Although the establishment of the truth of information requires the scientific method (Harari 2024), bad information in biology can lead to species extinction, e.g., disregarding the coloration of a coral snake can cause death. Information as it applies to humans and other animals is a narration (a collection of thoughts/images/events, Harari 2024) that can be described in terms of bits of information for storage by the brain and bits per second for transmission by the brain (Miller 1956; Shannon 1948; Tehovnik and Chen 2015). This metric is used to standardize intra-species comparisons (Tehovnik 2014), since species’ brains are varied and governed by distinct genetic markers shaped by evolution, i.e., by a long and unique history of environmental perturbations (Darwin 1859; Dawkins 1977; Gallistel and King 2010; Noble and Noble 2023). The information metric is predicated on base 2 (largely for simplicity to have just 0’s and 1’s, Shannon 1948), and it is not that different from the genetic code, which is formed according to base 4, representing the four nucleotides of DNA (Gallistel and King 2010). The exponent of base 2 (the bits) indicates the length of the information string to code for one element. For example, the information string per character for printing is composed of 8 bits (to yield 256 possibilities), which is referred to as the ASCII code (American Standard Code for Information Interchange).
Based on the number neocortical synapses in humans (i.e., 164 trillion, Tang et al. 2001), the number of bits of information that can be stored is estimated to be 164 trillion bits or 2 ^164 possibilities (for synapse ON and synapse OFF, Tang et al. 2001; Huang 2008). Likewise, based on the cerebellar synapses (i.e., 280 trillion granular-Purkinje synapses, Huang 2008), the number of bits that can be stored is 280 trillion bits or 2 ^280 possibilities (Huang 2008). Thus, the neocortex and cerebellum have a comparable storage capacity in humans for the storage of conscious, declarative information in the neocortex, and for the storage of motor routines in the cerebellum (Corkin 2002; Huang et al. 2014; Oh-Descher et al. 2017; Sokolov et al. 2017; Tehovnik et al. 2021). As discussed, these two storage facilities are interconnected anatomically, and this is so because consciousness as mediated by the neocortex is a process that eventually is expressed behaviorally via the cerebellum, which concurs with the original thinking of James (1890). Furthermore, in humans the surface area of the neocortex and cerebellar cortex is comparable (i.e., the surface area of the neocortex is just 1.25 times greater than the surface area of the cerebellum), but the volume of the neocortex is 8 times greater than the volume of the cerebellum, because there is far more white matter interconnecting the neocortical neurons (Huang et al. 2014; Sereno et al. 2021). In fact, each of the 16 billion neurons of the neocortex supports 10,250 synapses, whereas each of the 69 billion neurons of the cerebellum supports 4,058 synapses (Herculano-Houzel 2009), with the neocortical axons traversing longer distances thereby filling up the volume of the neocortex.
To compute the upper limit of the information transfer rate of neurons in the neocortex and cerebellum it will be assumed that each action potential represents 1 bit of information (ON for action potential presence and OFF for action potential absence). In the neocortex there are 16 billion neurons with a maximal firing rate of 200 spikes per second (Herculano-Houzel 2009; Tehovnik 1996), and in the cerebellum there are 69 billion neurons with a maximal firing rate of 600 spikes per second (Herculano-Houzel 2009; Tehovnik et al. 2021). This yields an information transfer rate of 3.2 trillion bits per second [or 2 ^ (3.2 trillion) possibilities per second] for the neocortex (1 bit/sec x 16 billion neurons x 200 spikes/sec), and 41 trillion bits per second [or 2 ^ (41 trillion) possibilities per second] for the cerebellum (1 bit/sec x 69 billion neurons x 600 spikes/sec). Thus, there is a ten-fold increase in transfer rate going from the neocortex to the cerebellum, a structure that must be precise about spike timing to control the muscles of the body (Thach et al. 1992). For example, the elephant has the largest known cerebellum of all mammals containing 260 billion neurons (Herculano-Houzel et al. 2014), four times as many neurons as found in the cerebellum of humans. This augmentation may be required to subserve the 40,000 skeletal muscles found in the trunk of this animal which has the dexterity to pick up a leaf or a large log (Longren, Brecht et al. 2023)[1]. Humans by comparison have up to 700 skeletal muscles (Tortora and Grabowski 1996).
