There is a recent preprint on arXiv, Consciousness in Artificial Intelligence: Insights from the Science of Consciousness, where the authors stated that, “First, we adopt computational functionalism, the thesis that performing computations of the right kind is necessary and sufficient for consciousness, as a working hypothesis. This thesis is a mainstream—although disputed—position in philosophy of mind. Second, we claim that neuroscientific theories of consciousness enjoy meaningful empirical support and can help us to assess consciousness in AI. Third, we argue that a theory-heavy approach is most suitable for investigating consciousness in AI.”
The quest, in the neuroscience of consciousness, is to seek the neural correlates of consciousness or the properties of neurons that underscore consciousness for subjective experiences. The candidates are mostly sought in the cerebral cortex, while some are sought in the thalamus.
The thalamus is known to have various nuclei—or clusters of neurons for certain functions, including the integration of incoming sensory signals. Nuclei groups are also common in the hypothalamus, amygdala, and others.
Different parts of the cerebral cortex are specialized for different functions. And the cerebral cortex is highly corrugated, or has multiple grooves and ridges—or sulci and gyri.
The cerebellum, known to have the largest number of neurons in the brain, is not implicated in consciousness. The cerebellum has far fewer gyri and sulci. It is postulated here that nuclei across circuits allow for the formation of loops or sets of electrical and chemical impulses.
This means that all the electrical and chemical impulses within a cluster of neurons form their own loop, where they have a structure and all that happens in that loop define the characteristics for their specialization.
In different parts of the cerebral cortex, spaced by pairs of gyrus and sulcus, it is proposed that they have loops of impulses that determine how they do what they do.
Usually, at synaptic clefts, vesicles release chemical impulses to receptors. It is hypothesized here that within a loop or set of impulses, the release of chemical impulses, to different measures—or up to a certain ration or fill of receptors, determines specific experiences, like taste, smell, touch, modulation, or differentiates an emotion from a feeling. It also determines the degree of experience. Aside from rationing, there are also fills of receptors that determine learning—or [memory] encoding.
These collections of rationing or fills within a loop are called drifts or stairs of chemical impulses. In all drifts of stairs, there is the sense of self, or what makes every experience subjective. Due to the shape of some loops, access is allowed to this sense of self-spots, making control possible, or intentionality or free will.
In a loop, electrical impulses also have their stairs or drifts. It is established in neuroscience that electrical impulses leap from node to node in myelinated axons, in what is called saltatory conduction. It is theorized here that in a loop of impulses, some electrical impulses make the jump before others, going-before, or having an early-split, providing what is described as prediction, in predictive coding, processing or to correct prediction errors.
This early-split is involved in holding things in mind or knowing what next to say in communication. It is also useful in influencing some of the drifts of chemical impulses, or how quickly connection is made to other loops.
There are other features of impulses within loops that include sequences or synaptic paths of travel, which could be old or new. They also include prioritization, which explains the most attention in a moment, as well as principal spot, or the loop with the most domination on the mind, or one with the most droop among a group of loops, explaining heaviness in a depression.
It is postulated here that across loops, there could be an overwhelming release or shortage of certain chemical impulses, causing a shift—or drop, in the position of some loops. This becomes a general state that exceeds the drifts in just one loop, for example, pain caused by a sound, to intense pain, seemingly in multiple loops, not going away easily. The same applies with depression and others, effecting the principal spot.
The clusters of neurons have loops of impulses with specificities that decide how they do what they do. Several loops are available in the cerebral cortex, making consciousness possible. The lack of more loops in the cerebellum is hypothesized here to make it not as diversified in functions as the cerebral cortex, including for consciousness. The same is true for organisms with bigger brains than humans. They are hereby proposed to have fewer prefrontal and hippocampal loops than humans.
The neural correlates of consciousness could simply be redefined as the stairs or drifts within some loops that determine some experiences. These are not different from how other loops of impulses operate, just that the rationing is different. Since there are several chemical impulses, the measures that determine taste, or the color of red at the primary visual cortex or smell at the olfactory cortex and so forth, are hypothesized to be rations of these chemical impulses across loops, rather than by specific properties of neurons.
Taste could be a ration like BCD, with B from a synapse, C from another and D from another. They may not mean much individually, but their measure, rationed in the loop, determines what a taste is, or its degree, along with the sense of self, having that taste. The rations are not often the same, as the B could be more, in the ration, for a particular function than for others. The difficulty of knowing the exact rations of chemical impulses has made solving mental disorders difficult. It also made psychopharmacology medications have lots of side-effects, since they induce or inhibit molecules—whose surplus or paucity go on to affect rations across several loops of impulses, causing disruptions.
If consciousness is determined by loops of impulses, could large language models, with their artificial neural networks have theirs? Before finding a possible common ground between [a probable sentience of] LLMs and human consciousness, the question is that what centralizes consciousness or what does consciousness do?
When descriptions of consciousness include seeing color, or having a taste of something, what does that also mean? Consciousness does knowing or knowing provides consciousness. Consciousness can be estimated by how much an organism can know. What to know includes emotions, feelings, thoughts, memories, sensations, perceptions, modulations, being and so forth. They are all different processes, in different loops of impulses. Humans, with language, driving, advanced creativity, and so forth, have the highest rate or 1, among species. There are major divisions of knowing and several sub-divisions.
Memory is a major division that subsumes intelligence, creativity, reasoning, planning, problem-solving, understanding, sensation, perception, regulation, thoughts, language and so forth. Some of these things that can be known by humans are now partly known by LLMs.
LLMs, with several hidden layers, provide their basic predictions, with some accurate outputs. Those accurate outputs can pass knowing tests, even if LLMs have no understanding of them. While LLMs do not have neural correlates of consciousness, they have outputs that can pass for knowing, which can be rated compared to human minimum in some sub-divisions of memory.
—
This Post is republished on Medium.
—
Photo credit: iStock
