Back to TechJournal

Research journeys · Record 001

How my research into AI and environmental journalism began

A photograph shows a possible environmental problem. How does it become a reliable story? My first published research paper explored how AI and human review could work together along the way.

Torsten Olivi Tiltack · PhD Candidate at UTS

Talk
FTC 2025 · 7:24 min · English
Paper published
16 October 2025
Context
Early AIJIM research

The original talk · FTC 2025

AI in Environmental Journalism

The YouTube player loads only when you click. This sends data such as your IP address to Google/YouTube. Privacy information

The question behind it

A picture is a starting point. The reporting is still to come.

A photo shows dumped waste. But where is it? What can the image actually tell us? And who checks the interpretation before it becomes a report? You don’t need an AI background to follow those questions.

The paper connects automated image analysis, human participation and report generation. NamicGreen provided the application context for the described 2024 Mallorca pilot. The interaction between these steps is central to the approach.

Paper & sources

From photograph to report

The earlier approach in six steps, simplified from the freely available preprint. Human review is an explicit part of the described process.

  1. 01

    Observe

    A photograph with location information is submitted.

  2. 02

    Analyse

    AI looks for clues in the image.

  3. 03

    Draft

    An initial report is generated.

  4. 04

    Check

    People check the detections.

  5. 05

    Review

    Experts and community members review the report.

  6. 06

    Publish

    The reviewed report is shared.

Workflow in the preprint · Section 3.4

What the paper reports

A pilot in Mallorca.

The published abstract describes a pilot involving 1,000 images and 50 previously undocumented waste sites. The figures below reproduce the outcomes reported in the paper. They have not been newly measured or independently recalculated for this record.

Explore the pilot’s reported figures
reported detection accuracy
85.4%
reported agreement with expert annotations
89.7%
reported reduction in reporting latency
40%

These figures describe different measures within the original study setting. They are not a general AI reliability rate or a performance promise for current systems.

Source: published abstract at Springer

Where to go from here

One step on my research journey.

The early work grew out of environmental journalism. My current PhD at UTS explores the traceability of AI-assisted claims. You can find the current research and connected projects here, each with its own status.

Read the originals

The sources behind this record.

The published chapter, earlier preprint and talk are different versions and formats of the same work. The preprint is freely available; access to the Springer full text may require a subscription or purchase.

  1. 01

    Published conference chapter

    AIJIM: A Scalable Model for Real-Time AI in Environmental Journalism

    Torsten Tiltack · FTC 2025, Volume 2 · LNNS 1676 · pp. 398–416

    Online since 16 October 2025 · DOI 10.1007/978-3-032-07989-3_26

    Publication at Springer
  2. 02

    Preprint · freely available

    AIJIM: A Scalable Model for Real-Time AI in Environmental Journalism

    arXiv:2503.17401v5 · Version dated 28 April 2025

    An earlier manuscript version. It should not be treated as identical to the later published chapter or as a second independent study.

    Read the preprint on arXiv
  3. 03

    Conference talk · English

    AIJIM: AI in Environmental Journalism | FTC 2025 Research Talk

    Torsten Tiltack · FTC 2025 · existing recording

    The talk reflects the research at the time. This companion page places it in context today.

    Original talk on YouTube

Take another look

Research moves on. So does this record.

Later additions, corrections and my personal reflection will have a visible place here. These questions remain open for a future revisit:

  • What would I investigate differently today?
  • Which assumptions have changed?
  • What new observation would change my interpretation?
Personal reflection to follow

Talk, published chapter and preprint brought together. Historical findings distinguished from the current project context. No new field study or subsequent revisit results yet.

Stay curious.

Find more films about research, technology and the people behind them on TechJournal.

Visit the YouTube channel