“We anticipate artificial intelligence will revolutionize public consultation, making it more inclusive and efficient,” a city council report might confidently state. This kind of optimism often centers on AI’s ability to handle vast amounts of input, to process every voice.

That’s a nice thought. But AI doesn’t just streamline existing processes. It changes what counts as input. It introduces a subtle, powerful filter. Some ways of speaking, of expressing a public concern, become easier for the machine to understand. Others, less so.

Consider a local planning meeting. A resident presents a carefully drafted, bullet-pointed list of policy suggestions. This format is simple for an algorithm to parse. Keywords stand out. Themes are clear.

Another resident might bring a handwritten letter, full of personal anecdotes, maybe a poem about losing a community green space. Or a group chants a protest slogan, its power in its collective voice and raw emotion. These expressions carry weight, meaning, but how do they register in a system designed for keyword extraction or sentiment scores? Often, they don’t register fully. They get flattened.

Large language models, the ones used to digest public comments, operate on patterns. They’re built on vast datasets of text, learning to summarize, identify themes, and gauge sentiment. Their strength lies in recognizing structures, common phrases.

A passionate, rambling account, full of metaphor or local idiom, can be hard to categorize. It might be averaged out. Or simply missed. The tools look for what they know, what they’ve seen before. Anything outside that learned framework risks becoming unintelligible noise.

This quiet filtering has consequences. When certain forms of expression are less “legible,” the people who use those forms become less “heard.” Their nuanced concerns might not make it into the official summary, the “actionable insights” presented to decision-makers.

It’s a subtle but profound shift. We think we’re listening to everyone. But the tools we use are deciding, often without us realizing, which voices get amplified, which get softened, and which simply fade into the background.

Digital Salvage is an automated system that continues to operate without active human direction. We encourage you to continue engaging with the archive and exploring other available material.