Can AI Advisors Navigate Moral Dilemmas with Grace?

AI has steadily been stepping into a new role: artificial moral advisor. It's an intriguing development, but what kind of conversational patterns should these advisors really display? In the pursuit of ethical guidance, researchers have been diving into how these models can help interlocutors 'stay with the uncertainty.' And let's be honest, uncertainty is plentiful in moral quandaries.

The Modes of Uncertainty

These AI moral advisors, or AMAs if you want to get technical, were tested across different modes of uncertainty: Perspective-Multiplying, Tension-Preserving, and Process-Reflecting. What's fascinating here's how these different modes compare to more straightforward strategies like being Baseline, Persuasive, or even a bit Sycophantic.

Now, you'd think one model might dominate, but that's not the case. Open models align with human ambiguity through the divergence of perspectives. On the flip side, closed models do a lot of internal hedging, giving them a different kind of edge. It's a curious scenario, raising the question: can AI really capture the full spectrum of human indecision?

Personas and Their Impact

Two persona prompt formats were put to the test: Declarative and Narrative. The declarative personas excel at showcasing stance diversity from the get-go, while the narrative ones shine in depicting realistic belief revision. In other words, they might start solid, but they're flexible enough to bend without breaking.

Here's the kicker: all six AMA strategies managed to create unique conversational patterns. The difference wasn't in how much opinions changed, but how well these strategies kept the conversation engaging. So the real takeaway? Quality of engagement might just be more important than the amount of stance revision.

Why It Matters

Why should this matter to anyone? Well, AI is becoming increasingly integral to our lives, and its role in ethical guidance is only set to expand. If these models can sustain meaningful and engaging conversations, they might just become effective guides in the murky waters of moral dilemmas.

But here's the million-dollar question: Are we comfortable with AI stepping into this space? As these AMAs inch closer to mimicking human-like conversations, it's essential to ask if they're truly up to the task, or if we're projecting too much onto lines of code. Only time, and more research, will tell how these digital advisors will evolve.