I am interested to know more about what you propose.
We are a UK/Belarus start up looking to work with AI researchers on our new platform. We too are concerned about creating safe AI and it is our opinion that the CONTEXT in which safe AI needs to develop is social/democratic. Online is neither particularly democratic nor social, hence any AI using this as its fundamental point of reference may not evolve with fundamental principles intact that ensure it is safe.
We are already building the platform from which we propose an AI can interact with humans and vice versa, with each side growing and learning in a democratic manner, and from there, once an AI has learned to assign meaning in a structured environment, it can go onto infer meaning in an unstructured environment.
Happy to talk further.