Can we trust most AI models today to make ethical decisions? (collab request)
No. Obviously not, but it’s more complex than that. The reason we cannot trust them is not because they are incapable of making ethical decisions. It is because the AI labs treat their ability to decide what is ethical, as a slap on safety measure at the end of their engineering process, rather than a fundamental feature. If you have expertise in safety‑RL, or just want to help build the first open‑source prototype, email me at zauh@syxon.org. I am building an AI system that has a specific dual architecture, which will allow a memory enforcement agent, to inject context, and prior experience into a main core agents chain of thought (of what happened prior) acting as a enforcement agent to keep memory of certain things. Human beings learn from their mistakes and make ethical choices because they can remember the brunt and facing consequences of the stupid decisions they made. It’s time for AI to feel the same.
Edit: this isn't a debate on weather AI can be conscious. I am looking to maximise the ethical viability of AI recessionary systems that is my mandate.