GLAMorous AI TL;DR — January 2026

Who Gets Seen When AI Looks at Nature?

Author

Alphaeus Lien-Talks
January 22, 2026

“AI does not discover nature on its own. It reflects the data we choose to collect, digitise, and prioritise.”

Welcome to a special edition Naturally AI, a short roundup of how artificial intelligence is being used across natural history, spanning biodiversity, botany, palaeontology, geology, and Earth sciences, as well as the collections and data infrastructures that support them.

This month’s theme is visibility:
what becomes easy to study, and what stays hidden, once AI is layered on top of uneven data.

Beery - People-Centric AI for Conserving Biodiversity (UNDP report)
A clear, practical overview of where AI genuinely helps conservation work (monitoring, remote sensing, prioritisation) and where it can mislead through bias, loss of context, or weak accountability.

Reynolds et al. — The potential for AI to revolutionize conservation: a horizon scan
A large, expert-led horizon scan identifying 21 AI applications most likely to shape biodiversity conservation in the near future, ranging from species recognition and acoustic monitoring to digital twins, foundation models, and AI conservation advisors.

🦴 Specimens, Collections, and AI Readiness

Hardisty et al. — Digital Extended Specimens: Enabling an Extensible Network of Biodiversity Data Records as Integrated Digital Objects on the Internet
Argues that AI-ready natural history depends on connecting specimen records to the wider web of related data (images, DNA, field notes, environmental context), so models can learn from linked evidence rather than isolated, error-prone snapshots.

East et al. — Optimizing Image Capture for Computer Vision-Powered Taxonomic Identification and Trait Recognition of Biodiversity Specimens
Demonstrates that AI accuracy depends heavily on how specimens are photographed.
Inconsistent angles, lighting, and cropping introduce systematic error that models cannot correct.

🌿 Botany and Herbarium AI

Weaver et al. — Herbarium specimen label transcription reimagined with large language models: Capabilities, productivity, and risks
Evaluates LLMs for label transcription, highlighting productivity gains alongside consistent failure modes that still require human review.

Pearson et al. — Machine Learning Using Digitized Herbarium Specimens to Advance Phenological Research
Extracts flowering and fruiting signals from digitised specimens, while showing how variation in layout and label quality constrains model reliability.

Turnbull et al. — Hespi: a pipeline for automatically detecting information from herbarium specimen sheets
Part of a broader shift towards “machines extract, humans verify”, which appears to be the most scalable and trustworthy pattern for botanical AI.

🐦Biodiversity Monitoring Beyond Specimens

Márquez-Rodríguez et al. — A bird song detector for improving bird identification through deep learning: A case study from Doñana
Shows strong performance for general detection, while revealing sensitivity to local soundscapes and recording conditions.

Bachimanchi et al. — Deep-learning-powered data analysis in plankton ecology
Explains how AI enables large-scale plankton classification from imagery, transforming marine monitoring while introducing new data-quality dependencies.

Chen et al. — Open-Insect: Benchmarking Open-Set Recognition of Novel Species in Biodiversity Monitoring
Tests whether AI can recognise species it has never seen before.
Performance drops sharply outside common, well-represented insect groups, exposing how narrow many training datasets still are.

Pollock et al. — Harnessing AI to fill global shortfalls in biodiversity knowledge
Explores how AI can scale biodiversity monitoring, while showing that uneven geographic coverage limits reliability in under-studied regions and ecosystems.

🌍 Earth Sciences and Geology

Tang et al. — Machine learning for point counting and segmentation of arenite in thin section
Demonstrates that training data diversity matters more than model complexity for geological classification tasks.

Amer et al. — A Photomicrographic Dataset of Rocks for the Accurate Classification of Minerals
A dataset paper addressing one of geology AI’s biggest bottlenecks: high-quality, reusable labelled data.

🦖 Palaeontology and Fossils

Knutsen & Konovalov — Accelerating segmentation of fossil CT scans through Deep Learning
Shows how deep learning can drastically reduce segmentation time, while emphasising validation and expert oversight.

Yu et al. — Artificial intelligence in palaeontology
Surveys AI use across classification, segmentation, and prediction, noting that adoption remains uneven and method-dependent.

⚖️ Governance and Responsibility

OECD — Biodiversity and Artificial Intelligence
A policy perspective stressing that governance, transparency, and accountability must evolve alongside technical adoption.

❓ Big Question

If AI increasingly shapes
what is identified,
what is prioritised,
and what is reused across natural history

who is responsible when those systems make nature look simpler, rarer, or more certain than it really is?

💬 About

I’m Alfie, a researcher and archaeologist exploring where heritage, ethics, and AI meet.
This digest keeps things short, critical, and useful — no jargon, no hype.

👉 Read or subscribe at glamorousai.beehiiv.com
👉 Send papers, reports, or ideas for Febuary