Why Language Models Lack Conscious Meaning (and What Brains Do Differently)

Link

Abstract and introduction

1. Extents and ways in which AI has been inspired by understanding the brain

   1.1 Computational Models

   1.2 Artificial Neural Networks

2. Embodiment of Conscious Processing: Hierarchy and Parallelism of Nested Levels of Organization

3. Evolution: From the architecture of the brain to culture

   3.1 genetic basis and epigenetic development of the brain

   3.2 AI and Evolution: Consequences for artificial awareness

4. Spontaneous activity and creativity

5. Conscious vs non-conscious processing in the brain, or res cogitans vs res extensa

6. Ai awareness and interaction with social challenges rational thinking and language

Conclusions, recognition, and reference

5. Consciousness compared to the subconscious processing of the brain, or res cogitans vs res extensa

The voluntary activity specified above is of importance for the difference between consciousness and subconscious processing, which the Descartes have suggested by forming res cogitans compared to Res Extensa, two beliefs outlined that are not blurred outside of a dualist context as indicated by the use of the same term “res” (changeux & riceeur, 1998). In contemporary terms, this means that the “physical” awareness – or physiologically – is different from the subconscious. To distinguish between the two a masking task was introduced. When a representation emerged – for example, on the visual pathway – from V1 to the temporal cortex, a subconscious representation was obtained up to a sudden difference – the activity occurred around 200 to 300 ms after the start of the stimulus. When the stimulus becomes conscious, there is a strong progress and propagation of additional activity, especially when it reaches the prefrontal and parietal cortex (becomes «unusual»). This non-linear variation is defined as “ignition” (Dehaene & Changeux, 2005) and occurs regardless of the stimulus modality or paradigm used to manipulate consciousness (Del Cul, Dehane, Reyes, Bravo, & Slachevsky, 2009; Klatzmann et al. 2020a). The spontaneous fire can also be recorded in the absence of external stimulation (Kockuli, Rooy, Changeux, & Maskos, 2016; Moutard, Dehane, & Malach, 2015).

The first simulations of ignition in a global workspace architecture (Dehaene & Changeux, 2005; Dehane, Sergent, & Changeux, 2003) have included a major role for long-term repeated and top-down connection to maintaining a long-term ignition: more stable activity appears when the sensory stimuli is consciously noticed (Schurger, Kim, & & Cohen, 2015). Recently (Klatzmann et al., 2023), the hypothesis made that the rapid start and offset of excitatory currents combined with AMPA receptors could help sensory areas that quickly reacted to changes in external stimulation (self, Kooijmans, Supèr, Lamme, & Roelfema, 2012; Yang et al. Maintaining a specific stimulation of the prefrontal cortex and GNW (Wang, 1999; Wang et al., 2013).

When conscious, “representations” show single properties. For example, the associated properties that allow them to adjust to the chains and develop forms of reasoning that do not exist accidentally in humans, in monkeys and perhaps not on computers. The classical interpretation is that working memory is very shallow in monkeys (1-2 accurate 2.1 in chimps)) but higher in humans (7 +/- 2). More important in our point of view is that the awareness of representation possesses a single neural organization to allow them to establish “physical links” in an argument -to have a global meaning that includes each component. On this basis (more than not yet explained) one must build a sentence, suggest a reasoning, account for language formation (i.e., the ability to make sentences that have not been said, and to understand the unheard of sentences), and motivate as the ability to place a substance within another element of the same type and develop a tree, an element of the linguistic or linguistic element Repeatedly in a follow -up. In all of these instances, again, cortical and non-cortical neurons are involved by a large scale selection through reward systems.

LLMs have recently claimed that there are these capabilities, in particular the syntactic language development. It depends on how the generation is conceived and understanding the language. LLMs can come up with new combinations of words based on statistical patterns, and also identify connections between those words. This type of labor and understanding of language is syntactic and independent of semantics (that is, meaning characteristic perhaps by selective rewards), which identifies the labor and understanding of human language (Bennett, 2023; Marcus & Davis, 2019). Although LLMs have the ability to fully emulate a semantically coherent human language, they do so through a different approach than human intellect, which does not include the meaning of understanding the common sense (Wolfram, 2023).

The same can be true for reasoning and creativity. While the details of these features in the human brain should still clarify, the data relates to the indicative of a key role of intrinsic and external rewards in selecting and stabilizing neuronal adjustments underlying creative activities such as artistic (changeux, 2019). Also, the principle of decay (i.e., many neuronal combinations can result in the same functioning results) are the basis for plasticity and creativity of the human brain, and appear to be significantly limited to AI, despite the associated results that emerge from neural network pruning and neuromorphic hardware.

In conclusion, the properties of the brain that will help contemporary attempts to produce conscious AI systems are: a clear difference between consciousness and subconscious representations (demarkable, for example, by the phenomenon of ignition), and the consequences that are single associated properties of conscious representatives; The semantic ability of the brain, expressed, for example, in the capacity for symbolic understanding and assistance to one another; Singular characteristics of the organization of awareness of representation in significant references and eventually the development of reasoning and rational thinking.