So even when the generative ghost maybe hallucinated some detail or something that was factually incorrect according to, you know, the, the real person's life, the, the people interacting were still sort of willing to give it the benefit of the doubt, just sort of overlook those?

Yeah, there was a lot of suspension of disbelief, right? People enter the experience, um, kind of knowing and understanding that they're about to encounter something with some level of theatrics, um, kind of behind the scenes.

Uh, and we found that they-- there were elements of the conversations that really took people out of the experience, but they were almost never factual inaccuracies.

Uh, there was this whole other layer of the conversations that stuck out to them, um, that was a really fascinating finding for our study.

Things like lack of emojis or too long or too short of responses.

And we found that actually the way that they had communicated with their loved one when they were alive was really coloring how they were judging the LLM's responses.

If they had only ever texted with this person, they were much more likely to be seeking out short clipped responses with emojis or reactions.

So I'm curious, what's your read? Is this, uh, a healthy kind of interaction? And did anything you found in the study sort of challenge any prior assumptions that you had?

So even when the generative ghost maybe hallucinated some detail or something that was factually incorrect according to, you know, the real person's life, the people interacting were still sort of willing to give it the benefit of the doubt, just sort of overlook those.

Yeah, there was a lot of suspension of disbelief, right? People enter the experience kind of knowing and understanding that they're about to encounter something with some level of theatrics kind of behind the scenes.

And we found that there were elements of the conversations that really took people out of the experience, but they were almost never factual inaccuracies.

There was this whole other layer of the conversations that stuck out to them that was a really fascinating finding for our study.

Things like lack of emojis or too long or too short of responses.

And we found that actually the way that they had communicated with their loved one when they were alive was... was really coloring how they were judging the LLM's responses.

If they had only ever texted with this person, they were much more likely to be seeking out short-clipped responses with emojis or reactions, where if they only emailed with this person, they were much more willing to endure longer, more narratively driven responses.

So I'm curious, what's your read? Is this a healthy kind of interaction? And did anything you found in the study sort of challenge any prior assumptions that you had?

Mm-hmm.

When I was young.

Had a heart condition and passed away in 2012.

Um, and one of the things that inspired me to seek out other researchers doing this kind of work is we spent years and years attempting to build a memorial, um, that felt meaningful and impactful, a place where the community could come together.

And after about five years, we had finally got it done, every weekend doing bake sales and, and bowling fundraisers.

And as I entered into my studies, I kept thinking to myself, "Surely this is the kind of thing that technology can help assist.

Um, surely we can lower this barrier to entry and allow more people to create these kind of meaningful experiences," and that kind of led me to find the work that we're doing now in Generative Ghosts.

Mm-hmm