TechJournal
Published work, original sources and questions I am still working on.
My research
In my PhD at UTS, I study how scientific claims change in human and AI-mediated communication: what evidence still supports them, and what do earlier assessments and approvals still cover? AIJIM is the separate research and publication protocol. SHARK is a scientific workspace and an independent implementation of this protocol. ViaoraAI S.L. is a separate company context. NamicGreen remains an independent environmental project outside my PhD.
My work spans an AI model for environmental journalism, an Evidence-First architecture for organisational decision support, and a machine-checkable specification of publication conditions. Two published conference chapters and a preprint that has not been peer reviewed document this research trajectory.
The papers examine different objects: findings from the early environmental study do not establish the performance of the later reference model or protocol.
AIJIM Protocol websiteResearch approach
Making sources, claims and accountability traceable
AIJIM explores how sources, changes and human decisions can remain connected to a published claim. Public papers and versioned research artifacts document its development.
Explore papers and sourcesIndependent implementation
Scientific workspace
SHARK independently implements the AIJIM Protocol. It focuses on work on the visible manuscript, traceable sources, changes and human decisions. The scope and verification status of individual workflows are reported separately.
Explore SHARKEnvironment & fieldwork
An independent environmental project
What remains of an environmental photograph when someone else takes over the case? NamicGreen aims to keep observations, evidence, decisions and the next inspection together. Originating in Mallorca, it is being developed for Mallorca and New South Wales, Australia. An old finding does not establish today's conditions.
Explore NamicGreenSources & publications
The original sources for the conference chapters, preprint and versioned research artifacts.
Conference chapter
Authors: Torsten Tiltack
Conference chapter
Authors: Torsten Olivi Tiltack, Yifei Dong, Kun Yu, Fang Chen
AIJIM specification paper
Authors: Torsten Olivi Tiltack, Yifei Dong, Kun Yu, Xu Wang, Wei Liu, Jianlong Zhou, Ren Ping Liu, Fang Chen
Software & research artifacts
Software & research artifacts
Looking ahead · in development
A systematic review in development: how do scientific claims change as they travel through human and AI-mediated communication? No results are reported here yet.
What can dated sources establish about how scientific work handles new evidence? I am also interested in how publication delays affect our view of that process. Ongoing manuscript work, not a published study.
Illustrative example
State C · Limited claim
The plants nearer the radio are taller. The cause remains open.
Published version · State C
Evidence: observation of heights
Human decision for this version · State C
Publication prerequisites checked – for this version.
Predecessor: the original observation: The plants nearer the radio are taller.
What do I see? Four plants. Two are taller. Both are closer to the radio. That's the observation. In this example, this version is reviewed and approved.
Suppose an AI rewrites it: Music made these plants taller. Wait. That's a different claim. The old approval stays with the old version.
And the lamp? The light could matter too. This image alone doesn't tell me the cause. The observation doesn't support the stronger claim.
So: the plants nearer the radio are taller. The cause remains open. The observation is still usable. This new version is assessed again.
Before publication, evidence, checks and human approval must match this version. All publication conditions must be met. More cautious wording alone isn't enough.
This exact version is published. The observation stays. The old approval doesn't follow. AIJIM connects evidence, checks and approval to what is actually published.
The plants nearer the radio are taller.
The original version, evidence and human decision belong together.
Evidence: observation of heights. Human approval for this exact A.
Music made these plants taller.
A describes a height difference. B additionally claims that music caused it. The observation stays the same; no causal evidence is added. The human decision stays bound exclusively to A; B has no approval.
No approval for state B.
Music made these plants taller.
All four plants, the radio and the light remain visible and unchanged. Light or other factors could also explain the difference. No cause is established. The same evidence can be considered again, but does not automatically support the stronger claim. Separately, the decision for A must not appear as a decision for B.
The plants nearer the radio are taller. The cause remains open.
A person narrows the claim and makes a separate decision for this version.
The plants nearer the radio are taller. The cause remains open.
The wording stays identical. All prerequisites bind this exact version; no abbreviated checklist replaces the complete rule.
The plants nearer the radio are taller. The cause remains open.
Editorial marks disappear without changing the wording. Evidence, checks, the decision and the predecessor remain as marginal references.
The old approval does not follow the new claim.
AIJIM makes this relationship traceable: which claim was checked against which evidence and authorized by a person?
The timeline is an explanatory simplification: human authorization and all other publication prerequisites bind the same exact state. The checking phase does not change the version and is not a separate, subsequent approval step.
A, B and C represent complete publication candidates, condensed here into one sentence. AIJIM does not require a separate permit for every sentence. Keep two distinctions: observing a difference does not establish its cause; and a decision for A does not automatically authorize the materially changed B. The protocol does not determine whether music, light or something else caused it. Human approval alone replaces neither evidence nor the other publication prerequisites.
The steps are explanatory moments. Stepping back rewinds the example; it does not revoke a publication.
Fictional illustration of the research question. Not a real review, a running AIJIM system or proof that the claim is true.
