Cautious commentators say AI might amplify existing societal biases. Yes, that might be true. But what about history-washing, or any other kind of "-washing" you care to name?

I'm interested in business history and have read a few Wikipedia articles. So I thought I would ask a commercial LLM for a quick overview of the history of a US company called IBM.

The app promptly gave me a neat, decade-by-decade summary. To my surprise, it looked odd; the "history" was entirely devoid of Edwin Black's findings on the strategic alliance of IBM's European subsidiaries with genocidal regimes in the 1930s.

All of this is well known and even IBM's Wikipedia article cites Black (2001) extensively. So why was none of this mentioned, and why did I get a selective highlight of IBM's involvement in the US government and military?

Beyond the technical challenges (or the enormous water and energy consumption), should the training and filtering mechanisms of LLMs not be open to scrutiny?