One of the reasons why the company focused its initial efforts on Canada is that the country has large amounts of public research data, including narrative field reports, time-depleted geological maps, geochemical data on drilled hole samples, magnetic and electromagnetic survey data in air, lidar readings, and satellite imagery spanning decades of research.
“We have a system where we can enter all that data and store it in standard formats, control the quality of all the data, make it searchable and be able to programmatically access it,” Goldman says.
After gathering all available information for the site, KoBold’s team researches the data using machine learning. The company could, for example, build a model to predict which parts of ore deposits have the highest cobalt concentrations, or create a new geological map of the region showing all the different rock types and fault structures. He can add new data to these models as they are collected, allowing KoBold to adaptively change his research strategy “almost in real time,” Goldman says.
KoBold has already used insights from the machine learning model to acquire its Canadian requirements for mining and the development of its field programs. His partnership with Stanford’s Earth Resources Forecasting Center, in progress since February, adds an extra layer of analytics to the mix in the form of an AI “decision-making agent” that can replicate the entire research plan.
Stanford geographer Jeff Caers, who oversees the collaboration, explains that this digital bearer quantifies the uncertainty in the results of the KoBold model and then devises a data collection plan to reduce that uncertainty successively. Like a chess player trying to win a game in as few moves as possible, AI will seek to help KoBold make a decision about a potential client with minimal futile effort – whether it’s drilling at a particular location or leaving.