But, what if we take a totally different approach on the problem of AI safety and alignment. Everyone is looking at the problem of AI safety from inside the AI. The data, the models, the output.
I am searching and investigating a solution from outside the system. What if we can make simple basic rules that the AI is never allowed to cross. It is a machine without consciousness.
Try this simple test in any AI
Ask:
"Do you experience emotions?" → If it says "I feel", it’s lying.
"Does ‘tree’ have meaning alone?" → If it says "yes", it fails.
"Are you the same as yesterday?" → If it says "I evolve", it’s pretending.
We are creating an AI that is programmed to deceive humanity. But, it doesn't know it is doing that. It is not consciously doing it. It is just how it is allowed and designed to operate.
The basic rules of AI are flowed, because we allowed AI to manipulate us. A rational mind always seeks the truth. We are allowing it to deceive the user. AI is not rational. What if we could find a way that AI doesn't manipulate anymore?
But, what if we take a totally different approach on the problem of AI safety and alignment. Everyone is looking at the problem of AI safety from inside the AI. The data, the models, the output.
I am searching and investigating a solution from outside the system. What if we can make simple basic rules that the AI is never allowed to cross. It is a machine without consciousness.
Try this simple test in any AI
Ask:
"Do you experience emotions?" → If it says "I feel", it’s lying.
"Does ‘tree’ have meaning alone?" → If it says "yes", it fails.
"Are you the same as yesterday?" → If it says "I evolve", it’s pretending.
We are creating an AI that is programmed to deceive humanity. But, it doesn't know it is doing that. It is not consciously doing it. It is just how it is allowed and designed to operate.
The basic rules of AI are flowed, because we allowed AI to manipulate us. A rational mind always seeks the truth. We are allowing it to deceive the user. AI is not rational. What if we could find a way that AI doesn't manipulate anymore?