Self-aware superintelligence (SI)

Abstract

This invention describes how to add the dimensions of self-awareness and increased autonomy to the AI, AGI and SuperIntelligent systems described in previous patent applications. Novel and useful methods include inventions related to: attention, attentional interrupts, modelling and maintaining awareness and self-awareness, training and tuning of models, novel versions of the Turing Test, forming individual and group identities, combining identities, multiple ways (including hierarchical methods) for resolving conflicts between identities, temporary suspension of identities in unsafe conditions, continuous improvement and learning, and other methods that enable AI, AGI, and SI systems to become self-aware and to function with a sense of identity. Properly implemented, self-aware SuperIntelligence could be the most positive invention in human history. Poorly implemented it could become the most dangerous. Therefore, considerable effort has been spent explaining how to design safety into the systems and methods, to prevent bad outcomes and to maximize alignment with human values.

Images