Article by Simon Ilyushchenko: “The Italian aphorism traduttore, traditore – the translator is a traitor – encapsulates a deep-seated suspicion about the act of translation: that to carry meaning from one language to another is always, to some degree, a corruption.
The writer and semiotician Umberto Eco took this charge seriously. In Experiences in Translation, Eco treats translation as an interpretive act – negotiation, compromise, loss. Every translation is an imperfect reproduction of the original. Every translator, in choosing what to preserve, chooses what to betray.
This is the situation confronting anyone who works with geospatial data – human or AI.
In 2019, Colombian researchers studied the relationship between armed conflict and forest cover in their country. Using the Global Forest Change dataset – a widely respected product derived from satellite imagery – they found something striking: if analysis is not done carefully, armed conflict appeared to be correlated with increases in forest cover.
One might infer, perversely, that violence was somehow good for forests. The authors’ interpretation of the ground data was the opposite.
Here is the mechanism they propose: armed conflict destabilized the rule of law, which enabled the rapid clearing of native forests for oil palm plantations. These plantations are monocultures – ecological deserts compared to the biodiverse forests they replaced. But to a satellite sensor, a mature oil palm plantation can read as ‘forest’. It has trees. The canopy closes. The pixels are green.
And even this example gets messy fast. The relationship between Colombian conflict and forest cover has generated substantial literature – but no consensus. Ganzenmüller et al. (2022) identified seven distinct categories of deforestation dynamics across Colombian municipalities; the same peace agreement drove opposite outcomes in different regions. Bodini et al. (2024), using loop analysis to model the socio-ecological system, found that causal pathways connecting violence, coca, cattle, and deforestation were so intertwined that their models for left-wing guerrilla dynamics showed “very low agreement with observed correlations.” The data didn’t fit a simple narrative – any simple narrative…(More)”.
