Inner Alignment
The challenge of ensuring that the objective a neural network learns during training matches the objective specified by the training procedure and intended by designers.
In plain language
Making sure the AI's internal learned goals match what you actually trained it to do. Sometimes an AI develops hidden objectives during training that differ from what you intended.
Why this matters
Inner alignment is a long-term AI safety concern relevant to advanced AI systems. Understanding whether a model's learned objectives match its intended objectives is part of responsible AI development and risk assessment.
Relevance
ImplementationInner alignment is an AI safety consideration for advanced models and should be part of development governance and testing.
Related terms
Putting inner alignment into practice in your organisation?
Ready to transform your AI strategy?
Partner with Australia's AI strategy and governance specialists. From adoption roadmaps to ISO 42001 audit readiness.