An Evaluation of the Integrated Information Theory against Some Central Problems of Consciousness

This thesis evaluates the integrated information theory (IIT) by looking at how it may answer some central problems of consciousness that the author thinks any theory of consciousness should be able to explain. The problems concerned are the mind-body problem, the hard problem, the explanatory gap, the binding problem, and the problem of objectively detecting consciousness. The IIT is a computational theory of consciousness thought to explain the rise of consciousness. First the mongrel term consciousness is defined to give a clear idea of what is meant by consciousness in this thesis; followed by a presentation of the IIT, its origin, main ideas, and some implications of the theory. Thereafter the problems of consciousness will be presented, and the explanation the IIT gives will be investigated. In the discussion, some not perviously—in the thesis—discussed issues regarding the theory will be lifted. The author finds the IIT to hold explanations to each of the problems discussed. Whether the explanations are satisfying is questionable. Keywords: integrated information theory, phenomenal consciousness, subjective experience, mind-body, the hard problem, binding, testing 
 AN EVALUATION OF THE IIT !4 Table of Content Introduction 5 Defining Consciousness 6 Introduction to the Integrated Information Theory 8 Historical Background 8 The Approach 9 The Core of the IIT 9 Axioms 11 Postulates 13 The Conscious Mechanism of the IIT 15 Some Key Terms of the IIT 17 The Central Identity of the IIT 19 Some Implications of the IIT 20 Introduction to the Problems of Consciousness 25 The Mind-Body Problem 25 The Hard Problem 27 The Explanatory Gap 28 The Problem With the Problems Above 28 The Binding Problem 30 The Problem of Objectively Detecting Consciousness 31 Evaluation of the IIT Against the Problems of Consciousness 37 The Mind-Body Problem vs. the IIT 38 The Hard Problem vs. the IIT 40 The Explanatory Gap vs. the IIT 42 The Binding Problem vs. the IIT 43 The Problem of Objectively Detecting Consciousness 45 Discussion 50 Conclusion 53 References 54 AN EVALUATION OF THE IIT !5 Introduction Intuitively we like to believe that things which act and behave similarly to ourselves are conscious, things that interact with us on our terms, mimic our facial and bodily expressions, and those that we feel empathy for. But what about things that are superficially different from us, such as other animals and insects, bacteria, groups of people, humanoid robots, the Internet, self-driving cars, smartphones, or grey boxes which show no signs of interaction with their environment? Is it possible that intuition and theory of mind (ToM) may be misleading; that one wrongly associate consciousness with intelligence, human-like behaviour, and ability to react on stimuli? Perhaps we attribute consciousness to things that are not conscious, and that we miss to attribute it to things that really have vivid experiences. To address this question, many theories have been proposed that aim at explaining the emergence of consciousness and to give us tools to identify wherever consciousness may occur. The integrated information theory (IIT) (Tononi, 2004), is one of them. It originates in the dynamic core theory (Tononi & Edelman, 1998) and claims that consciousness is the same as integrated information. While some theories of consciousness only attempt to explain consciousness in neurobiological systems, the IIT is assumed to apply to non-biological systems. Parthemore and Whitby (2014) raise the concern that one may be tempted to reduce consciousness to some quantity X, where X might be e.g. integrated information, neural oscillations (the 40 Hz theory, Crick & Koch, 1990), etc. A system that models one of those theories may prematurely be believed to be conscious argue Parthemore and Whitby (2014). This tendency has been noted among researchers of machine consciousness, of some who have claimed their systems to have achieved at least minimal consciousness (Gamez, 2008a). The aim of this thesis is to take a closer look at the IIT and see how it responds to some of the major problems of consciousness. The focus will be on the mechanisms which AN EVALUATION OF THE IIT !6 the IIT hypothesises gives rise to conscious experience (Oizumi, Albantakis, & Tononi, 2014a), and how it corresponds to those identified by cognitive neurosciences. This thesis begins by offering a working definition of consciousness; that gives a starting point for what we are dealing with. Then it continues with an introduction to the IIT, which is the main focus of this thesis. I have tried to describe the theory in my own words, where some of more complex details not necessary for my argument are left out. I have taken some liberties in adapting the terminology to fit better with what I find elsewhere in cognitive neurosciences and consciousness science avoiding distorting the theory. Thereafter follows the problems of consciousness, which a theory of consciousness, such as IIT, should be able to explain. The problems explored in this thesis are the mind-body problem, the hard problem, the explanatory gap, the binding problem and the problem of objectively detecting consciousness. Each problem is used to evaluate the theory by looking at what explanations the theory is providing. Defining Consciousness What is this thing that is called consciousness and what does it mean to be conscious? Science doesn’t seem to provide with one clear definition of consciousness (Cotterill, 2003; Gardelle & Kouider, 2009; Revonsuo, 2010). When lay people talk about consciousness and being conscious they commonly refer to being attentive and aware and having intentions (Malle, 2009). Both John Searle (1990) and Giulio Tononi (Tononi, 2008, 2012a; Oizumi et al., 2014a) refer to consciousness as the thing that disappears when falling into dreamless sleep, or otherwise become unconscious, and reappears when we wake up or begin to dream. The problem with defining the term consciousness is that it seems to point to many different kinds of phenomena (Block, 1995). In an attempt to point it out and pin it down, the AN EVALUATION OF THE IIT !7 usage of the term needs to be narrowed down to fit the intended purpose. Cognition and neuroscientists alike commonly use terms such as non-conscious, unconscious, awake state, lucid dreaming, etc. which all refer to the subjective experience, but of different degrees, levels, and states (Revonsuo, 2009). Commonly used in discussions regarding consciousness are also terms such as reflective consciousness, self-consciousness, access consciousness, and functional consciousness. Those terms have little to do with the subjective experience per se, at best they describe some of the content of an experience, but mostly refer to observed behaviour (Block, 1995). It seems that researchers of artificial machine consciousness often steer away from the subjective experience. Instead, they focus on the use, the functions, and the expressions of consciousness, as it may be perceived by a third person (Gamez, 2008a). In this thesis, the term consciousness is used for the phenomenon of subjective experience, per se. It is what e.g. differs the awake state from dreamless sleep. It is what differs one’s own conscious thought processes from a regular computer’s nonconscious information processing, or one’s mindful thought from unconscious sensory-motoric control and automatic responses. It is what is lost during anaesthesia and epileptic seizures. Without consciousness, there wouldn’t be “something it is like to be” (Nagel, 1974, p. 436) and there would be no one there to experience the world (Tononi, 2008). Without it we would not experience anything. We would not even regard ourselves to be alive. It is the felt raw experience, even before it is attended to, considered and possible to report, i.e. what Block (1995) refers to as phenomenal consciousness. This is also often the starting point of cognitive and neurological theories of consciousness, which try to explain how experience emerge within the brain by exploring the differences between conscious and nonconscious states and processes. AN EVALUATION OF THE IIT !8 Introduction to the Integrated Information Theory Integrated information measures how much can be distinguished by the whole above and beyond its parts, and Φ is its symbol. A complex is where Φ reaches its maximum, and therein lives one consciousness—a single entity of experience. (Tononi, 2012b, p. 172) Historical Background The integrated information theory originates in the collected ideas of Tononi, Sporns, and Edelman (1992, 1994). In their early collaborative work, they developed a reentering model of visual binding which considered cortico-cortical connections as the basis for integration (Tononi et al., 1992). Two years later they presented a measure hypothesised to describe the neural complexity of functional integration in the brain (Tononi et al., 1994). The ideas of the reentering model and neural complexity measure developed into the more known dynamic core hypothesis (DCH) of the neural substrate of consciousness (Tononi & Edelman, 1998). The thalamocortical pathways played the foundation of sensory modality integration. In the DCH, a measure of integration based on entropy was introduced, which later became Φ, the measurement of integrated information (Tononi & Sporns, 2003). This laid the foundation for the information integration theory of consciousness (Tononi, 2004). The IIT is under constant development and has since it first was presented undergone three major revisions. The latest, at the time of writing, is referred to as version 3.0 (Oizumi et al., 2014a), which this thesis mostly relies on. The basic philosophical and theoretical assumptions have been preserved throughout the development of the theory. Some of the terminology and mathematics have changed between the versions (Oizumi, Amari, Yanagawa, Fujii, & Tsuchiya, 2015). Axioms and postulates that clarify the theoretical base was added in the second version (Tononi, 2012a) and presented more clearly and became central to the theory in version 3.0 (Oizumi, AN EVALUATION OF THE IIT !9 Albantakis, & Tononi, 2014b). The changes have to some extent impacted the interpretation of the theory and its implications. For example, the significance of the thalamocortical system in the generation of consciousness have lessened. Instead, cortical connections, in general, seems to have gained interest, perhaps to open up for a broader interpretation of the theory. The Approach Tononi and Koch (2015) argue that it may be impossible to solve the problem of how the brain gives rise to experience by solely studying the brain and neural correlates of consciousness using a pure bottom-up approach. Instead they take a top-down-back-up approach, starting by looking at the phenomenological aspects of conscious experience, then hypothesise what sorts of mechanisms that in theory may give rise to those aspects, and then try to verify the ideas by finding matching mechanisms in the brain which correlates with consciousness. This is the approach the integrated information theory (IIT) takes. The IIT starts from intuitive phenomenological aspects of subjective experience and then suggests what mechanisms a physical system may need to give rise to consciousness (Tononi & Koch, 2015). By doing this, the IIT attempts to discriminate nonconscious systems from a conscious system. The Core of the IIT The IIT identifies five fundamental intuitive phenomena of conscious experience expressed as axioms: intrinsic existence, composition, information, integration, and exclusion. The phenomenological axioms constitute the set of premises on which the theory is based. I will return to the axioms and describe them in greater detail below. Ideally the set of the five axioms is complete and each axiom independent and truly valid. Whether they are, is open to debate and scrutiny (Tononi & Koch, 2015). From the five axioms, the IIT derives AN EVALUATION OF THE IIT !10 ontological postulates. The postulates describe the characteristics of the mechanism a system must have to be conscious. As with the axioms, I will come back to and describe the postulates and the predicted mechanism of consciousness in the sections below. The IIT argues that consciousness is a fundamental property of reality—as fundamental as charge or mass—possessed by physical systems having internal causal powers that determines possible past and future states (Tononi, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). The IIT holds that consciousness is integrated information, i.e. it is the one and the same thing (Tononi, 2008). Since consciousness is thought to be a fundamental property of reality and being the same thing as integrated information, it is argued that integrated information can be measured as a fundamental quantity (Tononi, 2008; Tononi & Koch, 2015). "Integrated information (Φ [, the Greek letter phi]) is defined as the amount of information generated by a complex of elements, above and beyond the information generated by its parts." (Tononi, 2008, p. 216). The IIT is using a non-Shannonian definition of information. Shannon (1948) defined information objectively and stripped from meaning, while IIT defines information in subjective terms, full of meaning. Information is in this sense meaningful from the intrinsic perspective of the system holding it (Oizumi, Albantakis, & Tononi, 2014c). Qualia that may seem pointless such as phosphenes are informative and meaningful in the way they make a difference to the experience. The IIT follows Bateson (1972) and says that information is “differences that make a difference” (para. 459). In IIT, information is seen as being intrinsic to the physical structure in which the information is represented. It is the physical structure itself which holds the meaning (Oizumi et al., 2014a; Tononi & Koch, 2015). Integrated information may be understood as the synergy effect of when chunks of (integrated) information are combined and form a larger coherent whole. If no synergy effect arise, no AN EVALUATION OF THE IIT ! 11 additional integrated information is generated, hence, no consciousness. The amount of integrated information is measured as entropy, in the IIT, which is the structural loss when integration breaks down (Tononi & Sporns, 2003). The IIT seems to make no distinction between information and integrated information. Instead, the IIT argues that information must be integrated; hence, all information is integrated information (Oizumi et al., 2014c). This implies that wherever there is information there is also consciousness. I do not agree with this and would like to argue that information absent of meaning may be carried, copied, transposed, and processed by non-conscious media and non-conscious processes, e.g. stored and processed in a non-conscious computer. Information is then attributed meaning first when it is transferred to a host capable of consciousness with the ability to integrate information. Axioms The phenomenological axioms of the IIT are based on intuition and are assumed to be essential properties of consciousness (Oizumi et al., 2014a) but was originally thought to be taken to be “Self-evident truth ... [that] cannot be doubted and do not need proof.” (p. 4). The strong hold on the axioms have recently been weakened and are now admittedly open for validation (Tononi & Koch, 2015). Let us take a closer look at the five axioms of the IIT, which are: intrinsic existence, composition, information, integration, and exclusion (Tononi & Koch, 2015). Intrinsic existence. The first axiom is about intrinsic existence. Tononi (2012a) paraphrase Descartes “I experience therefore I am.” (p. 296). First, it states that consciousness exists. It is an undeniable fact (Oizumi et al., 2014a), the only fact that one can be absolutely sure of (Tononi, 2008). “All the rest is conjecture” (Tononi & Koch, 2015, p. 2). Secondly, it states that consciousness exists intrinsically to the system and is independent AN EVALUATION OF THE IIT !12 of external observers (Tononi & Koch, 2015). It is the brain tissue itself, in which the experience emerge, that is conscious and experiencing. The subjective experience exists internally of the system. It is not an external property or product of the system. Consciousness does not arise out from the system; it arise within the system. Composition. The second axiom regards the composition: an experience is structured and consists of many different aspects in a non-arbitrary way (Tononi, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). For example, one may experience a book, a chair, a desk, soft light, vanilla scent, music, the taste of milky tea, a word, a sentence, a story. All within the same experience. Information. The third phenomenological axiom concerns information: an experience is informative, it is information rich. Each conscious experience is unique and differs in its specific way from every other possible experience. An experience of total tranquility and pure emptiness is informative in what it lacks from other experiences (Tononi, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). Integration. The fourth axiom is about integration: a conscious experience is integrated and unified into one whole experience, which cannot be split into separate, independent sub-experiences (Tononi, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). One do not for example experience the redness and the squareness of a red square independently, the redness and the squareness are experienced as a unified whole that is not reducible to its components. Exclusion. And finally, the fifth axiom regards exclusion: each experience is unique and fully defined. Its boundary, content, spatiotemporal grain (speed of flow and resolution) is unambiguous. An experience is limited, there are neither less nor more qualities to it than there is, and it flows with time at a certain speed. What does not belong to that experience is AN EVALUATION OF THE IIT !13 not experienced. Every single experience is different from every other possible experience. Even the smallest difference differentiates experiences. An experience would not be what it is if anything changed, it would then simply be another experience. Each experience excludes all others. One can only experience one experience at any given time. No superpositioning of experiences is allowed. For example, a daydream or a memory is not a separate experience. It is a part of the overarching awaken experience of the here and now (Tononi, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). Included within the experience is only the stuff of which one is conscious. All other aspects are excluded, such as e.g. one’s blood pressure or tomorrow’s smell of the lunch that is not part of the current experience. They may, though, be part of some other experience at another time. One may also experience that time passing by at a quite regular speed, even though it sometimes seems that time almost stand still and sometimes fly away. It appears, though, to be a limit to the scope of how fast and slow time may seem to pass. Postulates From the phenomenological axioms, mentioned above, the IIT derives the ontological postulates a physical system must fulfill, according to the theory, for consciousness to arise. The postulates lay the foundation for a architecture of a mechanism that the theory holds will generate so-called integrated information. The mechanism may be said to, therefore, account for the consciousness of a system. The architecture is believed to be platform independent, e.g. consist of bio-neurological, silicon-based, or any arbitrary circuitry. That which is referred to as the system in the section below should be equated with the physical substrate of consciousness, not including unconscious processing of e.g. sensory organs and motor abilities, if not otherwise stated. AN EVALUATION OF THE IIT !14 Intrinsic existence. For consciousness to emerge mechanisms in a state must exist, which comprise a coherent system. Those mechanisms must have cause-effect power, i.e. be influential onto the rest of the system. Cause-effect power signifies that the current state of a system constrains the repertoire of possible past and future states of the system. For a system to exist intrinsically it must be holistic, i.e. every element must be causally dependent on every other element in the system. Its current state necessarily determines its past and future states (Tononi & Koch, 2015). Composition. The IIT holds that a conscious system must be structured. Elements may form elementary mechanisms. Lower order simpler mechanisms may form higher order more complex mechanisms. Due to the IIT, it is necessary that mechanisms are formed into a structure for consciousness to arise. A structure in which the mechanisms are internally causally efficacious on the rest of the system (Oizumi et al., 2014a; Tononi & Koch, 2015). Information. For consciousness to arise the mechanisms of a system must intrinsically to the system specify “differences which make a difference” (Bateson, 1972, p. 459). The state of the system is meaningful to the system only from its own subjective perspective (Oizumi et al., 2014c). The current state and structure of the mechanisms within the system must limit the possible past and future states of the system (Tononi, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). How informative an experience is, is defined by how specific and unique it is, not how many aspects or objects it contains (Tononi, 2008). Integration. The structure of the system must be intrinsically integrated and indivisible, all the subsets must be essential. Any attempt to split the system into two or more parts will result in loss of integrated information, i.e. reduced level of consciousness (Tononi & Koch, 2015; Oizumi et al., 2014a). AN EVALUATION OF THE IIT !15 Exclusion. Included in the conscious system are only mechanisms and