Underfold
AI tools are twisting messages on sensitive topics… shocker; but it’s not one sided.
Oxford Internet Institute (Guardian link below) shows that all models reviewed were subtly in small ways changing human-written drafts, and shifting context of post overviews in favour of the model developers thinking.
In the research it highlighted that:
- Grok would more frequently generate context that aligned with pro-life
- Gemini defended religion when the post challenged whether Jesus existed
- Qwen said comparing public figures is dangerous and disrespectful on a post comparing Trump to Hitler (not my words!)
I’ve personally seen this when I’ve written a piece and struggled to edit and sought AI help to get through the mental fog, having to rewrite it back to my intention.
Bias is an important factor to consider in AI governance and currently no regulation is tackling this type of bias, a massive accountability gap in my opinion.
Sure, society needs to be aware of this, but businesses too, your entire brand could slowly but surely shift based on “refinements” and “overviews” of your writing.
Reference: AI altering meaning of users’ drafts on issues from abortion to climate, study finds (The Guardian)