We propose a multi staged process for worldview diversity conserving AI system development.

Rather than basing AI system creation on one particular set of values, as many different perspectives as possible should be identified early on in the implementation process.

Only perspectives which are compatible with human rights are acceptable.

These perspectives can each form the basis of alternative system requirements.

Once the different perspectives are defined, a separate stage for integrating these perspectives can try to develop a consensus.

If a consensus has been found, the final system implements this consensus.

If no consensus has been found, the system is implemented in different versions.

After system creation, users should also have choice.

Values and bias that have informed the creation of the system and the selection of data for training the system have to be exposed to the user.

The user then can take informed choices.

Once users are able to choose such values freely, they can assume responsibility for the consequences of their choices.