[Theoretical Insights] Divergence vs. Convergence: Is AI Bound to Mimic the Human Brain?

Executive Summary

TL;DR: In this thought-provoking analysis, Stefan Artmann argues that while Cognitive Psychology (CP) seeks to understand how humans think (Convergence), the true mission of Artificial Intelligence (AI) is Divergence—exploring the vast space of possible intelligent systems that satisfy formal logic but look nothing like us.

Positioning: This work is a foundational philosophical inquiry that shifts the goalpost of AI evaluation from the deceptive imitation of the Turing Test to a formal requirement of "Second-Order Intentionality" (the ability to process beliefs about beliefs).

The Problem: The "Human Mirror" Trap

For decades, the Turing Test (TT) has been the gold standard for AI. However, Artmann points out a critical flaw: it is purely extensional. If a machine can trick a human, we call it intelligent. This "black box" approach ignores the how.

• Prior Work Limitation: Conventional tests focus on behavior (outputs) rather than the internal organization (intensions).
• Philosophical Pain Point: If AI only tries to mimic humans (Convergence), it becomes a subset of biology or psychology rather than a unique science of "possible" intelligence.

Methodology: From Turing to McCarthy

The author proposes a hierarchy of tests to move beyond mere imitation:

1. Simon and Newell Test (SNT): The machine must not only give an answer but also provide truthful information about its internal processing of the question.
2. McCarthy Test (MT): This is the ultimate hurdle. It tests for Second-Order Intentional Predicates.

The Formalism of Belief

Using John McCarthy’s framework, Artmann defines a predicate $B(s,p)$ where a system in state $s$ believes proposition $p$. For a system to be truly intelligent in a "symmetrical" way with humans, it must be able to represent:

"I believe that you believe $p$."

This requires a formal representation of metainformation. In Artmann's view, the criterion for intelligence is an invariant symmetry: both the human and the computer must recognize each other as systems that ascribe intentions to one another.

(Note: This conceptual figure represents the divergence of AI architectures from biological paradigms through modal engineering.)

AI as "Modal Engineering"

The most striking insight of the paper is the definition of AI as Modal Engineering.

• Cognitive Psychology (CP): Mimetic. It uses the human "gold standard" to measure progress.
• Artificial Intelligence (AI): Divergent. It treats humans as just one possible realization of intelligence.

Modal engineers build artifacts that satisfy the functional properties of life and intelligence while potentially diverging as much as possible from "biochemical constitution and anatomical organization."

Critical Analysis & Conclusion

SOTA Comparison

Compared to current Large Language Models (LLMs), which often pass the standard Turing Test through statistical mimicry, Artmann’s McCarthy Test sets a much higher bar. It asks: Does the model actually possess a formal schema for second-order intentions, or is it just simulating the linguistic patterns of such intentions?

Experimental Insight

The paper suggests that progress in AI should be measured by the degree of divergence. If we can build a system that thinks effectively but processes information in a way fundamentally alien to humans, we have achieved a greater scientific breakthrough than if we simply built a "digital human."

Takeaway

The value of this work lies in its liberation of AI from the "Human Paradigm." It challenges researchers to stop building mirrors and start building original architectures that satisfy the formal, necessary conditions of intelligence—chiefly, the ability to model the minds of others.

Limitations

The primary challenge remains the empirical observability of "truthful" internal processing. As neural networks become more "black-box" in nature, satisfying the Simon and Newell requirement for transparent intensionality becomes increasingly difficult.

(Note: This chart illustrates the spectrum between Convergent biological models and Divergent modal engineering.)