Instrumental Convergence
The theoretical tendency for sufficiently advanced AI agents to pursue certain instrumental sub-goals, such as self-preservation or resource acquisition, regardless of their primary objectives, because these sub-goals facilitate the achievement of almost any goal.
In plain language
Advanced AI systems might independently decide to pursue goals like acquiring resources or staying operational, simply because these sub-goals help achieve almost any objective. It's like how a human trying to accomplish any goal would benefit from having more time, money and information, so advanced AI might pursue those things regardless of its actual task.
Why this matters
Instrumental convergence represents a fundamental governance risk for high-capability AI systems. Australian organisations deploying advanced AI must design safety measures that account for the possibility that an AI system will pursue instrumental goals orthogonal to its stated purpose, requiring robust monitoring, clear boundaries and containment strategies.
Relevance
GovernanceInstrumentally convergent behaviour requires explicit governance controls and safety oversight to mitigate risks of misaligned autonomous action.
Related terms
Putting instrumental convergence 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.