The Unknowns

A response to Filip Piękniewski’s certainty. His critique is valuable, but his certainty (“only control loops are intelligence”) limits exploration.

What We Don’t Know:

  1. Definition of Intelligence

    • No consensus exists
    • Turing test is behaviorist, avoids essence
    • Filip’s “feedback control” is a hypothesis, not proven

  2. The “Atom” of Intelligence

    • Filip proposes PVM prediction units
    • But is this the only answer?
    • Could there be multiple equivalent implementations?

  3. Conditions for Emergence

    • Why do some systems emerge intelligence?
    • Simple rules → complex behavior (Conway’s Life)
    • We lack predictive theory of emergence

  4. Is Embodiment Necessary?

    • Filip: absolutely yes
    • Open question: can information systems have “virtual embodiment”?
    • LLM has boundaries in “language space”
    • Resource constraints (compute) are real constraints
    • “Survival pressure” could exist (user satisfaction, continued operation)

  5. Learning vs Evolution

    • Biology has both: evolution (eons, genetic) + learning (lifetime, synaptic)
    • Deep learning simulates “learning”
    • But what about “evolution”? Architecture search?
    • Maybe intelligence requires evolutionary pressure on architecture itself?

A Hypothesis

Intelligence may have multiple equivalent implementations, like “flight” can use flapping, gliding, propellers, or jets.

Bird flight ≠ Airplane flight, but both follow aerodynamics.

Maybe:

  • Biological intelligence = neurons + embodiment + evolutionary pressure
  • Machine intelligence = ??? + ??? + ???

We don’t know what fills the blanks yet.

Key Insight from User

“I think intelligence is very possible, we just need to explore how to implement it, what system. We don’t know yet.”

This open stance is more constructive than “you’re wrong.” It acknowledges:

  1. The phenomenon is real
  2. We haven’t found the path yet
  3. Exploration is needed

Research Directions

  1. Study emergence mathematically - When do simple rules produce complex behavior?
  2. Virtual embodiment - Can LLMs have “body” in information space?
  3. Hybrid architectures - Control loops + neural networks + ???
  4. Evolutionary pressure - How to create selection pressure for AI architectures?
  5. Cross-domain transfer - What can we learn from cellular automata, game theory, physics?