What the Machine Saw: some questions on the ethics of computer vision and machine learning to investigate human remains trafficking
Bibliographic record
Abstract
This article represents the next step in our ongoing effort to understand the online human remains trade, how, why and where it exists on social media. It expands upon initial research to explore the 'rhetoric' and structure behind the use and manipulation of images and text by this collecting community, topics explored using Google Inception v.3, TensorFlow, etc. (Huffer and Graham 2017; 2018). This current research goes beyond that work to address the ethical and moral dilemmas that can confound the use of new technology to classify and sort thousands of images. The categories used to 'train' the machine are self-determined by the researchers, but to what extent can current image classifying methods be broken to create false positives or false negatives when attempting to classify images taken from social media sales records as either old authentic items or recent forgeries made using remains sourced from unknown locations? What potential do they have to be exploited by dealers or forgers as a way to 'authenticate the market'? Analysing the data obtained when 'scraping' image or text relevant to cultural property trafficking of any kind involves the use of machine learning and neural network analysis, the ethics of which are themselves complicated. Here, we discuss these issues around two case studies; the ongoing repatriation case of Abraham Ulrikab, and an example of what it looks like when the classifier is deliberately broken.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.063 | 0.100 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.005 | 0.003 |
| Science and technology studies | 0.007 | 0.091 |
| Scholarly communication | 0.012 | 0.040 |
| Open science | 0.004 | 0.006 |
| Research integrity | 0.014 | 0.019 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".