Instead of asking humans who they would vote for, try to understand the nuances of their thoughts and concerns, let those messages bubble up to candidates so they can adjust their campaign to meet voters’ demand, instead of that, why not just segment humans into a bunch of shallow stereotypes (the socialist Millennial, the conservative Boomer, the liberal city dweller, the rancorous rural voter who feels left behind…) and then have some AI agents replicate how those people would respond?
Surely nothing could go wrong.
Anyone who understands how LLM-based “AI” works is laughing at this. If you think polls are bad now, wait until ChatGPT hallucinations are part of the polling data set!