, and grade) taken from a few hundred public exploration reports, and a score for how well AI models pull out the same values.
Why it is still open
The British Geological Survey says that judging its own AI report reader "remains challenging due to the absence of comprehensive ground truth" for its archive.
Built with what exists
Open-file exploration reports from geological surveys, such as Western Australia's WAMEX system or the British Geological Survey archive; an OCR tool such as olmOCR, which the BGS team used; large language models such as Claude or Gemini (the note "Seven more flagged" covers Claude reviewing records); a spreadsheet.
First step
Download 30 open-file reports from one survey and key every drill intercept in them by hand.
Done looks like
300 reports and at least 3,000 drill intercepts keyed by two people, with their disagreements resolved and published under an open licence, plus field-by-field accuracy for at least two AI models.