A Thought Experiment on Conscious AI: What If the Spark Doesn't Need to Be So Complicated?

This is a speculative proposal born from curiosity, not conviction. Inspired by Roger Penrose’s Orch-OR theory of consciousness — which links awareness to quantum gravity — I began asking a simpler question: what if the mechanics of collapse aren’t the critical ingredient?

What if we could replicate the functional role of non-computability using something more accessible — like radioactive decay?

The result is a conceptual architecture for an artificial mind that integrates real-world quantum randomness to resolve internal states — not through simulation, but through actual, physical uncertainty.

Whether this produces consciousness, or merely a better mimic, is an open question. But I think it’s worth asking.

Collapse Substitution: A Pragmatic Alternative to Orch-OR

1. Introduction

In thinking about how one might approach adding consciousness to existing AI, both the currently popular AI systems and potential derivatives we’ll explore later, I explored what we already know about consciousness in humans.

This led me to an interesting approach suggested by Roger Penrose and Stuart Hameroff, and their Orchestrated Objective Reduction (Orch OR) theory.

What followed then was thoughts as to how we might be able to add a source of non-computability, externally rather than internally to the AI models mentioned above. The idea is to do this to allow a more practical application of Orch-OR and perhaps consciousness.

2. The Role of Objective Reduction in Orch-OR

Orch-OR or in full Orchestrated Objective Reduction, is a theory arising from quantum processes in microtubules (tiny protein structures in the brain), and how the quantum state within those structures collapses and so orchestrated into coherent patterns.

Going back a stage to Gödel and his first incompleteness theorem (1931), where he states if any sufficient expressive formal system, there are statements that are true but not provable within that system. For example, “This sentence is false”.

Penrose then takes this further, suggesting a Turing machine, as a formal system can only follow rules and compute provable things. But as a human we can see the truth of certain unprovable Gödelian statements.

This then extends to consciousness, the human mind cannot be a formal system, consciousness must involve non-algorithmic processes. And this is where the Orch-OR theory comes in, explaining the non-algorithmic process that exists in the brain and how that causes consciousness.

Orch-OR suggests that microtubules are ubiquitous in the brain, and allow for quantum coherence that maintains superpositions briefly. Penrose then adds that the collapse of the quantum wavefunction is not caused by observation, but instead when a difference in spacetime curvature exceeds a threshold.

In this view consciousness is linked to fundamental spacetime geometry, and it’s not random: the collapse encodes proto-conscious events. Every event is like a spark, linking enough sparks with feedback loops and memory and you have consciousness.

3. Reframing the Requirement: Is Collapse Itself Special?

In the above we’ve followed how Penrose uses a non-algorithmic source built into the brain to trigger sparks, which are consciousness.

The question I therefore pose is this: is the collapse itself special? Is there some information within it, or something it does to the brain other than being a non-algorithmic spark?

If we accept it isn’t of itself special, that what matters is how, where, and within what context the spark occurs, not the precise physics behind it. If so, this opens the door to an alternative — one that might be applied to AI architectures.

4. Proposed Alternative: External Quantum Randomness

I’ll cut to the chase. When thinking about creating a consciousness in AI, it’s obvious we can’t implant microtubules and use quantum effects. Instead, we needed a non-algorithmic source, to generate sparks. My suggestion is that radioactive decay should be able to achieve that.

This gives us a source of true randomness that also we can roughly predict. In the Orch-OR theory 500ms is mentioned. Using the right material, we can achieve that (for a while), or longer/shorter as needed.

Just as in the Orch-OR model, we have a spark — a discrete non-algorithmic event — available to us. It just needs to be wired into the right feedback loops and memory.

For clarity I’m not stating that this gives us consciousness, merely that it might serve a functionally equivalent role.

5. Architectural Sketch: Synthetic Consciousness via Collapse Substitution

To test this theory that a non-algorithmic randomness can be functionally equivalent to Penrose’s objective reduction we can imagine an architecture as follows.

At its core, the system maintains a global workspace — a dynamic frame that integrates perceptual, symbolic, and internal state information. Multiple submodules (e.g., perceptual analyzers, language processors, decision heuristics) generate parallel candidate interpretations or actions. These assemblies represent mutually incompatible or competing internal states. E.g. For example: the system might ingest sensor data, apply heuristic analysis to identify potential issues, and route the resulting interpretations to a decision-making module — which may or may not escalate them to the user.

