Mission to Mars

The astronaut has been conditioned over a very long period, since birth, to the Earth’s gravity, its atmosphere, its circadian cycle, and its seasons. The astronaut is going somewhere where all of that is wrong, and making a mistake in unconscious processing is potentially fatal. It would help to have a workmate who is fully trained in all the differences Mars represents, doesn’t get tired, bored, distracted, and doesn’t have the human’s Four Pieces Limit in times of stress.

Active Structure #AGI can provide such a workmate. Just as the human astronaut will be running unconscious processing on threats, Active Structure can run continuous threat processing against new threats that no one had thought to put in the Operating Manual.

The deep conditioning of being born and raised on Earth is impossible to undo.

So why can’t we use ANNs (Artificial Neural Networks) or LLMs (Large Language Models) to achieve the same result?

ANNs are a static piece of kit that has been programmed months or years before. Attempts can be made to avoid future threats, but only in a static way. It has no mechanism to track a rapidly evolving threat. A flying bird tracks other objects and notes any whose angle from the bird’s flight doesn’t change, indicating a collision course, and generating avoidance behaviour. A bird has dynamic memory (from real neurons, not ersatz ones) being updated every few seconds, which it can use to compare what was, and what is – an #ANN has nothing like that. A bird is a dynamic actor – an ANN is a static actor.

LLMs hold many thousands or millions of pieces of text, can summon a piece of relevant text relatively easily, and can cobble together several pieces of text with much less accuracy. An #LLM doesn’t understand the meaning of a single word, so what it constructs is very unreliable, and is predicated on the piece of text already existing. In a new and challenging environment, it is not likely to serve well as critical life support.

Article published on our AGI Defence blog.