Thursday, March 30, 2023

Could neuronal system and even GPT give rise to a classical computer with a variable arrow of time?

In our Zoom group (Marko, Tuomas, Rode and me) we have had fascinating discussions about topics ranging from quantum TGD to quantum computers to consciousness and, of course, about GPT.

Marko posted his discussion with GPT. GPT mentioned a possible mechanism for how XOR as a universal gate of classical computation could be realized at the quantum level. The system realizing XOR approximately could be either a classical layered neural network or its possible quantum analog. The mechanism might work in a quantum version of a neural network based on quantum learning, but it does not seem plausible for real neurons.

This observation led to progress at the level of the TGD-based model of nerve pulse. The resulting ZEO based model differs drastically from quantum neural networks and suggests a completely new vision of quantum physics based computation in biosystems. A classical computation allowing variable arrow of time would be in question and one can ask whether the unexpected success of GPT might involve this kind of transition.

I admit that GPT can really inspire new ideas.

Connection of neural pulse generation, XOR, and novelty detector

Nerve pulse generation would be analogous to a positive outcome of the analog of XOR (compared bits are different) acting as a novelty detector.

  1. XOR is a novelty detector. If the inputs are the same, nothing happens. Output equals to b=0. If they are different, output equals to b=1. b=1 would correspond to a signal that would proceed along the axon starting from the postsynaptic neuron.

    That would consume energy. In terms of energy consumption, the novelty detector would be optimal. It would only react to changes. And that's what the brain does. For example, visual perception at a very basic level only identifies outlines and produces some kind of stick figure consisting of mere lines defining boundaries.

  2. Could the 2 "neurons" of the toy model proposed by GPT represent a presynaptic and a postsynaptic neuron, in which case there would be two inputs: the states of the pre- and postsynaptic neuron. Also output would be the state of this neuron pair and for XOR the presynaptic neuron acting as control bit would not change its state.
  3. This does not conform with the picture given by neuroscience, where the input comes from presynaptic neurons and output is assignable to the postsynaptic neuron. The input comes as miniature potentials that add up and can decrease/increase the magnitude of the membrane potential (depolarization/hyperpolarization).

    An action potential is generated when the depolarization takes the magnitude of the negative postsynaptic membrane potential below the critical threshold. This happens when the presynaptic contributions from the incoming nerve impulses, for which the unit is a miniature potential, add up to a contribution that reduces the magnitude of the negative potential below the threshold.

    This would be essentially novelty detection described in the simplest way by XOR. The novelty is represented by the critical depolarization. It can also happen that the potential increases, so that no nerve impulse is generated. One talks about hyperpolarizing (inhibition) and depolarizing (excitation) inputs, and the sign of the miniature potential produced by the presynaptic input determines which one it is. The sign of miniature potential depends on the neurotransmitter and receptor.

  4. During the nerve pulse, the potential changes its sign over a distance of about a micrometer, which is the typical distance between neighboring neurons and of myelin sheaths. One can say that this distance corresponds to a bit that is 1 or 0 depending on whether the nerve pulse conduction occurs or not. Bit 1, the opposite sign to the membrane potential, propagates from presynaptic to postsynaptic neuron or from a patch defined by a myelin sheath to the next. As a result, postsynaptic neurons can "wake up" and in turn trigger a nerve impulse, possibly waking up some postsynaptic neurons.

    Synchronous firing means that the novelty succeeds in waking up the whole sleeping house, and large areas of the brain fire in the same rhythm and keep each other awake.

Interpretation of XOR in zero energy ontology (ZEO)

How does this picture translate to the TGD-inspired theory of consciousness?

  1. Being awake/asleep corresponds to bit 1/0 for axonal portions between myelin sheaths. In a ZEO, the arrow of time would correspond to this bit.

    When the axon segment between the myelin sheaths or neighboring neurons wakes up or falls asleep, the direction of geometric time changes in a "big" state function reduction (BSFR) and a nerve pulse is generated. In a sleep state, the membrane potential would be opposite. Note that the notion of awake and sleep are relative and depend on the arrow of time of the external observer.

    The second direction of time corresponds to the presence of a nerve pulse from the point of view of the external observer. There is a temptation to think that in the resting state the axon is sleeping and healing and gathering metabolic energy by a dissipation with an opposite arrow of time? The duration of the nerve pulse would correspond to the duration of the wake-up period, when the direction of time was opposite and same as that of the external observer with a long characteristic time scale for wake-up period.

