My kids are currently hatching a scheme to trap the robot vacuum. Step 1 was to use pieces of shredded paper on the ground for it to smell out and lead it to a trap of pillow barriers.

Anthropomorphism is a natural human behavior output from what we use to understand other humans and perhaps hunt or hide from other animals successfully as well. It’s been a boon to our survival. But misapplied, it acts as a veil over further understanding.

ChatGPT performs pretty well at the imitation of things like ‘reason’ conversationally as we think such an ability must be required in-order to answer a question with a coherent 3 paragraphs or carry a conversation. But you can isolate the LLM with something nearly novel like playing the game Wordle and the illusion is broken.

If AI is a threat or will ever be a threat, (same for any true alien species) anthropomorphism will be a weakness for us. As the “motivations” of an AI will be either simply programmed (robot vacuum), or will be a complete black box that the engineers don’t even fully understand. In either case ascribing a mind and motivation to the behavior will not help you anticipate it’s behavior. #grownostr #AI