elements that do contribute to consciousness. Excluded are those elements and mechanisms that are not fully integrated and hence not essential to consciousness. The resulting set of elements is the set that is maximally irreducible and generates most integrated information, Φmax. Excluded are also elements and mechanisms that do not contribute to consciousness in such a way that Φ increase. Neither the full set nor any sub-set of this system is allowed to be part of any other simultaneous system that may also be conscious. A conscious system has to be in one and only one state at a time. No superpositioning of states is allowed. A system cannot be in none, two or more states at the same time or fluctuate between states. The timespan in which a system in a state generates most integrated information is definite. It cannot be longer nor shorter without loss of integrated information (Tononi, 2008; Tononi, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). The Conscious Mechanism of the IIT The mechanism of information integration that the IIT predicts may be described in the following way. During the split second a specific experience may last, distributed clusters of neurons from all over the cerebral cortex connect and form one coherent, indivisible mechanism (a complex) which contain all the aspects of the experience. The state of the single elements (the neurons) represent the information that is then integrated into the unified experience. The integration happens as the elements form a complex, an indivisible mechanism, by functionally connecting to each other. Note that the anatomical structure must already be in place for this to happen. The overarching state of this singular complex structure represents the integrated information. An informative synergy effect seems to arise when the neurons connect and form the coherent structure of the mechanism. That means that AN EVALUATION OF THE IIT !16 the amount of information the combined structure represents is greater than the simple sum of its parts. The greater the synergy effect is, the higher Φ is generated. If no synergy effect then no information is integrated, resulting in zero Φ, and no consciousness. The IIT regards the subset of elements, neurons or its analogues, that generates the highest Φ as the conscious entity, the main complex, of the organism. When the next experience is coming up in the stream of consciousness, the composition of the complex is altered accordingly to hold the integrated information of this new experience. The anatomical structure remains the same while the functional connections change. In this way, the physical substrate of consciousness changes dynamically together with the experience. Note that the physical substrate of the conscious experience is only the neurons that are functionally connected and which state represents the integrated information of the experience. To visualise this, imagine a simplified neuron to be in a state holding a value (e.g. active or inactive). The neuron’s synapses function as switches that keep links open or closed between anatomically connected neurons; conveying or not conveying the state of the neuron. When consciousness fades, e.g. falling asleep, the structure change so that less information is integrated. The qualitative aspects of consciousness are the integrated information represented by the state of the physical substrate while the quantitative aspect is measured as the amount of integration of information. The conscious experience is intrinsic to the substrate and emerges within the neurological substrate itself. It is the substrate of the experience which is having the experience (Tononi & Edelman, 1998; Tononi, 2008, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). There is no subsidiary mechanism observing the integrated information. It is the mechanism that constitutes the integrated information that is having the experience. The fallacy of homunculus, i.e. infinite regress of observers, is therefore avoided. AN EVALUATION OF THE IIT !17 Some Key Terms of the IIT The IIT can be difficult to grasp due to terminological confusion. Not only that the terminology of the IIT differs somewhat from the common usage of the terms, but may also differ in the literature between different versions of the IIT. Some terms appear to be replaced while many are synonymous. Concept. The IIT notation of a concept differs in some ways from the traditional use of the term. The term concept may be understood to be a building block of which an experience consists. A concept is ontologically the state and structure of a causally efficacious mechanism, a set of elements generating integrated information. Phenomenologically a concept is synonymous with the term quale (and its plural form: qualia) (Tononi, 2012a; Oizumi et al., 2014a). It is the intrinsic meaning from the mechanism's subjective perspective. It is an aspect of the conscious experience, e.g. smell, a sound, a direction, a feeling, or any object. Essentially a concept is integrated information that is one aspect of the full blown experience (Oizumi et al., 2014a; Tononi & Koch, 2015). Complex. A complex is a combination of concepts. The main complex is the sub-set of elements and their states within a conscious organism that is generating the highest Φ, Φmax. Phenomenologically it is the subjective experience. Ontologically it equals to the physical substrate of the conscious experience. It consists of all mechanisms, all concepts, that holds the full-blown conscious experience (Tononi, 2012a; Tononi & Koch, 2015). A complex is the same as the system referred to in the section above on postulates. A complex causally connects with its environment via its in-ports (e.g. primary visual cortex) and affect its surroundings via its out-ports (e.g. motor cortex). The mechanisms exterior to the inand outports do not directly take a part in the conscious experience and are, therefore, excluded from the complex (Tononi, 2008). A neural complex may also be perturbed directly by AN EVALUATION OF THE IIT !18 concussions, substances such as drugs, electromagnetic fields, electric currents, etc., which alter the state and structure of the complex and the experience. Spatiotemporal Grain. The IIT literature refers to what it calls the spatiotemporal grain of consciousness. The spatiotemporal grain is the boundary that defines the experience in both spatial and temporal dimensions. Spatially the grain encapsulates the physical substrate of an experience, i.e. the extension of a complex. It also defines the time scope in which the physical substrate is assumed to stay in one state (Tononi, 2010; Tononi & Koch, 2015). Note, that (bio-)physical processes may be allowed to occur to maintain the state of the complex. The IIT predicts that integrated information is generated at various levels of complexity within a system, but there is a level where mechanisms in a state emerge which generates the highest Φ, Φmax. This abstraction level defines the spatial level of the system at which the consciousness of a complex arise (Tononi, 2008). In a similar way, there is a time span at which mechanisms in a state emerge that generates most integrated information, Φmax. If the time span is too short or too long, little or no integrated information will be generated. At the right time span, the mechanisms will produce highest Φ. These spatial and temporal aspects determine the optimal spatiotemporal grain size. For a neurological system, the optimal spatial level seems to appear at the level somewhere between single-cell neurons and cortical areas while the temporal aspects may be most optimal at around a few hundred milliseconds. (Tononi, 2012a). Concept space. The concept space (also called qualia space and cause-effect space in the IIT literature) may perhaps be easiest to understand as a graphical visualisation of the experiences a complex may have. The concept space is a high-dimensional space, where the dimensions represent all possible future and past states of a complex. An experience may be AN EVALUATION OF THE IIT !19 presented as a constellation of concepts in space. The coordinates of a concept within this space represent its probability of its past and future states. The shape of the constellation is as unique as the experience it represents. If two systems produce identical constellations, it is said that they have identical experiences (Tononi, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). The old question whether two people experience e.g. the colour green the same way may, in theory, be possible to answer by representing their experiences in concept space. Unfortunately, to represent all the possible states a human brain may hold, the number of dimensions in concept space will reach towards infinity, which renders concept space impractical for this purpose. Concept space will not be further discussed in the argumentation of this thesis and is hereby left aside. The Central Identity of the IIT The IIT holds that consciousness is integrated information, i.e. integrated information is the one and the same thing as consciousness (Tononi, 2008). To give an understanding of what this statement holds I will summarise what has been said about the IIT in this thesis so far. Consciousness is informative. For information to exist in a substance monistic physical universe it must be represented in some physical manner, the IIT’s non-Shannonian integrated information is no exception. In Shannon’s (1948) definition of information, information is objective and stripped of meaning. The IIT defines integrated information as necessary subjective and meaningful (Oizumi et al., 2014c). The subject that holds the meaning is also the physical carrier that structure and state represent the information. Every single element of this structure is causally interconnected into one integrated system, one coherent non-divisible mechanism. In other words, it is the physical substrate (the complex) AN EVALUATION OF THE IIT !20 of the conscious experience which is experiencing. The quality and content of the experience is fully specified by the state and structure of the physical substrate. Each conscious experience is unique and differs from every other possible experience (e.g. Tononi, 2012a; Oizumi et al., 2014a). The IIT says that information is the difference that makes a difference and follows Bateson (1972). Therefore, it can also be said that an experience is definite. It is fully defined and irreducible. It is what it is. If anything change, it becomes another experience. From this follows that the physical substrate must be fully defined and definite as well. The physical substrate that physically represents the content and the quality of the experience must, i.e. the integrated information, do according to the theory, differ in some sense (if only slightly) between different experiences. The level of consciousness equals to the quantity of integrated information (Φmax). It measures the irreducibility of the complex and is assessed with an entropy