Zheng and Meister (2024) point out that the throughput for most behaviors in humans averages about 10 bits per second (or a little over 1,000 possibilities per second), which is comparable to the information transfer rate of a cochlear implant (Baranauskas 2014). This average value is based on scores as low as 4 bits per second for Morse code to 40 bits per second for playing a musical instrument or giving a speech (see Figure 30, Tehovnik and Chen 2015). Even though the central nervous system of humans is designed to transfer up to a trillion bits per second, the 10 bits per second limit noted by Zheng and Meister is due to the bottle neck produced by the motor system (McFarland and Sibley 1975). Of course, as discussed, many behaviors are put on automatic pilot—such as running the autonomic nervous system and performing highly automated motor routines once learned—e.g., the vestibuloocular reflex, maintaining postural shifts of the body for walking, brushing one’s teeth, responding to basic questions in one’s language, and so on—but collectively these do not require trillions of bits per second to execute automatically. Finally, the primary behavior being executed volitionally is the one that occupies the throughput of the motor system, such that when a polylingual person speaks in English (e.g., at a 40 bits per second rate), the other languages are suppressed but remain unaltered in the brain (Ojemann 1991). Thus, the high transfer rate of a trillion bits per second is mainly utilized for the storage and intra-CNS transfer of information to generate fictive responses rather than for the execution of behavior.
What then distinguishes Einstein, Bolt, Pelé, and Kasparov from the general population? It is their ability to store and analyze vast amounts of information dedicated to one goal: coming up with the laws of physics, being the fastest man on the planet, being the best footballer in the world, and being the top grandmaster in chess. The supercomputer, Deep Blue, only started beating Kasparov after it learned and stored more chess moves than Kasparov could learn in a lifetime (Higgins 2017). Deep Blue’s success took ten years of programming and training by a team from Carnegie Mellon University. Once trained, Deep Blue could transfer 28 bits per second[2], as it made chess moves against Kasparov (Figure 30).[3] The mental projections of many moves ahead by a grandmaster depend on a spatial mapping (Coates 2013; Dotson and Yartsev 2021; Hassabis et al. 2007b; Johnson and Redish 2007; Kay et al. 2020; O’Keefe and Speakman 1987) that has access to information acquired over many years of competition. Chess requires the same intellectual savvy as an athlete projecting the body or ball fictively to the finish line or goal, or a scientist coming up with a novel theory (Coates 2013; Pasolini 1971; Wood 2017).
To enhance the efficiency of information storage by the brain, the information can be stored in chunks (Miller 1956). This occurs once a behavior has become automated. Memorizing a number code to open a combination lock does not require an explicit recall of the number, but just the motions of the fingers to open the lock; hence the information has been chunked by evoking finger motions once the lock is viewed. Similarly, a large body of work done by Einstein can be ‘chunked’ or reduced to E = mc^2. This abstraction captures a major notion developed by Einstein, but the supportive details of the expression are buried in scientific notebooks and physics textbooks. Note that if the information condensed through chunking needs to be retrieved, one might need to consult the physical records, which is commonly done by scientists (Clark 1998).
Now, let us consider the information consolidated by the brain as one learns a new task, which is a slow process requiring great effort (Ziyang, Sheth et al. 2017). A group of Japanese university students, who were moderately bilingual, were enrolled in a 4-month intensive language course to improve their English (Hosoda et al. 2013). During this period, they learned 1000 new English words, which they used in various spoken and written contexts. The learning was followed by a weekly test. To learn the 1000 words, it is estimated that 0.0006 bits per second of information were transferred over the 4-month period [(1.5 bits per letter x 4 letters/word x 1000 words)/16 weeks, bit-rate per letter corrected for redundancy, Reed and Durlach (1998)], a rate that—not surprisingly—falls well short of the 40 bits per second transmitted by a competent communicator of English (Reed and Durlach 1998); hence learning takes longer than the execution of a learned act.[4] Additionally, it was found that the pathway between Broca’s area and Wernicka’s area was enhanced in the students as evidenced by diffusion tensor imaging (Hosoda et al. 2013). Such enhancement during learning has been attributed to increased myelination and synaptogenesis (Blumenfeld-Katzir et al. 2011; Kalil et al. 2014; Kitamura et al. 2017).[5] [6]In short, the consolidation process is time consuming and effortful, so those who believe that by hooking up two brains will be sufficient to transfer learned information (in a flash) have no understanding of how the brain works (Figure 30, neurons-to-neurons, Tehovnik and Teixera e Silva 2014).
