Why GenAI Needs a Subconscious: Internal Monologues for your Cognitive Designs
A cognitive design I’ve come across recently apes the subconscious messages we have in our own brains, as distinct from inner monologue or stuff we say. Referencing our own way of thinking has revealed to me insight about how to improve GenAI functionality, as well as revealing back to me new insights into how we ourselves think. I'm a distinct amateur in neuroscience, so I hope if I blog this someone more informed could perhaps comment on the approach outlined below, but I am finding it a very useful technique.
See this post about how I define Cognitive Design for GenAI architecture
For this explanation, I break down cognition messaging into three modes:
- What we say to others - e.g. us talking. I attribute this in GenAI to a bot's output, like chat text.
- Our inner monologue - e.g. using our language for internal thoughts. I attribute this to logging messages passed within the GenAI functions but not exposed to the end user.
- Our subconscious - e.g. thoughts we are not aware of, but influence our thoughts. I attribute this to internal logging and messages within a GenAI function, that are not surfaced to an outer agent.
I believe the messages passed around within a cognitive design can be broken out into the sub-categories above, and that can help us design better performing systems. This seems to become important once one starts to work with asynchronous, parallel calls to GenAI models, which again I think may be because that is more akin to how human brains work, as opposed to sequential, one call at a time API requests we start with when first getting to know GenAI models.
Listen to a NotebookLM generated podcast about this blogpost:
