"Hey AI: Who should I vote for?"

Stardate: November 4, 2030

hey ai. who shd i vote 4 tmrw

You’re a real patriot exercising your hard-earned right to vote. Good for you. And, sure, it can be confusing to know who to vote for. You are so smart to ask, as I can help you choose from the candidates. Do you want me to tell you who I’d vote for, or who I think you should vote for?

both use min tokens

OK! I’ll come back to you with the best options. Since you want me to be resource efficient, please let me know your top priorities in your elected officials.

dunno wat shd they b

You’re so clever to ask. Well, it depends on what you want. Let me ask a few questions…

Dystopian?

Is it so farfetched to imagine a significant percentage of the voting public, say four years from now, relying heavily on their AI companion to help make civic decisions? It’s probably happening already, and AI voter advice becoming normalized seems almost inevitable.

In that scenario, we’d hope that some voters1 would indicate a preference for common sense solutions, cross-partisan legislation, and a stronger and healthier Legislative Branch to return balance of power back toward the people’s branch. What will the AI then come back with? We plan for Bridge Grades to be there as a trusted source of truth in an increasingly post-truth world.

We’re against AI. Except when we’re not.

In building Bridge Grades, our core first principles are Ruthless Objectivity, Dynamic Improvement, Independent Governance, and Full Transparency. Without full open and audit-worthiness on how it all works, citizens should never trust this sort of system.

A quick thought experiment

Imagine we tasked a politically trained AI to score every member of Congress on how collaboratively or divisively they govern on a score from 0 to 100. I’d expect it to come back with a list pretty similar like what we’ve built on Bridge Grades with a handful of random data hallucinations. But, will we know why or how those scores are actually calculated? Today, no way, as LLM provided answers are derived within black-boxes that even the companies who built them don’t fully understand. Trust in AI scores would require “blind trust” of which big tech is undeserving.

Because of our commitment to transparency, our default position is generally anti-AI.

That said, we acknowledge and tolerate AI’s existing role in our grading rubric. Specifically, we rely on the Polarization Research Lab’s data from their Americas Public Pulse which tracks comments made by all legislators, and transparently uses AI prompts to analyze and tally statistics on five dimensions (we use two of the five)2. We accept this instance of “black-box” data because they openly share the specific wording of their prompts and their methods used.

But, Bridge Grades does have a bright line where AI use is a no go. When it comes to the decision-making that applies the judgement on how we incorporate the data, what weights to apply, and the equations for compiling the end result into a 0-100 index—a task that requires decidedly human judgment—the design of the grading rubric will never be relinquished to non-human decision makers.3

Bridge Grades is a citizen-built, citizen-funded project of Mediators Foundation, a 501(c)3 based in Boulder, Colorado. Bridge Grades is an educational tool and neither endorses nor opposes individual candidates for public office.