Integrating information in the brain's EM field: the cemi field theory of consciousness

A key aspect of consciousness is that it represents bound or integrated information, prompting an increasing conviction that the physical substrate of consciousness must be capable of encoding integrated information in the brain. However, as Ralph Landauer insisted, ‘information is physical’ so integrated information must be physically integrated. I argue here that nearly all examples of so-called ‘integrated information’, including neuronal information processing and conventional computing, are only temporally integrated in the sense that outputs are correlated with multiple inputs: the information integration is implemented in time, rather than space, and thereby cannot correspond to physically integrated information. I point out that only energy fields are capable of integrating information in space. I describe the conscious electromagnetic information (cemi) field theory which has proposed that consciousness is physically integrated, and causally active, information encoded in the brain’s global electromagnetic (EM) field. I here extend the theory to argue that consciousness implements algorithms in space, rather than time, within the brain’s EM field. I describe how the cemi field theory accounts for most observed features of consciousness and describe recent experimental support for the theory. I also describe several untested predictions of the theory and discuss its implications for the design of artificial consciousness. The cemi field theory proposes a scientific dualism that is rooted in the difference between matter and energy, rather than matter and spirit.

‘Love is a smoke made with the fume of sighs’William Shakespeare, Romeo and Juliet‘What’s the best way to fix a bicycle that has a rope caught in its spokes?’Gary Marcus ‘Deep Learning: A Critical Appraisal’ (Marcus 2018)

The ‘binding problem’ is that of understanding ‘our capacity to integrate information across time, space, attributes, and ideas’ (Treisman 1999) within a conscious mind. The problem is often posed in terms of understanding how the disparate components of a visual scene—colours, textures, lines, motion, etc.—that are processed in distinct regions of the brain are yet brought together to form a unified conscious percept. However, binding is a general feature of consciousness in all its modes. The first text quotation above contains four discordant nouns, one denoting an emotion, the second a dark vapour, the third a noxious smell and the fourth, an utterance. Yet, Shakespeare’s genius bound each word into a single line of poetry that effortlessly evokes, in the conscious mind of the reader or listener, a singular, integrated, yet complex insight into the most tender of human emotions. In the second quotation, artificial intelligence (AI) researcher and pioneer of Deep Learning, Gary Marcus, laments the fact that AI currently lacks this ability, as illustrated by the intractability of problems, such as untangling a rope from the wheel of a bicycle that is, nevertheless, grasped and solved by any infant on her first exposure to the task.

Our subjective experience is that this kind of problem, which involves planning and executing several sequential steps, is nonetheless instantly grasped and solved in its entirety, as integrated information. This intuition is borne out by many studies that demonstrate that the binding provided by consciousness is indeed required to solve general intelligence problems, particularly sequential tasks that require working memory, such as memory trace conditioning (Carter et al. 2006), multi-step calculations (Dehaene and Cohen 2007), goal-directed behaviour and strategic planning (Dehaene and Naccache 2001), learning over time (Fuster 1991), language (but not word) comprehension (Hagoort and Indefrey 2014), social intelligence and interactions (Dunbar et al. 2010; Lieberman 2012) and creativity (Kaufman et al. 2010). As has been pointed out by several researchers (Tononi and Edelman 1998; Treisman 1999; Edelman and Tononi 2008), conscious binding requires the integration of complex information in the brain. The problem is to understand how the brain achieves this integration.

‘Philosophy is a battle against the bewitchment of our intelligence by means of our language’. Wittgenstein (2009, p. 109).

What do we mean by ‘integrated information’? To answer this question, we must first agree on a definition of information. I will here use that described by Claude Shannon and known as ‘Shannon information’ (Shannon 1948); which is essentially a measure of correlation between the degrees of freedom of a sender and receiver of a message, measured in bits. Neuronal firing rates thereby encode information about the outside world because some of its degrees of freedom are correlated with degrees of freedom of the outside world. I note that, in some theories of consciousness, causation is required in addition to correlation (Landauer 1991; Tononi et al. 1998).

Next, we must agree on how to distinguish conscious from non-conscious mental activity. I will follow the approach pioneered by Dehaene and colleagues who insisted that ‘subjective reports are the key phenomena that a cognitive neuroscience of consciousness purport to study’ (Dehaene and Naccache 2001). So, bringing these two definitions together, then conscious neuronal information (the sender) is that information encoded in the brain that correlates with the information encoded in the subjective reports (the receiver) of a conscious observer.

However, a great deal of information, as degrees of freedom in the brain, may be correlated with subjectively reported information, including the motion of ions through neuronal membrane, the motion of neurotransmitters within the synaptic cleft, the opening and closing of ion channels, blood flow or electromagnetic (EM) field perturbations generated by the motion of electrically charged particles. Each of these neuronal microstates knows, in the Shannon sense of its state being correlated with, some aspect(s) of the visual scene or subjective reports of that scene. Which is a likely physical substrate for the integrated information that must be encoded by conscious minds?

Before answering this question, it is first necessary to define what we mean by integrated information. This might appear to be an easy task as the term is widely used, so much so that there is evidenced by the United Nations Expert Group on the Integration of Statistical and Geospatial Information (http://ggim.un.org/UNGGIM-expert-group/), the International Society of Information Fusion (http://isif.org/), Information Integration Theory (Anderson 2014), data integration systems (Genesereth et al. 1997), numerous statistical and data mining methods that seek to integrate information from multiple sources (Maimon and Rokach 2005), as well as the integrated circuits of computers. However, as the physicist, Rolf Landauer, famously insisted, ‘information is physical’ (Landauer 1991) so integrated information, if it exists at all, must be encoded by a physically integrated substrate. Moreover, if it is to have an output then it needs to be causally competent (Pawłowski et al. 2009): the integrated information must, as an integrated unit, change something physical. Yet, none of the above examples of integrated information are physically integrated. Their information is causally integrated in time, rather than physically integrated in space, as I will now illustrate with a familiar problem from philosophy.

Ryle (2009) insisted that it is a category error to suppose that structures, such as the University of Oxford, have material existence. To make his point, he imagined a visitor to Oxford who visits the library and colleges but then asks ‘But where is the University?’ The visitor’s error is to assume that the university is a member of the category of material objects, rather than an institution which exists causally only in the minds of the students, staff and visitors to the university. Ryle calls this kind of mistake a ‘category error’. An analogous argument can be made for integrated information. University institutions, such as registry, finance, the libraries, exam boards, colleges or executives, integrate and process vast amounts of varied information ranging from student entry criteria, applicant qualifications, book catalogues, exam performance, timetables or salaries. However, this integration, like the institution itself, is causal, rather than physical, in the sense that downstream effects, such as the posting of offers of university places, depend on a multiplicity of upstream informational causes, such as the arrival of application forms and their scrutiny by academics and administrators. The integration is via a causal chain of operations in time, rather than physical integration in space.

This form of temporal integrated information is also a universal feature of computation, such as those performed by Boolean logic gates and instantiated in Turing machines, such as modern computers. For example, the single bit output of an AND logic gate integrates the two bits of information encoded in its two inputs (Fig. 1a) to output a single bit that represents an integration of the gate’s inputs. In reality, the integration is causal in the sense that the state of the output bit at time t2 is dependent on both input bits at time t1. Of course, the single output bit cannot encode both input bits: it is not physically integrated information. Note also that, once a signal has been transmitted from input to output via, for example, a current or voltage change, then the inputs are free to accept new data. So, by the time the signal has reached the output (t2), its state may no longer be correlated with the inputs. There is therefore no physical state that corresponds to the integration of information performed by the gate. This does not, of course, prevent the gate from participating in any complex algorithmic operation that can be considered to integrate information in the sense of the conventional temporal use of the term. Consider, for example, a computer connected to a camera and running an image recognition program whose job is to identify photographs of the actress Jennifer Anniston amongst hundreds of electronic images presented to the camera. By the time the complex series of computations that analyse and integrate features such as hair colour, eye colour, shape of nose, chin, complexion, orientation of the image, etc., have reached a gate that finally commits the program to deliver an audio output of ‘this is Jennifer Aniston’, the input logic gates may have gone on to the next photograph.

There can therefore be no physical state that corresponds to the integrated information at any single point in time. The integration exists as a correlation in time, not as a physical integration in space.

It is important to stress that no EM field theory of consciousness denies that much or most brain information processing proceeds via conventional neuron/synapse transmission. However, the same argument described above for integrated circuits, applies to the processing of complex information along complex neuronal pathways. They, like logic gates, input sensory information, such as photographs, and process that information along chains of neuronal networks until they reach a group of neurons, or even a single neuron that fires to generate a verbal output of ‘this is Jennifer Aniston’. Nevertheless, just as for logic gates, once a neuron has done its job of processing its many inputs to generate the single output of a firing rate, it is free to accept new data. Downstream neurons involved in the triggering motor output need not be correlated with the simultaneous state of input neurons or neurons involved in the processing of that sensory information by the time the output signal is generated. A hypothetical neuron that prompts the final output of a verbal report is also no more physically linked to its sensory inputs such, as in the retina or ear, than it is to a subject’s digestive system or skull. And even neurons that respond to highly processed and integrated information, such as the famous Jennifer Aniston neuron (Quiroga et al. 2005), only encode a single firing rate that cannot represent anything more than a tiny fraction of the information present in conscious percept. Its firing rate is, of course, all that any single neuron can know. Neurons integrate information but, as for logic gates, the integration is temporal, not physical, as the information to be integrated is separated in both space and time. Also, the so-called integrated information is far more complex than a single neuron’s coding capacity.

This kind of temporal integration cannot correspond to ‘our capacity to integrate information across time, space, attributes and ideas’ (Treisman 1999).

Of course, a huge number of firing neurons may fire in response to Jennifer Aniston’s face or some other potentially integrated conscious percept. It may be that each of these firing neurons encodes different aspects of the conscious percept—perhaps the particular colour of the actress’ eyes, the shape of her nose or the texture of her hair. One could imagine a network of such neurons that, together, encode and ‘integrate’ the conscious percept. But how do the firing neurons integrate their information? From the perspective of functionality, they need not even be physically connected so long as they, collectively, deliver their signals to a motor neuron(s) that execute(s) the verbal report ‘that is Jennifer Anniston’s face’. In reality, they need be no more physically connected to each other than are the different colleges of the University of Oxford. All that is required is a causal chain.

Nevertheless, nerves are, of course, physical connected, but, in the example, those encoding Jennifer Anniston’s face are no more physically connected to each other than they are to any other neurons or even cells of the body, particularly as matter-based signals, such as hormones, neurotransmitters or action potentials, do not attenuate with distance. If physical connectedness was sufficient for consciousness, then we would be aware of all of the information encoded in our entire body at all times. Neural networks, on their own, cannot be responsible for physically integrating conscious information because, like integrated circuits, they integrate information only temporally, not physically.

There are, however, physical systems that encode information integrated over space in a single moment of time. We know this form of information as force fields. The most obvious is the gravitational field that, at any point on the Earth’s surface, provides a force that effectively integrates the magnitude and distribution of local masses such as those of the Earth, Moon and Sun. Similarly, the EM field at any point in space represents an integration of information concerning the type, distribution and motion of local charges. In contrast to the temporal integration described above, force fields physically integrate complex information that may be simultaneously downloaded from any point in the field. This is apparent to anyone who views a TV show that has been transmitted from a single transmitter to their smartphone, alongside a thousand other people who may simultaneously view the same program on their phones in a thousand different locations. Moreover, an EM field can, like an integrated circuit, compute.

Consider for example, the arrangement of iron filings sprinkled over a magnet. A conventional computer could calculate their configuration by inputting the initial random configurations of the filings into an algorithm that implements either Maxwell’s equations or the equations of quantum electrodynamics to output their final equilibrium configuration. Yet the EM field at each point in space, generated by the electron spin of iron atoms within the magnet, instantly computes this solution. In this sense, the field represents an algorithm in space, rather than the algorithms in time that are implemented by Turing machines. And, most importantly, the information involved in the computation is simultaneously available in the space of the magnet and its surroundings. It is spatially, rather than temporarily, integrated information. The EM field’s information is complex information that is physically bound.

Magnets can also encode visual information. The artist, Andrzej Lenard, paints magnetic pictures, such as the portrait of Robert Downey Jr (Fig. 2 from https://www.youtube.com/watch? v=PHzz81yapcc). Note that the magnetic field-encoded information would be present whether the iron filings were there or not. The coding of the image would exist in space as invisible integrated information. This kind of coding is, I argue, much closer to the physical reality of our thoughts, than a firing neuron. However, it is static, rather than dynamic, and, unlike the signal from a WIFI router, the visual information is locally discrete rather than distributed throughout the space of the image.

Figure 1b illustrates how dynamic EM field information can integrate information and function as a logic gate. The figure illustrates an EM field AND gate with two inputs and a single output, each encoding either a zero or one. The inputs are dipoles that act as EMF transmitters that oscillate between two states. For ease of presentation, Fig. 1b illustrates the different inputs represented as firing (oscillating corresponding to input = 1) or non-firing (not oscillating, input = 0) states. The output is then an EMF receiver that implements the AND rule to output a signal. Note that the receiving node can be located anywhere in the space of the field. So, and most importantly, the entire AND computation is distributed throughout the space of the overlapping field of the inputs. The computation and integration of input information is implemented in space, rather than time. It is physically integrated information processing.

Combinations of EMF transmitters and receivers that implement different logical functions could, in principle, implement any complex algorithm, for example, an algorithm that recognizes images of Jennifer Aniston. In contrast to the above static magnetic image of Robert Downey Jr, this EM field algorithm would be distributed throughout the space of the field; such that, like a WIFI signal, it could be downloaded from any point in the space of the field. In this sense, the field possesses features in common with holograms that similarly store distributed information. But in the case of the cemi field, the information exists as an algorithm in space, rather than time. It is physically integrated information. Field-implemented algorithms such as these, but in the brain, are, I argue, the physical substrate of conscious thoughts.

It is, however, important to recognize that, although EM fields are maximally connected—the universe has a single EM field EM – waves travel at the speed of light across huge distances. However, their strength is subject to an inverse square (electric component) or cube (magnetic component) law so that the EM field perturbation of a single neuron rapidly falls off with distance. In my earlier paper, I estimated that the EM field electrical perturbation from the firing of a single neuron extends into a volume of only about 80 µm encompassing a maximum of about 200 neurons (McFadden 2002a). So, in contrast to matter-based signals that do not attenuate with distance, signals passing through the cemi field will tend to act only locally, unless boosted by chains of synchronization (see below).

Note however, and very importantly, that, in contrast to temporally integrated information, an algorithm in space can function only when its computational nodes fire synchronously so that their inputs are simultaneously available to all the components of the network. Therefore, a key prediction of the proposal that consciousness is distributed EM field-based algorithms is that conscious information will be correlated with synchronously firing neurons.

It has been known since the 19th century that the brain generates its own EM field, which can be detected by electrodes inserted to the brain. Its source is electrical dipoles within the neuronal membranes caused by the motion of ions in and out of those membranes during action potentials and synaptic potentials. The periodic discharge of neurons—firing or action potentials—generates EMF waves that propagate out of the neuron and into the surrounding inter-neuronal spaces where they overlap and combine to generate the brain’s global EM field that is routinely measured by brain scanning techniques such as electroencephalography (EEG) and magnetoencephalography (MEG). The human brain therefore possesses around 100 billion EMF transmitters.

The human brain also possesses at least 100 billion EMF receivers as each neuron in bounded by a membrane embedded with thousands of voltage-gated ion channels whose firing is triggered by EM field fluctuation across the membrane. Although these channels are generally assumed to respond only to large fluctuations of tens of millivolts across the membrane, much larger than the global EM field strength, EM field potential changes of less than 1 mV across the neuronal membrane are nevertheless capable of modulating neuronal firing (Schmitt et al. 1976). Moreover, for neurons poised close to the critical firing potential, the opening of just a single ion channel may be sufficient to trigger firing (Arhem and Johansson 1996). This degree of sensitivity suggests that very tiny changes in membrane potential, of similar strength to spontaneous fluctuations in the brain’s endogenous EM field, may influence the firing of neurons that are already close to firing.

The conscious electromagnetic information (cemi) field theory claims that the brain’s EM field is the physical substrate of consciousness. It was first outlined in a book published in 2000 in which I proposed that the brain’s ‘EM field … integrate[s] information from all of the calculations … performed by … [its] logic gates (McFadden 2000). The theory was further developed in two papers published in 2002 (McFadden 2002a,b). Similar theories were proposed around the same time by neurobiologist Pockett (2000, 2002), the neurophysiologist John (2001, 2002) and the neurophysiologists Fingelkurts and Fingelkurts (2008) and Fingelkurts et al. (2001). In 2013, I provided an update on the cemi field theory incorporating more recent experimental evidence (McFadden 2013a) as well as arguing that the theory accounts for the gestalt properties of meaning, in an accompanying paper (McFadden 2013b). In 2014, Adam Barrett similarly argued that the brain’s EM field integrates neuronal information to provide the substrate of consciousness (Barrett 2014).

The idea that the seat of consciousness is simply the brain’s EM field may initially sound outlandish but is no more extraordinary than the claim that the seat of consciousness is the matter of the brain. All it involves is going from the right to the left hand side of Einstein’s famous equation, E = mc2 thereby replacing the notion that consciousness is encoded by matter of the brain, with that of proposing that it is encoded by the energy of the EM fields generated by the motions of its charged matter. (Note that, by illustrating this idea with Einstein’s equation, I am not, of course, proposing any interconversion of matter and energy in the brain.) Matter and energy are equally physical; but, instead of being composed of material, the cemi field theory proposes that our thoughts are composed of the brains EM field energy. This is a kind of dualism, but it is scientific dualism based on the physical difference between matter and energy, rather than a metaphysical distinction between matter and spirit.

Although, as I have argued, shifting from matter to energy of the brain is conceptually trivial, when searching for an appropriate substrate in the brain that can physically integrated complex information, the move draws an immediate payoff, as it effortlessly solves the binding problem. Whereas information encoded in the matter of neurons is, as I have argued, always localized and discrete both in space and time, information in the field, as illustrated in Fig. 1b, is always integrated yet distributed, in the sense that it may be downloaded from any point within the field. Since, in this case, ‘the field’ is the brain’s global EM field, it also provides a feasible physical substrate for the notions of working memory and/or the global workspace that have been proposed in many other theories of conscious (Baars 2005). And, as illustrated in Fig. 1(b) EM fields may also implement algorithms. This capacity, known as ‘field computing’ (MacLennan 1999) sometimes as quantum-like computing (Khrennikov 2011), has several features in common with quantum computing such as ease of implementation of mathematical functions such as Fourier transforms, compared to digital computers. Moreover, as illustrated in Fig. 1b, this form of field computing—algorithms in space rather than in time—could only be implemented by neurons (either EMF transmitters or receivers) that fire synchronously. So, the theory predicts that, if consciousness is implementing field computing, then consciousness should be highly correlated with the synchrony of neural firing rather than firing rates.

Several decades ago, work conducted by Wolf Singer and colleagues demonstrated that neurons in the monkey brain fire synchronously when the animal attends to the stimulus (Kreiter and Singer 1996). Many additional studies confirmed and extended these findings to many different experimental systems. For example, work in David Leopold’s laboratory at Max Planck Institute for Biological Cybernetics, in Tubingen, Germany (Wilke et al. 2006) investigated awake monkeys trained to respond to a visual stimulus—the removal of a red dot from a target area—by pulling a lever (to receive their fruit juice reward). The researchers monitored both neuron spiking and changes in local EM field potentials in V1, V2 and V4 regions of the monkey’s visual cortex. They demonstrated that spiking of neurons in cortical areas V1 and V2 was totally uncorrelated with the monkey’s perception of the target; however, low frequency (alpha range, particularly 9–30 Hz) modulation of local field potentials—presumed to be generated by synchronously firing neurons—in these same regions did correlate with perception. It seems that though neuron firing rate in the primary visual cortex does not see the stimulus, synchronicity of neuron firing does indeed see the target.

Many subsequent studies have also demonstrated that neural synchrony also correlates with conscious perception in humans. For example, neural synchrony patterns were found to correlate with conscious recognition by subjects exposed to optical illusions (Lutz et al. 2002). More recent work has demonstrated that conscious auditory perception is correlated with long-range synchrony of gamma oscillations (Steinmann et al. 2014). Synchronization between the anterior and posterior cortex has been shown to correlate with consciousness levels of patients who have suffered traumatic brain injury (Leon-Carrion et al. 2012).