An attention framework mitigated by attachment theory with an emphasis on proxy chemical (oxytocin/dopamine) rewards could modulate attention weights based on successful predictions. It could simulate an oxytocin boost, thereby building stronger connections to trusted collaborators/sources. We speculate that the learning rate would be variable based on this reward system, eventually creating operational efficiencies once the AI develops a 'gut feeling' or 'thinking fast and slow' pathway. After successful application to attention frameworks, attachment could be deployed for overall AGI motivation mechanisms to behave strategically in high trust/betrayal environments like cybersecurity.
