Integrated Information Theory (IIT)

Contents

Consciousness

Consciousness is a very subjective feeling privileged only to its owner, it can never be described fully to an outsider. Each person taking a look at the picture in Figure 01 would evoke an unique response: it may be perceived as a beautiful scenery, or invokes a feeling of serenity, or just a boring photo, or ... depending on the wiring in the brain. On the finer level, there would also be disagreement on the hue of the trees for example, whether they are green, dark green, dark, or anything in between; actually we will never know the "greenish" experienced by someone else. We can define "green" only by referring to a third object. Simply put : consciousness is an ineffable experience. This is often referred as the "hard problem in consciousness".

Figure 01 Different Perception of A Picture [view large image]

Neural Correlates of Consciousness (NCC)

		Study of the human brain owes the fortuitous development of Magnetic Resonance Imaging (MRI) in 1973. The non-invasive scanning technology called fMRI (f stands for metabolic function, i.e., the chemical process) allows the association of certain brain region (with spatial resolution the size of a pea containing about one million nerve cells) to specific mental task as shown in Figure 02 for seeing.
Figure 02 Active Brain Region	Figure 03 Neural Correlates of Consciousness	Such image is too coarse for the concept of Neural Correlates of Consciousness (NCC), which relates consciousness to the level of individual neurons (Figure 03).

In NCC, it is assumed that :


• The physical basis of consciousness is closely linked to specific interactions among neurons and their elements.
• Although consciousness is fully compatible with the laws of physics, it is not feasible to predict or understand consciousness from these laws alone. It doesn't seem to be an emergent phenomena either as suggested by some neuroscientists. On the applicability of quantum theory, it embraces the unpredictable nature from quantum indeterminacy because erratic motion confers evolutionary advantage to both preys and predators.
• Many animals, mammals especially, possess some of the features of consciousness. They see, hear, smell, and otherwise experience the world.
• The repertoire of conscious states must somehow diminish with the diminishing complexity of an organism's nervous system. Without a sound understanding of the neuronal architecture necessary to support consciousness, it is not possible to separate sentient creatures from those that do not feel anything.
• Consciousness is substrate-independent, it can be implemented by neurons in biological system; or by the wires, transistors etc. in an electronic circuit; or by something entirely different as long as it performs the same functions faithfully (that's the essence of what's known as functionalism).

Differentiation

The percept is always presented to its owner as a whole. There are evidences to show that this whole can be differentiated into parts with the most elementary one coming from just one neuron. It is called quale by philosophers of mind. The quale of the color red is the most basic element in such disparate percepts as seeing a red sunset, the red flag of China, arterial blood, a ruby gemstone, etc. The common denominator of all these subjective states is redness. The sensation of redness is an "intrinsic" and private experience for which nobody can share. Figure 04a depicts two visual examples : one sensory neuron in the cortex receives an electrical pulse from the Ruffini's ending in the skin and produces a sensation of heat; while another sensory neuron in the cortex generates a feeling of pressure triggered by a signal from the Pacinian Corpuscle cell in another part of the skin. This example shows that the experience is "structured" containing many different things. It is also "definite" in that only one conscious experience occurs at a time.

Figure 04a Differentiation [view large image]

In addition to the motor, sensory, and pyramidal neurons shown in Figure 04a, there are two more types known as inter and Von Economo neuron respectively (Figure 04b). An inter neuron (also called relay neuron, association neuron, connector neuron or local circuit neuron) is a neuron that forms a connection between other neurons. It can be in the spinal cord, the cortex, or cerebellum. The Von Economo neuron (VEN) was first identified in 1926 by Von Economo. It was re-discovered 80 years later to be closely associated with consciousness. These neurons make their rare appearance in two small regions : the anterior cingulate cortex (ACC) and the fronto-insular (FI) cortex (see Figure 04b) - the area for smells, flavours and other mental tasks.

Figure 04b Types of Neurons [view large image]

Circumstantial evidences indicate the same occurrence in other big-brained, social animals. It is associated with the sense of self, and environmental monitoring to adjust behavior. It comes with prominent size - up to 200% larger than typical human neurons.

Integration

Conscious states share another property of seamless "integration". As mentioned in the previous section the percept can be differentiated into parts; however these parts can never be separated. No matter how hard you try, you cannot see the world in black and white; nor can you see only part of the horse as shown falsely in a painting by Magritte (Figure 05a). It is postulated that underlying this unity of consciousness is a multitude of causal interactions among the relevant parts of the neurons. If areas of the brain become fragmented, disconnected, balkanized, as occurs under anesthesia , consciousness fades. Since there are more than 20 billion neurons in the cortex with their numerous interacting paths and changing environment, each percept is virtually unique. A very particular state is selected from countless possibilities - the essence of "information".

Figure 05a Integration [view large image]

The seemingly effortless generation of integrated image is not in place at birth. The infant acquires it through the five bodily sensors with a lot of training in the form of play (Figure 05b). The corresponding five senses for sampling the environment are :


• Light - the physical attribute of which is photon with various energy, while the subjective feeling is colors.
• Sound - it is wave with different frequency, but turns into pleasing or awful tone in personal experience.
• Smell - this comes from the various kind of molecules, but perceived as agreeable or dis-agreeable scent.
• Taste - similarly originates from different kind of molecules, but graded personally as whether it's palatable.
• Touch - physically it is the amount of pressure applied to the surface, while the personal touch is in terms of soft or rough.

Thus, although information in the environment exists in the form of raw materials with no preference for anything, humans and most animals convert them into some sort of sensations, which may not be unique among different species or even among different individuals of same species (the "hard problem in consciousness"). Data from various sources are used to modify an internally generated draft image in a time interval ~ 0.1 sec (for saving processing energy) as though they occur simultaneously (this short duration provides an opportunity for magicians to perform tricks). The brain also learns to identify the location of the sources of these data.

Figure 05b Integration Training [view large image]

Therefore, we always see an integrated image. This is particularly important for predators (misplacement means missing a meal), and preys (error in detection turns itself into dead meat).

only the outlines of objects in black and white without any details as shown in the insert in Figure 05c. He still has basically the same problem fifteen years later. He has difficulty with reading words and recognizing facial expressions. He has to enlist aids from the other senses to make sense of the imperfect visual perception. Figure 05c compares MM's brain activities with a normal one in seeing objects (blue) and faces (red). One explanation for the problem suggests that the seeing neural network has not stabilized at the age of three. The absence of activity in seeing allows the apparatus to be taken over by other functions such as hearing and touching - those functions useful for blind person. Thus, the sensation would becomes useless if it is not integrated properly at young age.

Figure 05c The Vision of MM [view large image]

See "Treating Blindness Takes More Than Meets The Eye" for details.

Information

According to one definition of "information", the meaning of the messages is completely ignored but takes into account only the variations that is possible in the message. Figure 06 shows an example which should be very easy to understand. The image on the left portrays Bob as a dummy who can only utter Ba Ba Ba Ba to Alice. There is no variation in the message resulting in no "information content". On the other hand, the picture on the right shows Bob to possess a wonderful command of vocabulary. Alice is surprise to hear the actual message out of so many possibilities. Therefore, this definition

Figure 06 Information Content [view large image]

uses the "average amount of surprise" as its criterion. It doesn't take into account the "meaning" of the message.

Effective Information

		The effective information defined in the original paper of "Integrated Information Theory (IIT)" is essentially the same as those From Eqs. (1) - (3). They have introduced some terminologies that can be identified to those in the previous section such as : "a priori repertoire" = K, "a posteriori repertoire" = k. The effective information Ie is defined as :
Figure 07 Effective Information [view large image]	Figure 08 Examples [view large image]	Ie = -I = K - k ---------- (4) to get rid of the troublesome negative sign. It is also referred to as entropy by neglecting a proportional constant.

Integrated Information

When isolated subsystems combine together to form an integrated whole through causal connections, it generates lot more information and thus richer experience - the essence of integrated information. Roughly speaking, while the "a priori repertorie" K in effective information is the sum of individual elements, it is the product for the case of integrated information. Mathematically, the integrated information is defined similar to Eq.(4) : = KI - kI ---------- (5) where KI = k1k2k3 ... , the ki is the "a posteriori repertoire" for each of the isolated subsystem, and kI is the "a posteriori repertoire" of the integrated whole.

Figure 09 Integrated Information [view large image]

Diagram A in Figure 09 depicts two independent subsystems in partitions M1 and M2. The integrated information is 0 because there is no connection between the two.

Qualia Space

The qualia space (or Q-space) is a multi-dimensional space with 2K dimensions for a system of K connected elements. Each point in the space represents a particular percept - the mental perception only experienced by its owner. It is not possible to portray such space in general intelligently. Figure 10 shows a 2-dimensional qualia space having only one element with 2 states - the 0 and 1, each coordinate is assigned numbers from 0 to 1 representing the probabilities in that state. For the case of random firing, initially each state has the probability of 1/2 (rod dot). A perturbation at time t0 transfers the element to state 1 (firing, the blue arrow) at time t1 by some kind of mechanism. In general the trajectories trail out a polytope (or crystal) in the qualia space, in which the history of our experiences is recorded. In reality, we could only remember vaguely the events in the past.

Figure 10 Qualia Space [view large image]

Prospects

The Integrated Information is in its infancy and lacks the graces of a full-fledged theory. Following lists some challenges for its future development :

		According to IIT, consciousness is universal as long as the neurons are integrated (connected). Therefore, all the biological species in Figure 11 have more or less consciousness depending on the value of . Figure 12 shows the possible percepts of light for different animals. It is just a reasonable re-construction from their eye structures. Nobody really knows what they "see". If the idea of functionalism (substrate-independent) is applied to consciousness, a conscious robot can conceivably be constructed (wired) to mimic nature.
Figure 11 Non-human Consciousness	Figure 12 Animal Percepts [view large image]	In July 2012, the first annual Francis Crick Memorial Conference declared that "non-human animals have the neuroanatomical, neurochemical, and neurophysiological substrates of conscious states

along with the capacity to exhibit intentional behaviors. Consequently, the weight of evidence indicates that humans are not unique in possessing the neurological substrates that generate consciousness. Non-human animals, including all mammals and birds, and many other creatures, including octopuses, also possess these neurological substrates" (Figure 13).

IIT has to explain the benefit of consciousness for the survival of the organism, although it is generally assumed that high- circuits confers quick reaction to changing environment. The Integrated Information is exceedingly difficult to compute even for very small systems such as the C. Elegan which possesses 302 nerve cells. There may be an easier alternative to compute . The mechanism to transfer one state to another should be elaborated in detail. Presumably, this mysterious process can be identified to the electrical pulse running along the dendrite or the axon. IIT makes specific predictions about which brain circuits are involved in consciousness and which ones are peripheral players, even though they might contain many more neurons. The theory should let doctors build a consciousness meter to measure the extent to which severely brain-injured patients are in a vegetative state, and which ones are partially conscious but unable to signal their pain and discomfort.

Figure 13 Animal Consciousness