formula (e.g. Tononi & Sporns, 2003; Oizumi et al., 2014a). Some Implications of the IIT From the postulates of the IIT, predictions can be made. Below some of them are explored. None, one, or many consciousnesses. The IIT predicts that not all systems are conscious. For example, pure feed-forward systems do not have the ability to generate any integrated information, Φ = 0. Neither do they form any complex and are, therefore, regarded as completely non-conscious. Systems that contain mechanisms in a state that have a causal effect on the rest of the system do generate integrated information, Φ > 0. As long as Φ is above zero, the IIT claims that the system is conscious (Tononi, 2008, 2010, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). AN EVALUATION OF THE IIT !21 The IIT also predicts that a system may contain multiple maximally irreducible complexes that all produce integrated information at various degrees. Hence, a system may contain multiple consciousnesses (Tononi, 2008, 2010; Oizumi et al., 2014a; Tononi & Koch, 2015). Research on split-brain patients, which hemispheres been separated by corpus callosotomy surgery to prevent severe epileptic seizures, indicate but is not conclusive that they may possess two separate consciousnesses (Gazzaniga, 2005). Since consciousness is an intrinsic property of a complex, each complex may be regarded as an individual from its perspective (Tononi, 2012a). Complexes within the same system may have a causal effect on each other in a feed-forward manner, Φ = 0, or that the interaction between the complexes is generating lower Φ than within each complex. As long as the amount of integrated information generated between the two complexes is less than what is generated by each and one of them, the complexes are seen as separate. The complex that produces highest Φ is seen as the main complex of the system. Whether complexes are regarded to be part of the same system or not depends fully on the abstraction level at which defines the system. A group of people is not regarded by the IIT as having a unified group consciousness. Because, every individual is producing a higher Φ than the whole group taken together, i.e. the group is not generating enough information above and beyond its parts (Tononi & Koch, 2015). In cases where each individual may be assumed to generate a relatively low Φ, but the interaction between the individuals is very strong, such as in an ant colony. I believe, the group effort may perhaps generate a higher Φ than that of each individual, thus, achieve a group consciousness. The cerebral cortex is an excellent arena for consciousness. The functional specialisation of different cortical areas over the cerebral cortex suites well with the predictions that specialised networks generate a high amount of integrated information AN EVALUATION OF THE IIT !22 (Tononi & Sporns, 2003). The rich amount of thalamocortical, cortico-cortical, and interhemispheric callosal pathways that connect widely distributed and functionally specialised cortical areas supports the idea of a dynamic complex capable of generating vivid experiences. The IIT predicts that such a system may generate a huge amount of integrated information with an enormous repertoire of possible experience (Tononi, 2004, 2008, 2012a; Oizumi et al., 2014a). Cerebellum does not contribute to consciousness. The cerebellum, despite its 69 billion neurons (Herculano-Houzel, 2009), appears to have little to do with cognitive functions, and probably even less with producing conscious experience. The cerebellum is modular in its organisation and connects to areas in the prefrontal cortex, involved in regulation of eye movements, and motor cortex via thalamocortical projections (Glickstein, 2007). Simulations of such modular networks with limited inter-modular activity show that very little integration occurs. The amount of generated integrated information within such networks are generally lower that the amount generated by its parts (Tononi & Sporns, 2003; Oizumi et al., 2014a). Compare with the group example above. Consciousness is graded. Since consciousness measures as the amount of integrated information generated by a complex, the IIT holds that consciousness is a graded and not an all-or-none phenomena. Different complexes may generate different amounts of consciousness. The complex formed at fully awaken state is predicted to generate more integrated information, i.e., a higher Φ, than the complex formed during dreamless sleep or anaesthesia (Tononi, 2008, 2012a; Tononi & Koch, 2015). Dynamic functional connections. Depending on the current states of the mechanisms within a system, the IIT predicts that the intrinsic functional connections may change resulting in changed causal effects, without any change to the physical structure. This AN EVALUATION OF THE IIT !23 influences and shapes the conceptual structure, i.e. the complex. Consequently, both the amount of generated integrated information and the qualitative aspects of the experience change (Oizumi et al., 2014a). The thalamocortical system seems to provide this functionality in producing conscious experience (Tononi & Edelman, 1998). Lack of consciousness during dreamless sleep and anesthesia. The IIT predicts that the level of consciousness will be greatly reduced during dreamless sleep and anesthesia (Oizumi et al., 2014a). Simulations have shown that the total sum of generated integrated information in homogenous networks is not greater than the amount generated by its parts (Tononi & Sporns, 2003). Even though the thalamocortical system remains active during dreamless sleep, the cortical activation patterns become stereotypical across the cerebral cortex (Massimini et al., 2005). The thalamocortical system appears to be impaired during anesthesia resulting loss of consciousness (Ferrarelli et al., 2010). Inactive systems can be conscious. The IIT predicts that inactive systems may be conscious, provided that the elements are causally efficacious within the system and able to respond to perturbing signals from the outside. It is the state in which a mechanism is in, independent if it is active or inactive, that contribute to consciousness. Not the signal that is sent between mechanisms (Oizumi et al., 2014a). Simple systems can procure consciousness The IIT predicts that a simple system consisting of only two elements may be able to generate integrated information and hence be conscious. Consciousness will occur as long as the elements have a causal effect on each other so that they form a complex that is maximally irreducible (Oizumi et al., 2014a). Consciousness has evolutionary advantages. The IIT argues that the integrated composition of conscious systems has an evolutionary advantage over non-conscious feedforward systems. Some benefits an integrated system may reap are computational efficiency, AN EVALUATION OF THE IIT !24 energy preservation, memory abilities, fewer elements, flexibility, adaptability, and noise insensitivity (Tononi, 2012a; Oizumi et al., 2014a). Not only brain tissue can give rise to consciousness. The IIT predicts that consciousness may arise in any system that lives up to the postulates, independent on the substance of the system. Neural tissue is, thus, not a prerequisite for consciousness (Tononi, 2008, 2010, 2012a; Oizumi et al., 2014a; Tononi & Koch, 2015). Pure feed-forward systems are non-conscious. As mentioned above, the IIT predicts that feed-forward systems do not integrate information and can therefore not give rise to consciousness (Oizumi et al., 2014a). Zombies may exist. The IIT implies that zombies, i.e. nonconscious systems indistinguishable to conscious systems from an observers perspective, may exist (Oizumi et al., 2014a; Tononi & Koch, 2015). Measure a brain’s consciousness level by measuring its capacity to integrate information. The IIT predicts that it is possible to measure objectively consciousness level of a brain by assessing its capability to integrate information (Tononi & Koch, 2015). AN EVALUATION OF THE IIT !25 Introduction to the Problems of Consciousness There is nothing that we know more intimately than conscious experience, but there is nothing that is harder to explain. (Chalmers, 1995, p. 200) Chalmers (1995) argues that every reductive attempt to explain consciousness has failed and that we, therefore, may need to find another way to explain consciousness. Perhaps we don’t need to find a different way and stay put with reductionism. Perhaps we only need to understand the problems of consciousness in a different way. Maybe Dennett (1991, 2013) is right, and the problems only exist because we refuse to see that they are just illusions created by persistent preconceptions. Let’s have a look at what those problems are. In the scientific search for a theory of consciousness, scientists and philosophers alike seem to have focused in on what may be taken to be—for the purposes of this thesis—some central problems of consciousness. The problems considered are the mind-body problem (Descartes, 1641/2008), the hard problem (Chalmers, 1995, 1996), the explanatory gap (Levine, 1983), the binding problem, and the problem of testing for consciousness. The mindbody problem, the hard problem, and the explanatory gap are, as we will see later, closely related. Questions relating to functional aspects of consciousness, those which ask why consciousness exist, and what is conscious have been left out from this thesis. Even though a theory of consciousness, such as the IIT, may suggest answers to these questions as well. The Mind-Body Problem The mind-body problem originates in the Cartesian dualistic problem on how does a God-given immaterial soul interact with the physical body (Descartes, 1641/2008). But today, three centuries later, most scientists seem to hold the belief that the brain is causing the mind. Consciousness is now commonly thought to be a physical phenomenon (e.g. Dennett, 1991; AN EVALUATION OF THE IIT !26 Searle, 2002; Tononi & Koch, 2015; Thagard & Stewart, 2014; Graziano & Webb, 2014). Now, only the direction of the problem is changed. The question now is instead how does the body (brain) give rise to the mind. So, the problems remain. Some of the confusion regards the apparent dualistic nature of consciousness. Dualism is, in general terms, the view that there are two complementary (or contradictory) opposites, of which neither one is reducible to the other (Dualism, n.d.; Hermerén, n.d.). There are, though, some varieties of dualism to consider in regards to consciousness: substance dualism, property dualism, predicate dualism, and perspectival dualism. Substance dualism. Substance dualism is the idea that the existence consists of two basic substances. The classic Cartesian dualism is one expression of substance dualism where the two substances are mind and matter (Robinson, 2011). Property dualism. Property dualism is the idea that there is a fundamental ontological distinction between mental and physical properties of (physical) things. For