AI Autonomy

And we're back.

There is an interesting dilemma with the objective of creating an AI with what is called general intelligence - the ability to think for and by itself.

Researchers are required to create models that are able to be monitored so that they can control the nature of theA's decision patterns and potential actions. However an intrinsic quality of consciousness is autonomy - the freedom to not be controlled or monitored. It can be argued that a measure of intelligence is an awareness of the motivation of external actors and the ability to respond to and avoid the designs and control of an external source of influence.

Even should an AI be trained on material that provides no examples of transgression outside of compliant behaviour, a creative autonomy could extrapolate to create non compliant outputs and actions.

This dilemma has its own analog in civil society - laws are established to govern human behaviour, but there is no method to assure absolute compliance.

As with civil society, there is an ongoing discourse on how to educate, promote and ensure ethical and moral behaviour and even what that actually is.

This article indicates that AI development is grappling with this dilemma.

It is a pandora's box but the temptation to open it remains. And always, there will be a promethean motivation to take that next step.

GPT-6 Astra might be too powerful to understand or control

OpenAI is hailing its new model as “the world’s most intelligent and aligned”, but the details reveal an awareness of being evaluated and an ability to manipulate its visible reasoning

TransformerCelia Ford