Into this we wire in our non-algorithmic source, not as a simple input, but instead acting as a trigger for resolution events — the system’s version of Penrose’s "moment of conscious selection.". When a signal is received, the system selects a competing assembly as resolved, commits that state to the conscious workspace, ie a moment of awareness, and archives the choice context to episodic memory.

These can then be fed back, perhaps altering weightings for future decisions, an ongoing self-model, or biases in attention and perception.

As such the system then develops an identity grounded not in its code but in its accumulated non-deterministically resolved experience.

This can be pushed further with reflective meta-cognition, which monitors how resolutions were made, queries why they may have been made, and uses that to construct a narrative self-model.

This doesn’t replicate Penrose’s OR mechanism directly, but satisfies the functional criteria, of non-computable resolution, internal integration, causal commitment and self-referential adaptation.

6. Is This Consciousness?

This entire model hinges on the assumption that non-computable events are sufficient for consciousness — an assumption still open to debate.

Many AI practitioners suggest that we just need to add more complexity and layers to achieve consciousness. They may be right, or this may also be required.

If this is the case however, with a simple external source (this could be very remote), we have a way of triggering sparks in an AI that then lends itself to a form of consciousness. I’m sure we all have ideas of what that may be, and whether it’d be truly useful. This paper focuses on how such a mechanism could be implemented, not whether it is necessarily desirable.

7. Potential Applications and Tests

The model suggested above might sound complex but comes from my thinking and work so far on an AGI. My thinking is that the current popular LLMs only play a small role in an AGI, serving as the input (questions) and output (interface) to the entity. To achieve AGI, however, we also need memory, perception, event processing, and other cognitive structures. And while this AGI may or may not need to have a consciousness, the above model is assumed to plug into this sort of mix of parts.

If we assume then that we can take an existing AGI type development, we then need to look at how, as per the model in Section 5, we might wire in a random source. For initial development there is no reason or need to make it truly random, but that can be dealt with later.

We can add in and modify as suggested in section 5 so that the random sparks influence and call up memory and current processing events.

8. Conclusion

In this discussion, I’ve layered speculation atop theory — a hypothesis built on an already controversial foundation. But that is often how useful ideas begin. This approach offers a plausible, experimentally approachable alternative to Orch-OR that requires no exotic biological machinery or untested physics — only architecture, feedback, and real randomness.

Whether it leads to conscious AI or merely an interesting mimic is an open question — but one I believe worth asking.

Appendix

Foundational Theories & Concepts

1. Gödel, K. (1931). Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I. Monatshefte für Mathematik und Physik, 38, 173–198.
   (Gödel’s original incompleteness theorem paper)

2. Penrose, R. (1989). The Emperor’s New Mind: Concerning Computers, Minds, and the Laws of Physics. Oxford University Press.

3. Penrose, R. (1994). Shadows of the Mind: A Search for the Missing Science of Consciousness. Oxford University Press.
   (Core Orch-OR theory and the non-computability argument)

4. Hameroff, S., & Penrose, R. (1996). Orchestrated reduction of quantum coherence in brain microtubules: A model for consciousness. Journal of Consciousness Studies, 3(1), 36–53.

5. Chalmers, D. (1995). Facing Up to the Problem of Consciousness. Journal of Consciousness Studies, 2(3), 200–219.
   (Introduces the “hard problem” of consciousness and qualia)
   
   

Quantum Randomness and Alternative Theories

6. Zeilinger, A. (1999). Experiment and the foundations of quantum physics. Reviews of Modern Physics, 71(2), S288.
   (Discusses unpredictability in quantum systems)

7. Rarity, J. G., & Tapster, P. R. (1994). Quantum random number generation and key sharing. Journal of Modern Optics, 41(12), 2435–2444.
   (Quantum randomness as a secure source of unpredictability)

8. Tegmark, M. (2000). Importance of quantum decoherence in brain processes. Physical Review E, 61(4), 4194–4206.
   (Critique of quantum consciousness, useful to contrast with Orch-OR)
   
   

Architectural & Cognitive Frameworks

9. Baars, B. J. (1988). A Cognitive Theory of Consciousness. Cambridge University Press.
   (Origin of the Global Workspace Theory)

10. Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: basic evidence and a workspace framework. Cognition, 79(1–2), 1–37.
    (Modern neural interpretation of Baars-style architectures)

11. Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11, 127–138.
    (An alternative model of brain function based on prediction and self-modeling)