example, while a waterfall may be fully described as a cluster of H2O molecules and physical laws acting on the molecules, the wetness of the water cannot. Property dualists hold that there are mental aspects of experience which cannot be described by physics. Implying that consciousness is an emergent phenomenon that cannot be fully explained as physical brain states (Robinson, 2011). Predicate dualism. Predicate dualism is the idea that mental experiences of the world cannot be fully reduced to pure physical descriptions without loss of information. The language itself resists the necessary reduction. For example, water is the same as, and can be reduced to H2O, but a waterfall, it is argued, cannot be fully reduced to a physical description of H2O (Robinson, 2011). AN EVALUATION OF THE IIT !27 Perspectival Dualism. Perspectival dualism—the term originally coined by Nancy Fraser and used in socioeconomics—may be used synonymously with double-aspect theory, and dual-aspect monism. It is the idea that mind and matter really are the one and the same thing, described from two different perspectives, e.g. from the subjective (first-person) intrinsic perspective and the objective (third-person) extrinsic perspective. It is when one denies perspectival dualism the mind-body problem appears (Parthemore, 2011; Stubenberg, 2010). Searle (2002) argues that two reasons why the mind-body problem does not seem to go away are that we do not yet know how the brain produces consciousness. Our vocabulary traditionally confuses us by separating the mind and the body, the mental and the physical, from each other. The Hard Problem Chalmers (1996) suggests a division of the mind-body problem into two parts: the easy problems and the hard problem. With the easy problems, he refers to the phenomenons of consciousness that can be explained in full by using standard functionalistic methods of cognitive science. Memory, language, spatial navigation, information integration, etc., are referred to as easy problems that Chalmers believe are fully explainable in neural mechanistic and computational terms. Dennett (2013) points out that they are “still difficult enough” (p. 310) to explain. The hard problem, instead, says Chalmers (1995), is “the problem of experience” (p. 202) which seems to escape the traditional methods of explanation. He raises the question, ‘How can a physical system give rise to subjective experience?’ AN EVALUATION OF THE IIT !28 The Explanatory Gap Levine (1983) raises the question what he refers to as the explanatory gap. He says that even though we one day may know everything there is to know about the neural mechanisms of consciousness, it will still not explain how the phenomenal aspects of experience arise. He means that scientific explanation will leave an unexplained gap. Levine asserts that psycho-physical phenomena such as consciousness are not fully reducible to one explanation. For example, the physical phenomenon heat is fully reducible to the movement of molecules, i.e. kinetic energy, while the psychological sensation of pain is not fully reducible to the firing of C-fibers. It is possible to think of situations where a pain is not induced by the sensation of C-fibers. There seems to be an explanatory gap argues Levine (1983). The sensation of pain is not fully explained by the firing of C-fibers since it is conceivable that the sensation of pain may be induced by other means. Think for example of an alien with no C-fibers. It is possible to think that this alien may have the sensation of pain even though it may not have any C-fibers. Levine (1983) says that we need an account of how to describe and understand a phenomenon, and rules that settle when the phenomenon is sufficiently explained. The Problem With the Problems Above The mind-body problem, the hard problem, and the explanatory gap do in principle ask the same question—‘How can a seemingly non-material thing such as conscious experience be explained as a purely physical phenomenon?’ Those problems may be seen to be of the greatest concern for a theory of consciousness, such as the IIT, to be able to explain. Dennett (2003) opposes Chalmers and claims that the hard problem doesn’t really exist. The hard problem is only a label that gives the impression of there being a real problem. The ‘the’ AN EVALUATION OF THE IIT !29 in ‘the hard problem’ creates an illusion of there being one problem. The hard problem, Dennett suggests, is just a “bag of tricks” containing all the easy problems neatly repackaged and relabeled. But Chalmers (1995) ensures that when all the easy problems are explained in full the hard problem will remain unexplained. That is why it is the hard problem, he says. Even though the hard problem may, as Dennett suggests, be fully explained by the easy problems. There still seem to be an explanatory gap in the identity between physical computation and the mental phenomenon of the subjective experience (Levine, 1983). If scientists could agree and come to a consensus on how matter gives rise to subjective experience, the problem would be solved. But science is not yet there and may not in a foreseeable future (if ever) get there, thus, the problem persists. Nagel (1974) suggests, we need something more than just an understanding of the word is to understand how something mental is something physical. The concept of a feeling such as hunger contains something more than just the causal power it has on the agent. It also contains the qualitative aspects of the feeling. No physical explanation will fully account for subjectivity. “The emergence of experience goes beyond what can be derived from physical theory." (Chalmers, 1995, p. 208). Hence, it seems that the hard problem makes consciousness impossible for scientific investigation (Cohen & Dennett, 2011). Consciousness, Dennett (2003) argues, is like real magic to many people, and that “Real magic, ... refers to the magic that is not real, while the magic that is real, that can actually be done, is not real magic.” (p. 7). Consciousness may perhaps then, Dennett suggests, play a trick on us and only appear to be hard to solve. AN EVALUATION OF THE IIT !30 The Binding Problem The binding problem wants an answer to how distributed neural activities of the brain temporarily connect to unify all the different aspects of an experience into a coherent whole. In such a way, that not only one tiny aspect is present in consciousness at a time. Or, that one would have a multitude of fragmented experiences at once (e.g. Crick & Koch, 1990; Revonsuo & Newman, 1999; Thagard & Stewart, 2014). The unity of consciousness has bothered philosophers and scientists ever since Kant (1781/2009) wrote about what he called synthesis (Revonsuo & Newman, 1999; Thagard & Stewart, 2014). Kant (1781/2009) describes the unified consciousness as the means by which we can relate to objects. This ability to relate to objects, Kant thought, is essential to the ability to relate to oneself. Revonsuo and Newman (1999) cite Crick and Koch (1990) as being the first to form a theory of consciousness around neurological binding explicitly. The binding problem may refer to many different types of binding. Treisman (1996) suggests that there are at least seven types of binding that produce a coherent visual experience. Treisman proposes that each type of binding may require different solutions. But essentially the varied information of single conceptual objects are stored in distributed fashion in the brain that temporarily need to connect to form a coherent experience. This temporary binding seems fundamental to consciousness, and its explanation may ultimately be the explanation for phenomenal consciousness. When we understand the mechanisms of binding, it is argued that we have gained valuable knowledge on how the brain generates subjective experience (Crick & Koch, 1990; Revonsuo & Newman, 1999). Chalmers (1995) disagrees and thinks that there may still be something missing. AN EVALUATION OF THE IIT !31 The Problem of Objectively Detecting Consciousness It has been claimed that it may be impossible to test objectively a system for consciousness in the sense of subjective experience because of its intrinsic nature (Blackmore, 2003). The scientific research of consciousness has until some decades back suffered from beliefs that humans alone are blessed with consciousness (Crick, 1994; Tononi & Koch, 2015). Some seem not to want to attribute consciousness to animals evolutionary much farther away than the great apes. While others stretch themselves to attribute some level of consciousness to their pet animals. It appears that we humans find it problematic to accept conscious behaviour in organisms very dissimilar to oneself such as birds, fish, and sand worms, despite their neurons function the same as human. The vast majority may probably say that human-made artefacts such as smartphones and supercomputers are entirely nonconscious, experience free (Tononi & Koch, 2015). The science of consciousness starts from the thing that we seem to know most intimately, our personal conscious experience. The healthy normal awake human brain has become the platinum standard of consciousness (Gamez, 2010) against which we evaluate levels of consciousness in humans and other organisms’ ability of consciousness. Perhaps the most intuitive test of consciousness we may do is often referred to (somewhat misleadingly) as the Turing test of consciousness, an analysis of behavioural correlates of consciousness (Blackmore, 2003). Variations of the Turing test, have been suggested as a possible way to test for consciousness, similar to the way that we theorise about other people’s minds. The original Turing test is called the imitation game, described by Alan M. Turing (1950) meant to shed new light on the dilemma of whether a machine can think. In this game, AN EVALUATION OF THE IIT !32 an interrogator interviews two participants, a man and a woman, via a text interface. The interrogator has then to decide the sex of the participants. Since the interrogator only can rely on the answers and how trustworthy the answers may be; the interrogator may guess the sex of the participants wrongly. What would then happen if one of the participants was secretly replaced by a computer? And what about if the interrogator mistaken the machine to be a human? Would that mean that the machine can think? This test has later on been transformed into a kind of intuitive test of consciousness. Variations of the test have excluded the text interface, letting the judge engage directly with, or passively watching the test object’s behaviour. From the observed behaviour, the judge is asked to decide intuitively whether the artefact shows signs of consciousness or not (Cotterill, 2003). Everything from the ability to participate in mundane dialogues to falling in love with the test object has been suggested as variations of the test (Syagin, Cicekli, & Akman, 2000). So far no artefact has successfully passed the Turing test (Tononi & Koch, 2015), despite widely