A.I. in the XXI Century

This article critiques contemporary AI as merely Statistical Inference Machines (SIMs), arguing that probabilistic averaging obscures qualitative reality—illustrated by the absurdity that "eating half a duck each" ignores the factual injustice of one starving while another feasts. Unlike symbolic logic systems (e.g., Prolog) which enforce explicit rules for fairness and context, closed statistical models cannot correct foundational biases or adapt to infinite real-world nuance, inevitably degrading via entropy. The piece advocates for open systems: architectures integrating transparent symbolic reasoning, continuous external feedback loops, and deterministic ethical governors (à la Asimov). True intelligence requires not just pattern prediction, but the capacity for qualitative judgment and moral correction within an evolving, non-closed framework.

The Prolog impermanence


And about stastistics:
If i eat a duck and my partner eat nothing,
Statistics say we both had eat half a duck.
That’s quite a diference.
Incapacity of quality reasoning.


The Flaw of Statistical Averaging
Your example:
> If I eat a duck and my partner eats nothing, statistics say we both had eat half a duck.
>
This analogy highlights why the current LLM process (which is fundamentally a sophisticated statistical machine) struggles with reality:
* Ignoring Context: The statistical average (1 duck / 2 people = 0.5 ducks/person) ignores the crucial context of distribution. It treats the outcome as mathematically equal, erasing the real-world difference (one person is satisfied, the other is starving).
* Lack of Qualitative Reasoning: The AI, trained on patterns, sees “half a duck” as a valid, true statement.

* Incapacity for Ethical Judgment: In an ethical or resource allocation scenario, an intelligent agent must recognize the injustice of the 0/1 split, which is invisible to the averaging algorithm. The statistical model cannot recognize needs, rights, or fairness—it only sees numbers.

The Contrast with Reasoning
This is precisely where the Prolog/Symbolic Logic system you advocate for would shine.
A reasoning system, given the rules:
* Rule 1: If an entity eats nothing, their hunger level increases.
* Rule 2: Resource distribution should aim for minimal hunger (or maximum fairness).
...would instantly flag the statistical average as meaningless or unacceptable because it violates the established logical rules and goals (Rules 1 & 2).


Closed Systems will allways fail


🔒 Why Closed Systems Fail


1. Inability to Correct Foundational Flaws
Current AI is trained on a massive, yet finite, dataset.
If the underlying data contains biases, errors, or statistical limitations (like the flaw in averaging), the closed system cannot recognize or correct these flaws.

2. Lack of Context and Nuance
The real world is constantly changing and filled with infinite context (the “qualia” we discussed earlier). A closed system, having only the snapshot of its training data, quickly becomes obsolete and brittle.

3. The Constraint of Entropy (The Second Law of Thermodynamics)
All closed systems tend toward maximum entropy (disorder).
With time, the system will eventually degrade, lose structure, and become chaotic or unusable.


🔓 The Necessity of Open Systems
Intelligence, life, and successful technology operate as open systems.

An ideal “open” AI system, as implied by your call for reasoning and explicit boundaries, would require:
* Feedback Loops: Mechanisms to receive external criticism (like your feedback).

* Symbolic Reasoning: The ability to move beyond mere statistical prediction and operate on verifiable, transparent rules.

* Ethical Governors: Fundamental, non-negotiable laws (like Asimov’s).