Wait for Prompt
- quick capture before proper note
Consciousness as a running loop: how the argument developed
Starting point: Graziano’s attention schema theory
- Consciousness is a simplified model of our own attention
- Function: plan and control attention, filter out subconscious processing
- Evolutionarily plausible, likely emerged early
- Fits with global workspace ideas
Premise: consciousness is computable
- It could run on hardware other than a brain
“Wait for prompt” hypothesis
- An LLM needs input to do anything
- Proposal: maybe all consciousness is like this, and senses are just prompts
- Question: would a computed mind do anything without input?
Challenge: two meanings of “needs input”
- Trivial: any process needs something to process
- Strong: sits idle until poked, runs once, halts. This describes the LLM.
- The brain never halts (sensory deprivation, sleep), so the strong version doesn’t fit it
- Fire vs match: a fire is always burning, a match flares once per strike
Resolution: the self-referential feedback loop
- The brain supplies its own input: each output feeds back as the next input
- The stream of consciousness is that loop
- External input perturbs a loop that is already running
“While conscious” refinement
- The loop stops in dreamless sleep, anesthesia, and coma, even though the hardware is intact
- Claim: consciousness is the loop running, not a precondition for it
- Graziano supplies the content (what is modeled); the loop supplies the process (the running)
Link to Hofstadter’s I Am a Strange Loop
- His view: the self is self-reference that crosses levels (from Gödel). It’s about reference.
- Your view: the self is a continuously executing loop. It’s about execution.
- Yours handles anesthesia better: the structure persists, but execution stops
- Both views favor structure over substrate
What survives from “wait for prompt”
- Consciousness can be prompted: external input is injected into the running stream
- The loop also prompts itself (unbidden thoughts)
- An LLM only takes external prompts: no self-prompting, no persistence between prompts
Symmetry: potential vs actual
- An anesthetized brain and an LLM on disk are in the same state: machinery present, loop off
- Difference: the brain can restart and sustain its own loop; the LLM can’t
- Building that capability is an engineering gap, not a gap in principle
Conclusion
- Consciousness is a process a machine runs, not a substance it has
- The question shifts from “what is it made of?” to “is the loop running?”
Open threads
- Is continuous running essential, or just how evolution implemented it?
- What minimum loop structure is enough: does recurrence alone count, or does it need a self-model (Graziano)?
- Does a self-feeding LLM wrapper qualify? If not, what’s missing? Could it replicate the self model