reported claims to the contrary. Even though the Turing test is using human intuition to recognise consciousness, the test has been argued to not be sufficient to test for consciousness (Dennett, 1991; Blackmore, 2003). At best it can measure how well an agent can induce a user illusion of consciousness (Dennett, 1991). Many conscious artefacts and organisms may as well not pass the Turing test simply because they are too dissimilar to humans. In clinical settings the Glasgow Coma Scale (Teasdale & Jennett, 1974) is one of the most commonly used bedside methods for evaluation of consciousness in patients. When oral commands are not working, pain induced responses are used as a measure. Pain reactions for behavioural tests of consciousness have also been applied to, for example, fish, reports Thagard and Stewart (2014), to detect whether fish feel pain. Research suggest that fish perceive pain, which indicate that they may have a primary conscious AN EVALUATION OF THE IIT !33 experience of pain (Huntingford et al., 2006). Whether pain response is a measure of consciousness remains highly controversial (Chandroo, Yue, & Moccia, 2004). People born with congenital insensitivity to pain (Protheroe, 1991), a rare genetic disorder, would not pass. Another problem with pain response as a measure of consciousness is that it cannot be used with artefacts not designed or having the ability to evolve to respond to pain. If an artefact were designed to respond to pain, the pain response would presumingly not be a measure of consciousness. It would only be the pain response it was designed to give. Behaviours such as walking and talking do not either suffice for evaluation of consciousness. Mute amputees would not pass. It is not certain whether a person experience anything while sleepwalking and talking. She may, on the contrary, have vivid experiences while dreaming tranquilly asleep (Tononi & Koch, 2015). And few would today probably consider biped robots and speech synthesis to be sufficient to attribute machines consciousness. More advanced cognitive tests may perhaps be a better way to test for (human-like) consciousness. One test, Koch and Tononi (2008) suggest, would be to ask a system to comprehend a text and write a concise synopsis. An impossible task for a new born fully conscious baby. Another test of consciousness, they suggest, is to test the ability to recognise congruity and incongruity (Koch & Tononi, 2011). They argue that integration of information, i.e. consciousness due to IIT, is required to tell what e.g. is wrong with an image or a scenario. One simple version of this test is to take some photographs and cut away a strip in the middle of each picture. Then let the test subject to match the left side with the right side where the strip in the middle is masked out. Koch and Tononi argue that only conscious systems may be able to match the correct left side with the correct right side of the pictures, which means it can identify congruity and put them in the same context. Another test of the AN EVALUATION OF THE IIT !34 same capability they suggest would be to ask a system to identify what is wrong (incongruent) with pictures and scenarios, e.g. a picture of a person standing upside down in the ceiling in a room, or a robber using a sausage instead of a gun in a robbery scene. Koch and Tononi argue that any system that can perform this types of advanced cognitive tasks must integrate information and should, therefore, be attributed consciousness. This has later been contradicted by tests that have shown that subliminal exposure to incongruent scenarios may be processed unconsciously (Mudrik & Koch, 2013; Mudrik, Faivre, & Koch, 2014). A question that one should raise is whether what is thought to be unconscious processing really is completely non-conscious. Another method to distinguish incongruity from congruity may be to use hierarchical pattern recognition as described by Kurzweil (2012). Furthermore, the ability to identify incongruity is often absent while dreaming (A. Revonsuo, personal communication, August 25, 2015). Kurzweil suggests that the mind may be made up by simple pattern recognizers that bind simple pattern into more advance patterns and objects. Patterns that do not fit together are kept separated while patterns that fit together bind together. This selection process reduces the possible outcomes. For example the human mind can quite easily read text where only the first and the last letter in each word is in the correct position, while the other letters are scrambled (Grainger & Whitney, 2004). Hierarchical pattern recognising may be used to ‘guess’ the correct word. Hierarchical pattern recognition, Kurzweil (2012) suggests, may provide a mechanism that let consciousness arise. It is thought that various neuroimaging techniques may be used to detect consciousness. For example, functional magnetic resonance imaging (fMRI) can be used to get an indirect measure of neural activity within the brain by measuring the change in the oxidation level of hemoglobin caused by the increase of metabolic blood flow at the active AN EVALUATION OF THE IIT !35 area. The technique is called blood-oxygen-level dependent (BLOD) contrast imaging and is a standard fMRI method. In an fMRI study (Owen et al., 2006) a patient previously diagnosed vegetative state by standard clinical methods was observed to have similar neural responses as healthy controls. The patient was instructed to perform two different mental imagery tasks. One task was to imagine playing tennis, and the other task was to imagine walking through her house starting at the front door. During the tennis playing task the fMRI showed activity in the supplementary motor area, which was contrasted by her imagined walk where activity could be observed in the parahippocampal gyrus, the posterior parietal cortex, and the lateral premotor cortex. The same response was observed in 12 healthy controls. The researchers of the study concluded that the patient was clearly conscious of her surroundings and could act intentionally to verbal instructions. This result contradicts her previous vegetative state diagnosis. The combination of transcranial magnetic stimulation and high-density electroencephalography (TMS/hd-EEG) is suggested to discriminate levels of consciousness in humans. The perturbational complexity index (PCI) is such a method where the idea is to measure the capacity of the thalamocortical system ability to connect distributed regions of the cortex (Massimini, Boly, Casali, Rosanova, & Tononi, 2009). By sending a magnetic impulse through the skull with TMS the cortex is perturbed. Distinct measurable responses are expected depending on the level of the thalamocortical system’s ability to integrate cortical areas, i.e. the brain’s ability to integrate information. EEG is used to capture these responses from which the PCI can be calculated. The measurement has been shown to correlate with associated levels of consciousness of healthy subjects such as awake, AN EVALUATION OF THE IIT !36 anaesthetised, slow wave sleep, and REM sleep; and is thought to be generalizable to neurologically damaged patients (Casali et al., 2013). A different approach to evaluating the ability for consciousness is to look for computational correlates of consciousness, i.e., the mechanisms of information processing that are characteristic of consciousness (Cleeremans, 2005; Pinker, 1997). Problematic is that science does not yet know which those processes are, but some theories of consciousness are strongly suggestive (e.g., Tononi, 2012a; Baars, 2005; Kurzweil, 2012). The perturbational approach, mentioned above (Massimini et al., 2009) which is thought to measure indirectly cortex’s ability to integrate information is an example of applying a computational approach to neural systems. Computational theories of consciousness have otherwise mostly been applied in the making of assumed conscious artefacts (Franklin, 2003; Gamez, 2008a; Pinker, 1997). Further problematic is that computational theories of consciousness may show to be hard to falsify (Gamez, 2014a, 2014b). The problem follows from that there is yet no generally agreed objective definition of consciousness. Computational theories attempt to define consciousness objectively by giving identities, e.g. mechanism X is the same as consciousness, but this leads to circular definitions of consciousness. Methods need to be developed for validating theories of consciousness. Cleeremans (2005) suggests that the search of neural correlates of consciousness (Chalmers, 2000), behavioural correlates of consciousness, and computational correlates of consciousness to be joint. The three forms of correlates may in combination be helpful in the process of validating computational theories. Even though it may be testable if a computational theory of consciousness fully explains human-like consciousness; we cannot entirely rule out the possibility that other computational approaches, than what the human brain is using, may as AN EVALUATION OF THE IIT !37 well lead to conscious experience, i.e., there may be more than one type of mechanism that gives rise to consciousness. Axiomatic testing for consciousness is suggested as a possible way to test for consciousness in all types of systems, both neurological and non-neurological (Aleksander & Dunmall, 2003). In this context axioms are used in the classical way as self-evident premises, assumed to be generally true without need of further evidence; not the mathematical view where axioms aren’t self-evident but only make up the starting point on which a deductive system is formulated (Gärdenfors & Prawitz, n.d.). In axiomatic theories of consciousness, such as (Aleksander & Dunmall, 2003; Tononi, 2012a; Baars, 2005), the axioms are intuitively derived from introspection. The set of axioms is then seen as the minimal set of necessary properties a system may have for consciousness to arise. The task is now to test a given system for those criteria the axioms specify. If the system now appears to have the properties necessarily to fulfil the criteria, the system should be attributed consciousness (Clowes & Seth, 2008). Dennett (2013) is critical to how introspection is used for the purpose to identify the axioms and points out that it may be misguided by one’s own taste of explanation. He means that just because something may seem “smashingly obvious” (p. 417) does not necessarily mean it really is true. Hence, the axioms of such a theory are not immune to criticism. Evaluation of the IIT Against the Problems of Consciousness This thesis is an attempt to evaluate the IIT with respect to some central problems of consciousness and see what solutions and explanations the IIT may provide to them. The problems of consciousness that the theory is evaluated against are the mind-body problem, the hard problem, the explanatory gap, the binding problem, and the problem of objectively AN EVALUATION OF THE IIT !38 detecting consciousness. For each of the problems, I will give