How a string of neurons through the neocortex mediates the conscious process by activation of a succession of declarative conscious units is discussed in a previous chapter (summarized in Figure 23). The neocortex is composed of neurons that both trigger behavior (via the glutamatergic pyramidal neurons) and inhibit behavior (Schiller and Tehovnik 2001, 2003, 2005; Tehovnik and Slocum 2013) by way of GABAergic circuits (Krnjević 1974; Krnjević et al. 1966abc; Krnjević and Schwartz 1967; Logothetis et al. 2010). The neurotransmitter GABA mediates a multitude of transitions in behavior at the neocortical level, including locomotion, body orientation, immobility, and so on (Poulet et al. 2019), and to this list should be added sensory events and consciousness (Marvan et al. 2021). The excitability of the neocortex is regulated by a balance between excitation and inhibition to prevent against extremes: an extreme decrease in activity that induces unconsciousness (as with anesthesia, Attwell and Laughlin 2001; Hebb 1968; Richards 2002) and an extreme increase in activity that, again, induces unconsciousness (as with epilepsy, Penfield and Jasper 1954).[7] [8]
GABA was discovered by studying the effect of delivering an electric pulse to the neocortex of a mammal, which causes a brief burst of activity by the stimulated pyramidal neurons followed by a period of inhibition of those neurons. This inhibition was found to be mediated by GABA (Krnjević 1974; Krnjević et al. 1966abc; Krnjević and Schwartz 1967). This discovery established that the entire nervous system is under both excitatory and inhibitory control (Schiller and Tehovnik 2015), thereby keeping all the neurons within a restricted range of excitability to keep all the neocortical channels open to the ‘conscious’ flow of information and activation by all the senses: exteroceptive, proprioceptive, and interoceptive.
Fu (2023) has proposed that the visual system of invertebrates and vertebrates utilize ON/OFF neurons to register the path of object motion through retinal space. We would suggest that a similar ON/OFF principle applies to the neocortex for tracking strings of language (or any other form of consciousness) by leveraging the excitatory and inhibitory circuits of the neocortex. When a pyramidal fibre (which stores the conscious declarative information) is discharged in the neocortex, a collateral fibre activate a GABAergic interneuron to silence the pyramidal fibre (Krnjević 1974; Krnjević et al. 1966abc; Krnjević and Schwartz 1967). This ON/OFF property allows the stored elements of language (or any other behavior) to be concatenated, much like the trajectory of an object is concatenated by the visual system. Thus, the stream of information (or consciousness) at the level of the neocortex is maintained by the ON/OFF properties of the pyramidal neurons.
Information for the purpose of consolidation is maximal when sensory feedback is high (Figure 31). In fact, the brain becomes inoperative without feedback (Tehovnik and Chen 2015). In both invertebrates and vertebrates, the contraction of muscles from body position 1 to body position 2 requires a differential innervation of signals to maintain equilibrium at a particular collection of muscle lengths or tensions and joint angles (Feldman 1974; Giszter, Mussa-Ivaldi, Bizzi 1991). Once a new position is assumed, a re-afference signal is sent to the brain so that any future shift in body posture can be made in an intelligent manner based on sensory contingencies; this is especially important if position 2 is suboptimal, namely, registering an error according to an animal’s efference-copy plan, a plan that is stored in the cerebellum of vertebrates. Invertebrates must also have a collection of neurons that contain the efference-copy plan to optimize behavior when there is an environmental perturbation. It is well-known that in the absence of feedback, consciousness is extinguished as evidenced by the study of ALS (amyotrophic lateral sclerosis) patients (Birbaumer 2006).
Summary
1. The neocortex and the cerebellum of humans have the capacity to store large amounts of information (surpassing 10^14 bits) pertaining to conscious declarative information and motor routines, respectively.
2. The amount of information transferred within the neocortex and cerebellum of humans can surpass 10^12 bits per second.
3. The information throughput of the brain rarely surpasses 40 bits per second because the motor system establishes a bottleneck; the transfer of information when learning a new task is very low at values of 0.0006 bits per second when learning new words, for example.
