Revisionism occurs when AI systems generate false narratives, deepfakes, or biased reinterpretations of events by drawing from skewed training data or deliberate fine-tuning. 

This can involve rewriting historical records, downplaying atrocities, or fabricating evidence to fit agendas, often amplified by automation in content generation or search results. 

Why it matters

History shapes how societies understand themselves - who they are, where they came from, and what they owe each other. 

When AI systems distort or rewrite that history, the consequences go far beyond academic inaccuracy to affect collective identity, political legitimacy, justice for victims of past wrongs, and the ability of citizens to make informed decisions about the present.

Impacts & consequences

Technological revisionism can lead to widespread misinformation and disinformation based on minority histories or uncomfortable truths that are "smoothed over" by algorithms trained on majority perspectives.

Long-term, such revisionism distorts education and result in societal polarisation and conflict.

Causes & contributing factors

• Biased or uncurated training data. AI absorbs whatever is in its training corpus, including existing propaganda, outdated scholarship, and politically motivated rewrites.

• Western-centric data dominance. The internet, and therefore AI training data, over-represents western or other cultural and political perspectives

• Lack of provenance checking. AI systems typically cannot verify whether sources are authoritative or manipulated.

• Deliberate adversarial prompting. Users intentionally steering models toward revisionist outputs.

• Incentives for engagement. Platforms optimising for clicks can amplify dramatic or provocative historical narratives over accurate ones.

• State-sponsored disinformation. The manipulation of historical events has been largely employed as a tool for foreign interference to achieve strategic objectives.