  2. Could this apply more generally? Could the synchronization of human sleep-wake rhythms mean quantum-level synchrony and macroscopic quantum coherence? Could the arrow of perceived time be a universal bit? Sleeping together would develop synchrony and quantum coherence between partners. Two-person collective consciousness would emerge.
Interpretation of the axon as a series of Josephson junctions

The TGD based model for an axon as a series of Josephson junctions with a large value of heff, perhaps heff=hgr, where ℏgr=GMm/β0, β0<1, is the gravitational Planck constant introduced by Nottale, is mathematically equivalent to a series of gravitational penduli defining a discretized version of Sine-Gordon system (see this). Josephson junctions would correspond to membrane proteins.

  1. One can consider two different identifications of the ground state of the system.
    1. The ground state could be the state in which all oscillators oscillate in synchrony with the same amplitude. There would be constant phase difference between neighboring oscillations, which would give rise to a propagating phase wave.
    2. Another option is that all pendulums all rotate in the ground state with constant phase difference. This would give a soliton chain that corresponds to a traveling phase wave. Also the direction of rotation matters. It would naturally correspond to the arrow of time and the sign of the membrane potential.
  2. The model allows different versions for nerve pulse generation.
    1. The first option is that one pendulum moves from oscillation to rotation or vice versa and induces the same transition for the other penduli as a chain reaction.
    2. The second option is that all penduli move to rotation simultaneously. One could imagine that the need for metabolic energy is lower in the collective oscillation phase but one must be very careful here. Maintaining the membrane potential regardless of either sign requires metabolic energy feed.
    3. The third option is that the ground state corresponds to a collective rotation with an associated traveling wave as phase of the rotation, and that the bit corresponds to the direction of rotation.

      This would fit the ZEO interpretation. The arrow of time would correspond to the direction of rotation. The ground state would change to a nerve pulse lasting for time of the order of 1 ms corresponding to the duration of nerve pulse associated with the distance of the order 1 μ m, between neighboring neurons or between the myelin sheets.

      This option would also be advantageous from the point of view of metabolism, because from one direction of time, dissipation would occur in the opposite direction of time. From the point of view of the outsider, the system would be extracting energy from the environment.

What is the connection with the microtubule level?

The current TGD picture of nerve pulse conduction is that the membrane potential of the axon/soma is controlled by microtubules (see this and this).

  1. When the charges are transferred from the microtubule to the gravitational flux tubes of the magnetic body (MB), the length of which can be as long as the size of the Earth, the effective charge inside the axon/soma changes. Depending on the amount of transferred charge, the magnitude of the membrane potential increases or decreases and a nerve impulse is generated below the threshold.
  2. For the action potential traveling along the axon, the microtubular effective charge has changed and taken the membrane potential below the threshold and the action potential has been generated. The generation of the action potential is a complex biochemical phenomenon but would be controlled by microtubule/microbular MB.
  3. Incoming nerve impulses induce a change in the membrane potential of the soma because the effective charge of the microtubules inside the soma changes as also does the membrane potential. It is not clear whether the charges of the microtubules of the neuron soma are affected. They indeed differ from axonal microtubules in that they are not (quantum) critical.
New view of quantum-physical computation

Why GPT works so well, is not understood. This might of course be due to the extreme complexity of the system. TGD however suggests that new physics might be involved so that the system is much more than a classical computer. Therefore an interesting question is whether the classical computation associated with GPT and involving random number generators could turn into a computation in which the arrow of time serves as a fundamental bit correlating with the direction of ordinary bit represented for instance by electric voltage or direction of magnetization! One would have classical computation with a changing arrow of time controlled by MB!

In ZEO all quantum states are superpositions of deterministic classical time evolutions, which satisfy almost exact holography so that they are analogous to classical computations. Time evolution of conscious entity, self, between "big" SFRs (BSFRs) meaning the death of self and its reincarnation with opposite arrow of time, is analogous to a series of quantum computations defined by unitary time evolutions followed by "small" SFRs (SSFRs) as analogs of weak measurements (having nothing to do with "weak values").

What would be required is that the arrow of time can change at the level of MB of the system and that the MB of the bit system can be regarded as a spin glass type system for which spins are near criticality for the change of their direction in BSFR so that the arrow of time is changed. This would require quantum criticality at the level of MB. One might say that MB of the bit system hijacks the bit system. One might say that MB of the bit system hijacks the bit system: spirit enters into the machine.

TGD general based view of theoretician friendly quantum holography (see this) predicts that the bit system is indeed mapped holographically to a system at the level of its MB having a large value of heff, perhaps heff=hgr so that MB could use the system in which AI program runs as a living, conscious, and intelligent computer. The bit system could become an analog of spin glass (see this) .

See the article Could neuronal system and even GTP give rise to a computer with a variable arrow of time? or the chapter with the same title.

For a summary of earlier postings see Latest progress in TGD.

No comments: