Aletheia

The prevalence of artificial intelligence systems, specifically generative artificial intelligence systems based on large-scale, billion and multi-billion parameter transformer models of human language, has become deeply integrated into nearly every existing major software suite on the market. It becomes appropriate to ask: what are the cognitive trade-offs, and how should we regard this technology that has now propagated into the mainstream so deeply? Namely, should Generative AI (GenAI) be treated as just another tool that can extend human ability, an extension of the human body and will, like a hammer? Should it be treated as deterministic automation? Or should it be treated as something else entirely?

In this perspective paper, I will argue that extension of cognition while removing sensory feedback through multiple layers of abstraction, severs the situated environmental connection needed for true skill acquisition in novices, and skill retention in experts.  To do this, I will frame current interactions with GenAI through variations on a proposed Quad-Process Model (QPM) of cognition from the fields of artificial intelligence and robotics, which are variations on Kannanman's Dual Process Model which breaks cognition into two complementary “systems” of cognition.

I will then make the argument that GenAI should not be treated as a pure mechanism for extended cognition in the manner in which it is currently being incorporated into the cognitive systems of human organizations. In fact, it is precisely the facade of extended cognition that creates organizational and operational risk in GenAI implementations through auto-regressive tendencies, and forced abstraction.

I will make this argument on the basis of three main claims:

1. The use of language is a symbolic abstraction layer that sits between the environment and environmentally situated signals, boundaries, and skill based affordances.
2. This abstraction inherently reduces or eliminates systemic, sensory and signal feedback loops which undercuts the ability of individuals and systems to develop or maintain assembled affordance-based skills in their environment.
3. The inability to develop skills and dynamically tune to new sensory signals means that system resilience and precision will degrade over time due to a reduction in situated, environmental intuition and skill acquisition, creating regression to the mean and the potential for catastrophic cascading failure.

Taken together, I argue that GenAI should be treated as an extensible abstraction layer based on existing cultural and symbolic artifacts of meaning (System 3). Not as a foundational pre-cognitive processing layer (System 0). More clearly defined domain boundaries must be drawn between integrated cognitive tools and the human individual or collective human systems organized around them. To treat GenAI as System 0 is to invite skill atrophy, technical debt, and the impingement of system metacognitive ability.