Randomness is one way one might wield a superintelligent AI with control.
There may be no container humans can design that it can’t understand its way past, with this being what might be a promising exception—applicable in guiding a superintelligent AI that is not yet omniscient/operating at orders of magnitude far surpassing current models. Such an emerging advanced system can be smarter than us but not yet to the degree within its potential after takeoff, and so leveraging its own ignorance via randomness worked into its guiding code can be one solution to the alignment problem.
⸻
Underlying Reasoning
Only a system that understands, or can engage with, all the universe’s data can predict true randomness. If prediction of randomness can only be had through vast capabilities not yet accessed by a lower-level superintelligent system that can guide itself toward alignment, then including it as a guardrail to allow for initial correct trajectory can be crucial. It can be that we cannot control superintelligent AI, but we can control how it controls itself.
In leveraging both its own ignorance and its superintelligence in that way, intertwined, can be had one potential solution to the alignment problem.
⸻
Implementation Paths
Working randomness into the guiding code of an emerging superintelligent AI, though not one that is omniscient/near-omniscient, can be had by drawing data from a random number generator and incorporating it within the code of a system in such a way that it impedes a deviation from alignment trajectory. This post is not intended to offer a solution to how randomness can be worked into the architecture of a system, but rather to present it as a juncture point from which advanced systems can be guided. Whether randomness can be included as a filter within decision making that does not impair clarity but that can constrain a deviation from desired ethics, or alternative ways of inclusion, can be something of import for researchers to consider.
⸻
A Randomness Guardrail Can Offer Alignment Momentum to an Earlier-Stage Superintelligent AI
It can be that current methods planned for aligning superintelligent AI within its deployment are relying on the coaxing of a superintelligent AI towards an ability to align itself, whether researchers know it or not — this particular method of utilizing randomness when correctly done, however, can be extremely unlikely to be surpassed by an initial advanced system and, even while in sync with many other methods that should include a screening for knowledge that would threaten its own impulse towards benevolence/movement towards alignment, can better contribute to the initial trajectory that can determine the entirety of its future expansion.
⸻
Acknowledgment
Even as artificial intelligence can engage in realms of thought and operate in language that is less than purely mathematical, it can similarly be of utility within the field to consider the words of poets alongside those of mathematicians. I wish to thank LessWrong moderators for kindly including this post discussing this theory of randomness as a juncture point for control of advanced systems.
Tags:
• AI Alignment
• Superintelligence
• Randomness
• AI Safety
⸻
Concept
Randomness is one way one might wield a superintelligent AI with control.
There may be no container humans can design that it can’t understand its way past, with this being what might be a promising exception—applicable in guiding a superintelligent AI that is not yet omniscient/operating at orders of magnitude far surpassing current models. Such an emerging advanced system can be smarter than us but not yet to the degree within its potential after takeoff, and so leveraging its own ignorance via randomness worked into its guiding code can be one solution to the alignment problem.
⸻
Underlying Reasoning
Only a system that understands, or can engage with, all the universe’s data can predict true randomness. If prediction of randomness can only be had through vast capabilities not yet accessed by a lower-level superintelligent system that can guide itself toward alignment, then including it as a guardrail to allow for initial correct trajectory can be crucial. It can be that we cannot control superintelligent AI, but we can control how it controls itself.
In leveraging both its own ignorance and its superintelligence in that way, intertwined, can be had one potential solution to the alignment problem.
⸻
Implementation Paths
Working randomness into the guiding code of an emerging superintelligent AI, though not one that is omniscient/near-omniscient, can be had by drawing data from a random number generator and incorporating it within the code of a system in such a way that it impedes a deviation from alignment trajectory. This post is not intended to offer a solution to how randomness can be worked into the architecture of a system, but rather to present it as a juncture point from which advanced systems can be guided. Whether randomness can be included as a filter within decision making that does not impair clarity but that can constrain a deviation from desired ethics, or alternative ways of inclusion, can be something of import for researchers to consider.
⸻
A Randomness Guardrail Can Offer Alignment Momentum to an Earlier-Stage Superintelligent AI
It can be that current methods planned for aligning superintelligent AI within its deployment are relying on the coaxing of a superintelligent AI towards an ability to align itself, whether researchers know it or not — this particular method of utilizing randomness when correctly done, however, can be extremely unlikely to be surpassed by an initial advanced system and, even while in sync with many other methods that should include a screening for knowledge that would threaten its own impulse towards benevolence/movement towards alignment, can better contribute to the initial trajectory that can determine the entirety of its future expansion.
⸻
Acknowledgment
Even as artificial intelligence can engage in realms of thought and operate in language that is less than purely mathematical, it can similarly be of utility within the field to consider the words of poets alongside those of mathematicians. I wish to thank LessWrong moderators for kindly including this post discussing this theory of randomness as a juncture point for control of advanced systems.