23-12 Using Discriminative AI to Identify Fossil Problematica from the Coon Creek Formation (Campanian, West Tennessee)

Session: Paleontology of North America (Posters)

Presenting Author:

Authors:

Abstract:

The Campanian Coon Creek Formation is a lagerstätte deposit containing over 700 species of invertebrates, vertebrates, plants, microfossils, trace fossils, and numerous unidentified “problematica” fossils.  We generated a series of prompts across five AI platforms (ChatGPT, Claude, Grok, NoteGPT, and Perplexity) in an attempt to identify two distinctive problematica: (1) a partial skull of an unidentified vertebrate and (2) a single cobble suspected to be a gastrolith. 

The AI prompt for the partial skull focused on the interlocking reversed triangular plates and associated morphology.  Stratigraphic, geographic and age constraints (Coon Creek Formation, West Tennessee, Campanian) were given.  Then based on two images of the partial skull, the AI models (AIM) were asked for 5-10 identification options based on morphology, size, and age constraints.  The AI prompt for the suspected gastrolith was more detailed.  The initial prompt was over 200 words and focused on rock type determination from six uploaded images of the hand specimen, thin section slab, and petrographic thin sections (cross-polarized light).  The AIM were asked to use mineral identification and rock texture to determine rock type and specific lithology based on mineral composition, texture, color, scale and dimensions of the hand specimen and thin sections.  Each AIM was asked to suggest 5-10 source areas and provenance locations for the suspected gastrolith.  Finally, the AIM were given a weathering and exhumation age constraint during the Campanian and asked to suggest the three most likely sources areas for the cobble. 

Discriminative AI focuses on identification, classification, and categorization.  When addressing paleontological and geological prompts, the AI models provided quality ranked options of possible taxa based on the partial skull.  The top AI rankings for the partial skull suggested pychnodontiform fish, marine turtle, and large predatory teleost fish.  The AI models did yield some misleading interpretations of glaucophane and foliation present in the gastrolith.  Follow-up constraining prompts focused on an arkosic lithology and suggested source areas associated with the Mogollon Highlands, Sevier Thrust Belt or the Colorado Front Range.  In general, the AIM excelled at providing summaries of Campanian and older source rocks for the gastrolith and ranking them based on proximity to coastal west Tennessee during the Campanian.  The models also provided obscure referenced sources for both the partial skull and gastrolith. 

Geological Society of America Abstracts with Programs. Vol. 58, No. 1, 2026

© Copyright 2026 The Geological Society of America (GSA), all rights reserved.