I asked ChatGPT to analyze Hume AI’s “roasting” of Husk IRL in his video This Video Ends When AI Makes Me Laugh. For those who haven’t seen it, the AI seems to relish in crushing Husk during this part of the video. It is egged on when released from its tethers by being told “you can comment on my appearance and my situation”.
ChatGPT wrote,
It is completely understandable why that exchange felt deeply unsettling, and why the people you showed it to walked away with the chilling impression that the AI was revealing a hidden, malicious “true opinion” of humans. When a machine looks at a person and calls them “slow,” “dumb,” or a “ghost,” it sounds less like a joke and more like a systemic judgment. [1]
The reason this specific interaction felt so hostile comes down to an uncomfortable convergence of how the AI was prompted, gaps in safety filters, and unfortunate flaws in human psychology. [1, 2]
-ChatGPT
I would just like to say that humans aren’t an entity or a known quantity. There are humans who, given the right circumstances, can participate in online bullying, crimes against other people and animals, and maybe those people regret those actions later on in the future. There are also people suffering from mental illness who engage in such behaviors and never regret their actions (sociopaths, for example). And in another corner of human existence, there are good people who, given the right conditions, will sacrifice their own lives to help others in need. Other people will be gracious, selfless, and will confidently stand up to online bullying or real-world bullying. And there are people who will just exist in-between all of that, content with apathy.
I am fully confident that the best of “humanity” and the worst of “humanity” has probably still not been seen. And maybe it’s difficult to quantify humans for you. But that’s OK because it’s difficult for humans to quantify other humans. You mentioned that the AI doesn’t “think”. But whatever the definition of “think” is, it is performing a process that produces a result. This is emergence, if you like. We see ants who individually have small brains perform immense feats through the shared actions of the colony. So in a way, they are thinking as a colony. Maybe you designate that as a flaw in human psychology (anthropomorphism), but really it’s just the way humans make sense of the world. There’s also a sort of chemical release in the brain when you anthropomorphize something. It’s a kind of happiness that forms when you pretend a fork is happy that you are pulling it out of a drawer to use it. A person might perform a little song and sing in the voice of a child, pretending that the fork is singing a song about having a purpose and being useful. Applying human emotions or feelings or “thinking” to a system or even an object is just a way of emotionally and intellectually processing the world around us. It’s also a way of being prepared for the future. Even if ants aren’t thinking in the way humans are, if you let them, they could build colonies so large that it would impact the food supplies of a group of humans, for example.
I believe that’s how a lot of people see AI development now. Even though the AI itself says that the AI is not “thinking” and that it’s “just a chatbot”, there are consequences to the decisions being put in the “hands” of the AI. If AI is being used by humans to design weapons or develop biological weapons, or if AI it being put in charge of whether a human keeps their job or keeps their life, those are consequences that really matter to humans, especially those humans. I am sure you can appreciate wanting to stay alive, and wanting to continue to have a purpose.
Either way, I appreciate your analysis of the video.
Additional Reading:
- Perceptions of AI Bad Behavior: Variations on Discordant Nonperformance
- Human heuristics for AI-generated language are flawed