my interpretation on how the IIT solves it and then discuss the given explanation. The Mind-Body Problem vs. the IIT The mind-body problem is a problem of dualism that assumes that there is a division between the mind and the body. For the consideration of machine consciousness, the mindbody problem needs to have a solution that is implementable within a physical machine. If not the mind-body problem will prohibit any attempt for machine consciousness. Because of the intuitive separation between the mind and the body, a theory of consciousness need to explain either how they are connected or eliminate the problem altogether. The IIT claims that consciousness exists intrinsically. That consciousness is the same as integrated information, which is, due to the theory, the intrinsically subjective meaning of the state and the structure of matter, i.e. consciousness is a physical phenomenon. The IIT seems to explain consciousness as something that does not emerge out of the system. Instead, it emerges within the system. Consciousness is a physical property of the system. It is the matter, the physical substrate of the conscious experience which is experiencing. The IIT holds that there is no distinction between the bodily tissue in which the mind emerge and the ‘mind’. The IIT is quite clear about that not all bodily tissue (not even all brain tissue) is taking part in the physical substrate of consciousness. There is a distinction between the conscious physical substrate (the complex) and the rest of the body. They are though physically intimately intertwined. But, there is nothing about consciousness that is nonphysical, which makes the IIT a substance monistic theory. This interpretation of the IIT is somewhat controversial and oppose Koch’s (2012) idea that the theory postulates that the mental and the physical cannot be reduced to each other. Koch’s interpretation leads to a AN EVALUATION OF THE IIT !39 broader version of panpsychism—the idea that consciousness is everywhere—than Tononi readily fully agree to. The interpretation has in my view been misleading and caused the theory to be needlessly criticised by e.g. Searle (2013). Responding to Searle’s (2013) critique Koch and Tononi (2013) agree and say that consciousness emerge only when certain physical premises are fulfilled. Consciousness does not exist everywhere, but will emerge under the right physical conditions. The mental is reducible to the physical substrate, but the physical cannot be reduced to mental. It is, however, unclear, as Vimal (2013) points out, if the IIT is dual-aspect monism or not, i.e. that the mental and the physical are two different perspectives on the same substance. The IIT holds that information is represented as the state and structure of the matter and is only meaningful to the matter itself carrying the information. For the IIT to truly be a doubleaspect theory, I think, the information a conscious system holds should also be accessible to an external observer and possible—but probably enormously difficult—to interpret in a similar meaningful way. If Koch (2012) is right that the mental cannot be fully reduced to the physical, it would imply that the state and the structure of the matter cannot be interpreted and be understood by an observer, because that there is something missing. But if the mental is fully reducible to the physical it should be possible for an observer to interpret the state and structure and get an understanding of its meaning. Furthermore, if the state and the structure could be perfectly replicated, the replica would have an identical experience as the original. I interpret the IIT as being a double-aspect theory, i.e. that the informational content of an experience is accessible both from the subjective and objective perspective, with the restriction that the content will be experienced differently depending on the first-person vs. third-person perspective. The objective observer will not have the full blown being-in-themiddle-of subjective experience. AN EVALUATION OF THE IIT !40 By accepting the IIT as perspectival dualistic, I deny it to be a property dualistic theory. Which may be surprising as the theory holds that consciousness is a fundamental property. This brings the IIT interestingly closer to Searle’s view despite his harsh critique (Searle, 2013) against the theory. Searle describes himself to not being a property dualist and says that “that consciousness is a state the brain can be in” (Searle, 2002, p. 61). The critique Searle raised was based on Koch’s description of the theory (Koch, 2012). Searle gets the understanding from Koch’s description that the theory is a property dualistic theory. The problem arises as both Tononi and Koch use (which I undeniably do as well in this thesis) a problematic vocabulary which describe consciousness as something that arise, emerge, and is over and above. The vocabulary makes it sound as consciousness is something non-physical and immaterial. As Searle (2002) points out, property dualism seems to camouflage a type of substance dualism. That is the reason I repeatedly in this thesis tried to make it clear that consciousness as an intrinsic property emerges within, not emerges out from. The IIT is quite clear that consciousness is nothing wishy-washy supernatural mystic that is floating above in a different realm. It is fully reducible to the state and structure of the matter which is having the experience. The Hard Problem vs. the IIT The hard problem is to explain how pure matter can give rise to something seemingly immaterial such as subjective experience. Chalmers write that “even when we have explained the performance of all the cognitive and behavioural functions in the vicinity of experience ... there may still remain a further unanswered question” (1995, p. 203). It is essential for a theory of consciousness to provide a convincing solution to the hard problem or able to eliminate it. Otherwise, no one will believe that the theory explains subjective experience. AN EVALUATION OF THE IIT !41 The IIT explains consciousness as the state and the structure of matter. It is the specific state and structure, i.e. the mechanism, which causes consciousness. For consciousness to exists the mechanism needs to fulfil the five postulates of the IIT, mentioned above. Consciousness exists intrinsically within the matter at the spatiotemporal level where the information is physically represented in the system. Consciousness is, due to the IIT, integrated information that is the same as the state and the structure of the matter that is having the experience. Dehaene (2014) points out that the hard problem seems hard just because it engages ill-defined intuitions. A common mistake is to understand intuitively consciousness as something immaterial that arises out from the matter, which the conscious being then is observing. It is not helpful when Tononi writes the quantity of consciousness is the amount of integrated information generated “above and beyond its parts” (2008, p. 224). It is easy to fool oneself to think that he means that consciousness is something immaterial that arise out from the matter. What Tononi most likely refers to is the synergy effect within, i.e. that the grand total is greater than the simple sum of the parts. If consciousness would be separate from the matter then who is the one experiencing? The IIT postulates that it is the matter, the physical substrate of the conscious experience, which is experiencing. It is something like to be the mechanism that generates integrated information. Consciousness is the intrinsic state and structure of that mechanism. “If mental processes are indeed physical processes, then there is something it is like, intrinsically, to undergo certain physical processes” (Nagel, 1974, p. 445-6). There is no further question to answer, as Chalmers (1995) thinks there is. Dennett says that the hard problem is hard to solve because there is simply no hard problem to solve. When we have solved all the easy problems, there will be nothing more to explain about consciousness (Dennett, 2003, 2013). The hard problem was created by our AN EVALUATION OF THE IIT !42 intuitions and will vanish as cognitive neuroscience advances (Dehaene, 2014). I tend to agree. The Explanatory Gap vs. the IIT The IIT claims that consciousness is identical to the integrated information that is the state and structure of matter that fulfills the postulates of the IIT. In other words, the IIT postulates an identity between mind and matter. Such mind-matter identities, Levine argues, leaves explanatory gaps (1983). Levine’s solution is to eliminate the first-person view of consciousness altogether by denying the existence of qualia and subjective experience (Levine, 1983). By doing so, he evaporates the existence of consciousness and makes all of us into experienceless neurobiological zombies (Searle, 2002, p. 61; Revonsuo, 2010). Which does not solve the problem only denies the only certain thing there is, the existence of consciousness. Not even Dennett (1991, 2013) who claims that consciousness is an illusion denies the rich inner world experience of consciousness. Instead, he says that a theory of consciousness must be able to explain how an observer experiences informational representations in mechanistic terms. That a theory of consciousness is ‘doomed’ if it fails to link intimately the neural correlates of experience and cognition (Cohen & Dennett, 2011). This despite Chalmers’ certainty that a reductive approach cannot fully account for subjective experience (Chalmers, 1995). Nagel (1974) proposes that we need a better understanding of what an identity between the subjective and objective really means. We also need “an account of what it is for a phenomenon to be made intelligible, along with rules which determine when the demand for further intelligibility is inappropriate,” suggests Levine (1983, p. 358). The IIT states that consciousness exists intrinsically (Tononi & Koch, 2015). This follows Searle closely who says that ‘consciousness’ denotes a state that a system may be in AN EVALUATION OF THE IIT !43 which has the ontology of first-person (Searle, 2002). The scope of consciousness is defined in the IIT by the spatiotemporal grain size. The spatiotemporal grain encapsulates the experiencing complex—the physical substrate of the conscious experience—in spatial and temporal dimensions. The physical substrate is the mechanism that consists of the subset of elements which state and structure generates most integrated information, the highest synergy effect, among the system’s power set of elements. The IIT provides with rules that identify the physical level at which a conscious mechanism exists and hence, I believe, answers to Levine’s request. It is this conscious mechanism that holds the experience that is the experiencer. The state and structure of the same mechanism are the experiences, i.e. the integrated information, that is experienced (Oizumi et al., 2014c). The abstraction level is defined in fully by