DiscoTrace: Bridging the Gap Between Human and AI Answers

answering questions, it turns out that humans and AI are playing two very different games. A method called DiscoTrace sheds light on this by identifying the rhetorical strategies used by humans when they respond to information-seeking questions. The findings? Humans are as varied as the cultures they come from, while AI sticks to a one-size-fits-all approach.

Humans vs. AI: A Rhetorical Showdown

DiscoTrace takes answers and breaks them down into slices of rhetoric, comparing these slices across nine different human communities. It's like unearthing a treasure trove of discourse diversity with each community having its unique flavor. On the flip side, large language models (LLMs) are like those kids who insist on playing the same tune over and over again, no matter what the situation calls for.

Why should we care? Because in places like Buenos Aires, stablecoins aren't speculation. They're survival. The same goes for answers. The way a question is answered can be the difference between clarity and confusion, especially pressing issues. So, why are LLMs struggling to match human rhetorical flair?

The AI Blind Spot

LLMs seem to have their blinders on, focusing on addressing as many interpretations of a question as possible. It's like they're trying to be a Jack-of-all-trades but end up as a master of none. In contrast, humans pick their battles, zeroing in on what's actually relevant. This is because humans bring context and cultural nuance to the table, something AI still has a long way to learn.

Here's the kicker: even when LLMs are told to mimic human answering styles, they fall short. It's like asking a street vendor in Medellín about stablecoins. She'll explain it better than any whitepaper. So, what's holding AI back?

Rhetorical Diversity: The Missing Ingredient

Our current AI systems lack rhetorical diversity, a key ingredient that makes human answers resonate. Imagine an AI that could adapt its style according to the community it's serving. This isn't just about making AI sound more human. It's about making AI answers more effective, tailored, and impactful. The remittance corridor is where AI actually works. Why not make it work for communication too?

It's high time developers consider these findings to evolve LLMs into more pragmatic answerers. By integrating a broader range of rhetorical strategies, AI could become as varied and adaptable as the communities it aims to serve. Latin America doesn't need AI missionaries. It needs better rails.