107 - Cognitive Architecture

My latest paper, covering, consolidating, and reframing some elements of what it takes to build a working cognitive architecture, is now available in pre-print. It covers how dynamic complexity emerges from human-like systems, which grow and evolve over time, and how those capacities were instantiated into software for our previous research system in 2019 (ResearchGate). That system demonstrated each of the necessary capacities for non-trivial definitions of AGI by January 2022, which are also the only definitions and systems that are compatible with alignment.

This paper covers concepts of computer science, neuroscience, and psychology, and places them in the context of complexity and how humans navigate it using emotions and cognitive biases, as well as how software is able to do the same. It is intended to lay the groundwork for educating narrow AI experts on the basics of how non-trivial forms of AI are fundamentally different from trashbots (LLMs) and RL systems, and the capacities that are unique to those differences.

Feedback is welcome, as are conference recommendations for where to submit it. I have another paper and thought experiment that could be submitted to conferences as well, but I generally don't go looking for them, so they'll merely be sent to some of my previous venues next year otherwise.

I have placed a couple of Easter Eggs in the paper, taking a lesson from Neil DeGrasse Tyson's MasterClass and integrating humor into scientific communication.