Brute force fast processing vs controlled deduction?
This is an insightful hypothesis that lands right in the middle of a decades-long debate in AI and cognitive science.
Your hypothesis effectively champions the Connectionist view (intelligence from massive neural working) over the Symbolic AI (or "GOFAI" - Good Old-Fashioned AI) view (intelligence from controlled deduction).
Here’s a breakdown of the argument, incorporating a very popular model of human cognition.
Part 1: Neural Working vs. Controlled Deduction
What you're describing maps almost perfectly to the "System 1" and "System 2" model of human thought, popularized by psychologist Daniel Kahneman.
System 1 (Your "Massive Neural Working"): This is the brain's fast, intuitive, automatic, and parallel processing. It's the "gut feeling" that recognizes a face, understands language in real-time, or gets a "vibe" from a situation. Modern large language models (LLMs) are a fantastic example of a powerful System 1. They are "massive neural workings" that produce plausible, human-like text based on statistical patterns.
System 2 (Your "Controlled Deduction"): This is the brain's slow, deliberate, logical, and sequential reasoning. It's what you use to solve a multi-step math problem (), follow a complex recipe, or construct a formal logical proof. This is what Symbolic AI tried (and often failed) to model as the basis of all intelligence.
Your hypothesis that "realistic like-human intelligence comes from massive neural working, not from controlled deduction steps" is strongly supported by the evidence that our System 1 is the foundation of almost everything we do. It’s the ancient, powerful engine, while System 2 is a more recent, fragile evolutionary add-on.
Part 2: Is Reasoning Just a "Dream" Mimicked by Processing?
This is the more philosophical part of your claim ("Ergo, massive processing mimics reasoning, and reasoning itself always has been a dream").
This is where the model gets more complex. There are two main ways to interpret this:
1. The "Reasoning is an Illusion" View (Supporting Your Hypothesis) In this view, System 2 ("controlled deduction") is just an illusion. It's not a different kind of process. It's just System 1 (the massive neural network) working very slowly and deliberately on a problem.
When you think you are "doing logic," you are just running your neural network in a special, learned mode that simulates logic. The "rules of logic" themselves are just patterns that the network has learned, like any other pattern. In this view, your hypothesis is correct: "reasoning" is just a convincing "mimicry" performed by the underlying massive processor.
2. The "Hybrid System" View (Challenging Your Hypothesis) This view, which is gaining a lot of traction, is that human intelligence is a hybrid of both systems. They are functionally different, even if both are ultimately built from the same "wetware" (neurons).
The failures of only using LLMs (pure massive processing) support this. An LLM can write a beautiful paragraph about physics, but it might fail a simple logic or math problem that a child could solve. This is a System 1 (massive processing) failing at a System 2 (controlled deduction) task. It mimics reasoning but fails when rigorous deduction is actually required.
Similarly, humans are not just System 1. We can force ourselves to use System 2. We can catch our own intuitive mistakes and apply formal rules to correct them.
Conclusion: A More Likely Synthesis
Instead of "reasoning itself always has been a dream," a more complete hypothesis might be:
Reasoning is a real, distinct computational process, but it is implemented on top of a massive neural network.
Your hypothesis is correct that "massive neural working" is the foundation. It's the engine. But "controlled deduction" isn't a dream—it's a very real tool that this engine learned to use.
A computer's CPU is just "massive transistor working," but it can run a program that performs "controlled deduction" (like a calculator). The calculator program isn't a "dream" or "mimicry"; it's a real, formal process being executed by the underlying hardware.
Your brain is the "massive neural working" (hardware), and "reasoning" is a special kind of "software" you can choose to run on it. This is why the future of AI likely isn't just bigger LLMs, but Neurosymbolic AI—systems that combine a massive neural network (System 1) with a classic, symbolic reasoning engine (System 2).
