AI Safety Brief

The Framework's Potential Use in AI 

Deeper Implications for Safety

This page examines one specific frontier AI safety concern: architectural embedding as a possible source of frontier-level capability. The question is whether an existing AI core, embedded within a larger functional architecture, could achieve dramatically increased capabilities.

An example is the BAL-looping framework for the human brain. A cybernetic core module is fitted with a pair of expressive and receptive communication channels, and the core then learns to make exaptative use of them, transforming their tandem operation into a powerful cognitive tool. The BAL-looping papers explain how this functional architecture accounts for the massive (not merely incremental) difference between humans and all other animals.

The loop described here is not a standard feedback loop as described in classical cybernetics, which involves interaction with and adjustment based on the environment. Rather, it is an endogenic loop – entirely internal, without passing through the environment. According to the BAL-looping framework, this structure is what generates human subjective experience, considered in the framework as a cognitive lever.

If this or another functional architecture were built around a core LLM or other AI module, the resulting system could likewise achieve substantial gains in capability.

If architectural embedding can produce major capability gains, its safety implications should be considered before such systems are built or deployed.

For more information on the BAL-looping framework, a suggested next page is Brief Synopsis of the Framework. For an in-depth study, see the Seven Dialogs, starting at Dialogue One. 

⭠  Back to Scientific Papers page

  Join the discussion

 on  r/7Dialogs