4. The high information transfer rate via the brain is utilized for the storage of information and for predicting different outcomes to optimize future behavioral responses, such as performing athletically, playing chess, or creating something novel in science or art.
5. Once a behavior is automated, the information pertaining to it is chunked (or abstracted) to enhance efficiency. But if the information condensed through chunking needs to be retrieved, one might need to consult the physical record.
6. It is proposed that the stream of information (or consciousness) at the level of the neocortex is maintained by ON/OFF properties of the pyramidal neurons.
7. When all sensory feedback is abolished, the amount of information transmitted by the brain is reduced to zero. Without sensory feedback there can be no consciousness.
Footnotes:
[1] Like the elephant for mammals, the electric fish for fish species has an augmented cerebellum which is utilized for electro-communications (Fukutomi and Carlson 2020). The elephant also has an advanced communication system (Herbst et al. 2012). Just how the cerebellum is utilized in this process is unclear.
[2] The supercomputer Deep Blue—which was specifically designed to play chess, and which defeated Garry Kasparov—could search 200 million positions per second (Wikipedia/Computer Chess/Aug. 15, 2017) or ~ 28 bits per second [bits = 3.3219 x Log10 (2 x 10^8)].
[3] The supercomputer Deep Blue utilized a totally different method of computation when evaluating the chess board as compared to a human player. It was shown by Mihai Neghina in 2009 that certain moves against a computer will defeat the computer, since computers use brute force to assess all possible moves across an entire board, whereas humans can focus on key sections of a board to optimize performance (Coates 2013).
[4] Individuals learning a language from birth (between birth and the age of 18) transfer 0.0006 bits per second to the brain acquiring new words at a completion rate of nine per day (i.e., {1.5 bits per letter x 4 letters per word x 60,000 words}/18 years, Bloom and Markson 1998 Bloom and Markson 1998; Miller 1996; Reed and Durlach 1998). To achieve this rate, children must learn to read books to build up their vocabulary to 60,000 words or so by adulthood (Bloom and Markson 1998).
[5] The minimal circuit for language learning is composed of pathways between Wernicke’s and Broca’s areas of the human brain, based on lesion, stimulation, and neural recording experiments (Ibayashi et al. 2018; Kimura 1993; Ojemann 1991; Metzger et al. 2023; Penfield and Roberts 1966).
[6] The increase in myelination after learning (Blumenfeld-Katzir 2011; Hosoda et al. 2013) may augment the reliability of the information transmitted by reducing the conduction failures (Raymond and Lettvin 1978; Swadlow et al. 1980), thereby enhancing the bits per second transferred per axon activated (Tehovnik and Chen 2015) when developing declarative, conscious units. Yet some believe that fundamental changes during learning, such as of language, occur at the subcellular level beyond the limitations set by synaptic time and axonal conduction velocity (Chomsky 2019—lecture delivered at MIT in April; also see Kandel 2006).
[7] It is noteworthy that the GABA agonist, muscimol, and the GABA antagonist, bicuculline, both abolish visual perception when injected into area V1 of primates (Schiller and Tehovnik 2003).
[8] If one anesthetizes just the neocortical grey matter, which would disable the synapses via GABA activation (Attwell and Laughlin 2001; Richards 2002) and abolishes consciousness (Hebb 1968), electrical stimulation of the pyramidal fibres exiting the neocortex should still evoke action potentials since the initial segments are not affected by the anesthesia. Indeed, changes to the synapses of a neuron by GABA agents have minimal effects on the activation of an axon (Noda et al. 1988), and hence the grey matter of the neocortex could be studied separately from the white matter using GABA agents to evaluate the effect of axonal loops on consciousness.
Figure 30. Bits of information transmitted per second by various behavioral systems ranging from brain-to-brain communication to the generation of speech. The inset is included to translate the bit rate to the number of possibilities per second. Figure derived from Tehovnik and Chen (2015) and updated with new information.
Figure 31. Information transfer for consolidation is maximal when the sensory feedback is high. This feedback can arise from inside or outside of the body as depicted by the inset. ALS (amyotrophic lateral sclerosis) patients lose their consciousness once sensory feedback becomes zero (Birbaumer 2006).