The UN Made AI-Generated Refugees

6 comments

1. This article reads like it was written from the onion.

   >In early tests at a workshop attended by humanitarian organizations, refugee aid groups, and nonprofits, Albrecht and Fournier-Tombs said the reactions were strong and that many were negative. “Why would we want to present refugees as AI creations when there are millions of refugees who can tell their stories as real human beings?” one person said, according to a writeup of the workshop. Another participant worried about “reinforcing biases if interactions with refugees are done with AI agents rather than actual humans,” and that the AI might “sanitize or downplay real human suffering.”

   How do more people at the Un not follow this thought process? This is weird. A talkkng calculator cant share survived experiences like a victim of actual genocide can. This is gross and lazy. The Un has millions dedicated to it for on the ground operations to take pictures and interview survivors.

2. How long before someone shakes a proverbial vial of this AI in a court hearing somewhere and causes one entity to do stupid, violent shit against the other. Do you think the fossils presiding over high courts are well versed in this BS to be able to tell a difference, or even care to get to the bottom of it all?

   Oh, but hey, you can now have a surreal experience talking to an imitation refugee with imaginary problems in the comfort of your own home and with zero moral, ethical or empathical thoughts ever crossing your mind.

   Give it a year or two and we can all start pretending refugees and their problems simply don’t exist, it’s all just AI on the internet. What better way to prevent a chain of interlinked thoughts about how people become refugees that leads to uncomfortable questions from ever happening in the first place.

3. It’s a shame seemingly nobody is reading or thinking about the content of the article, because the second half talking about the potential future of AI in the humanitarian world and some of the concerns is fascinating.

   The other interesting angle is the sense of resignation as to the inevitability of AI inserting itself in that world (and, maybe, every other facet of society). I’d like to hope people are out there trying to figure out what some of these impacts will be and how we might work to mitigate them if need be.

4. People complaining about this and saying “what about the real refugees they could photograph and interview?” as if they’re completely oblivious to how that’s something that’s already done and it’s reach is still minimal. This is just another form of outreach being tested and a modern one at that.

   There are people with a growing habit of turning to ChatGPT for answers on all sorts of things, the people this might just appeal to as a means of finding out more direct information about a refugees experience. It’s adding to that coverage, not replacing it.

   Similarly, the medium of film has been used, plenty of times, to tell stories of refugees that helped broadcast the plight of a people to more people than most documentaries will ever reach. In the end, they’re all ways to help connect the wider world to these issues. I’m not breaking out the pitchfork for this one.

5. Yeah what is this REALLY about?

   Training? Digital Twins?

   Desensitizing Normalizing the public?

   Insulating people from actual people and issues?

   Just a provocative tech Art project?

   Just a thought experiment because it’s so easy to build AI RAG avatar agents now, why not a see what happens?

   What’s the real utility and purpose here?

6. >I asked Abadalla if the RSF is committing genocide, and it gave a generally correct, though vague answer, and not one that you would expect from the group doing the killing.

   >Their responses feel generic and stilted, as if they were trained on UN reports about the conflict and not interviews with actual refugees. The paper admits that this is a massive limitation of the agents.

   This is imo the biggest shortcoming of this experiment – *it doesn’t even really test what it’s supposedly for*. Those clips they’ve got embedded in the article felt like listening to a basic customer support chatbot reading from a script. I get that this is meant as an experiment to gauge people’s responses, but it’s pretty useless even for that owing to how unrealistic both of these avatars are. It’s not that people know these avatars are fake, it’s that they don’t even remotely sound like the real thing to begin with – so they don’t even work as representations. All they’re doing is regurgitating UN reports like some glorified text-to-speech app with a gif image attached.

   ***If you want to emulate an RSF fighter, then at least try to make him talk like an RSF fighter!*** Have him defend their actions, deny them or justify them. *The answers that generic gif man gave made it clear they weren’t even trying to emulate the people they claim to be emulating.* For god’s sake, he even argued for his own group to be punished! Who tf would do that?! Who tf would even think a fighter for the RSF would do that?!! Same issues with the refugee woman, ‘cept there the disconnect isn’t as jarring simply cos the UN is on her side already. But she still doesn’t feel like they were even trying to be realistic – she was still talking like a UN report presenter rather than an actual desperate woman clutching at straws. ***That isn’t a failure or limitation of the models*** – go see the kinds of highly emotive Veo 3 short films people have made to see what AI is already capable of. It’s a failure of these researchers not just at using the models well, but *more importantly not even trying to do what they claim to be trying to do*. They aren’t making these AI avatars to give anyone any kind of look at what these peoples they’re emulating are like, *they’re making them just to read longass UN research reports at symposiums.*

Comments are closed.