AI in Geoscience: An International Union of Geological Sciences (IUGS) sponsored meeting on Geoscience Large Language Models (LLM) took place at the Geological Society of London on July 16th attended by 59 stakeholders world-wide.
I was asked to attend representing the IUGS Geoethics Commission. Of particular focus was the IUGS endorsed, Deep-time Digital Earth (DDE) LLM driven chatbot GeoGPT (and its API), the current version is yet to be generally released.
To ensure IUGS and the wider international geoscience community leads the way on open science and ethics in AI, meeting participants emphasised the desire for transparency and openness of data. Specifically, the need to publicly release, at an article (item) level, all of the geoscience data used to train DDE’s GeoGPT which is around 50Billion tokens.
“There was a strong recommendation that the geoscience corpus used to further train GeoGPT from its base model is transparent and be made available to the community.”
This includes making public, “The geoscience-specific training data to the article level. This could be crowd-sourced ‘checked’ by society publishers and other parties to assess the ethics and biases of the content used as well as the legalities, to build broad trust around a number of areas”.
A readout of the meeting under Chatham House rules has been made public. Link in the comments.
A readout of the meeting under Chatham House rules has been made public, see below. This will also be added to the IUGS website shortly. https://www.iugs.org/
A Geodyssey - Geoscience Text Analytics and Enterprise Search Research 
Leave a comment