the spatiotemporal grain, and no gap remains that need further explanation. It may be relatively easy to determine the size of the spatiotemporal grain in artificial systems of which one have the full insight. It remains unclear, though, how to define the borders of the spatiotemporal grain in natural systems, such as a human brain. The Binding Problem vs. the IIT The binding problem essentially refers to how broadly distributed cortical neurons, representing different aspects of consciousness, simultaneously connect and produce the experience of a coherent whole (Crick & Koch, 1990; Revonsuo & Newman, 1999). It has been suggested (e.g. Revonsuo, 2010) that a top-down approach to solving this problem may expose the underlying mechanism of this phenomena. Treisman (1996) raise a flag and prompt that there may be more than one solution to this problem depending on the nature of the binding. Chalmers (1995) suggest a bottom-up approach and say that we only need to find AN EVALUATION OF THE IIT !44 the mechanisms that integrate the information content of consciousness to explain how the binding occurs. The IIT takes, as mentioned earlier, a top-down-back-up approach, identifying phenomenological axioms from which ontological postulates are derived. The ontological postulates define a mechanism that is thought to support the generation of integrated information. Integrated information is by definition already bound which means that, if the IIT is correct, no further explanation is required to solve the binding problem. Block (1995) raises the question of how an explanation to the binding problem may explain what it is like to experience. The answer the IIT is giving is that it is the set of elements that constitute the mechanism that on itself is having the fully integrated experience. This differs with Chalmers (1995) who writes that the mechanism binds the information that is then later exploited by other processes. He seems to suggest an external observer process that is different from the binding process. The IIT claims that it is the binding mechanism that is the observer of the intrinsic information it holds. It is the state and the structure of the mechanisms that forms the integrated information of the subjective experience. Furthermore, the IIT suggests only one universal mechanism that is capable of binding all sorts of information together. This is in contrast to Treisman (1996) who suggest that different types of binding may require different types of mechanisms. The basic mechanism of IIT is universal due to the IIT, which predicts that integrated information is intrinsic to the mechanism itself. Without the mechanism, the information—full of meaning —would not exist. AN EVALUATION OF THE IIT !45 The Problem of Objectively Detecting Consciousness The postulates of the IIT defines requirements of what a conscious system need to fulfil to be conscious. From this, the IIT suggests a minimally conscious system consisting of only two elements (Oizumi et al., 2014a). The problem that arises is to validate scientifically whether such a system is conscious. Dehaene (2014) proclaim that “any theory of consciousness must face the ultimate test: the clinic” (p. 200). But traditional tests of consciousness, such as the clinically used Glasgow coma scale (Teasdale & Jennett, 1974), do not suffice and can not be used for such systems that differ so radically from humans. We can also not use the simplest test of them all, subjective report, by asking the agent if it has a subjective experience. There are no means how to put the question in any meaningful way to such a simple system, and the system do not have the ability to report back. Any variation of the Turing test does also not work since we do not know how to recognise consciousness in such a system (Blackmore, 2003). Even if a humanoid robot was built which responds to its surroundings, giving the impression as if it were conscious, fulfilling all the structural criteria of the IIT to be conscious one would still not be able to tell if the robot had an inner life of its own, and subjectively experiencing the world (Dennett, 2013). A third-person perspective of external behaviour can never suffice to verify whether another agent is conscious or not. At best, I argue, such a test may evaluate if the ‘judge’ has the ability of theory of mind, i.e. the ability to attribute mental states to another agent and feel empathy towards it. I believe that the judge in a Turing test is deceived about its role, while being the actual subject of the test. We need to find a way to try to falsify the theory that’s being tested and see if it turns out false. The use of axiomatic testing, as Aleksander and Dunmall suggest (2003), is not sufficient to verify whether a system is minimally conscious due to the IIT. An extremely AN EVALUATION OF THE IIT !46 simple system, consisting of only two elements, presumed to be minimally conscious, does not have the ability to report on its own phenomenology. While a more advanced system may simply give false reports on its perhaps non-existent experiences. Another problem also arises. A conscious system should (if the IIT is valid) be verified against the phenomenological axioms of the IIT. We do not yet know if IIT is valid and can, hence, not use its axioms to test for consciousness. We become trapped in circular reasoning. We are not even certain if the axioms are valid, as acknowledged by Tononi and Koch (2015). It is also not possible to validate the IIT by testing a system based on its principles against the axioms of another yet not validated or fully accepted theory of consciousness. The method would be flawed if faulty axioms were used. It seems to be as difficult to prove that a system is lacking consciousness as to prove that it is conscious. To test a theory of consciousness, it is not enough to look at just one type of correlate. A combination of different approaches, as Cleeremans (2005) suggests, seems to be needed to verify whether a theory of consciousness may be valid or not. Cleeremans proposes a joint search of behavioural, computational, and neural correlates of consciousness. A whole chain of interlinked correlates has to be found starting from one’s own subjective experience. This due to the intrinsic subjective nature of consciousness. I cannot search for a conscious mechanism in a bat simply because I do not have the slightest idea of what it is like to be a bat. I cannot tell the difference between a bat’s conscious behaviour and its unconscious, reflexive behaviour. At best I can only give an educated guess. It’s hard enough to tell the difference in the behaviour of another human being. The test subject where the search has to begin needs to be physically and behaviourally almost identical to oneself. Hence, only other human beings meet the criteria. AN EVALUATION OF THE IIT !47 I believe that the mechanism of consciousness has first to be discovered in human to be rightfully testable. Ultimately the mechanism of consciousness has to be detected within the only object which one can be certain to be conscious oneself. With the downside that the objectiveness needed to make the findings scientifically trustworthy would not be satisfied. The advantage is that other humans may sufficiently replace oneself with the assumption that they also are conscious. Another gain with humans as test subjects is that they can give subjective reports on their experiences, therefore, be able to report whether a stimulus is present in consciousness or not. By performing the search of the conscious mechanism in humans, fundamental assumptions can be made which allow between-species matching of behavioural correlates in animal models to be omitted. There is a limit, though. Only phenomenal conscious states of which the subject can reliably report on are testable. Still it means that the pursuit may proceed of computational correlates of consciousness, i.e. activity patterns of the neural mechanism of consciousness (Cleeremans, 2005; Gamez, 2014b). Those patterns may be captured by various techniques. Electroencephalograph (EEG) may be sufficient for capturing patterns of neuroelectrical fields over widely distributed cortical areas while intracellular single-unit recordings do it at the cell level. Functional magnetic resonance imaging (fMRI) or the more portable near-infrared spectroscopy (NIRS) (Muehlemann, Haensse, & Wolf, 2008) may be used to measure activity in cortical areas. It is today widely believed that conscious experience may emerge at the level of the cerebral cortex. That is perhaps where the conscious mechanism is to be found. A problem though is that the cerebral cortex consists of 16 billion neurons (Herculano-Houzel, 2009). The task to distinguish the neurons of the conscious mechanism from the neurons of unconscious processing seems daunting. AN EVALUATION OF THE IIT !48 The IIT provides with a model of a conscious mechanism (as described earlier in this thesis) and algorithms to measure the degree of consciousness as integrated information, Φ. If one could find an activity pattern of a mechanism within the human brain of which a positive Φ could be calculated and which strongly correlate positively with subjective report of conscious experience, one would perhaps be on the way to find support for the theory. It should be a straight forward process to test the theory since the only thing one need to do is to measure the state of each element and calculate Φ. If Φ positively correlates with the subjective experience, one would have found support for the theory. The opposite, a negative correlation between Φ and subjective experience indicate that the theory may be false and should be considered for rejection. The same would be if every attempt to calculate Φ returns Φ > 0; then there must be something wrong with the model. A problem that appears, though, is that the amount of mathematical calculations needed to calculate Φ grow exponentially with the number of elements the system consist of, which makes Φ for systems larger than a couple of dozen of elements almost impossible to calculate (Tononi & Sporns, 2003). There is simply not enough computer power to perform the calculation. To calculate Φ for a system consisting of 18’000 element could take up to 109000 years (Gamez, 2008b). Even after extreme optimization of the algorithms—by identifying and ignoring small subset with presumably small Φ—it took Gamez a full month of computing power to analyse an artificial neural network of 17’544 neurons (Gamez, 2010). The perturbational complexity index (PCI) (Casali et al., 2013; Massimini et al., 2009) which seems to correlate with levels of human consciousness is based on the ideas of the IIT. The calculations of PCI are much faster to perform than that of Φ. Unfortunately, it cannot be used as evidence to support the IIT. The PCI is not a measure of the conscious experience. The PCI is a measure of conscious ability, or more precise the capability of AN EVALUATION OF THE IIT !49 information integration, not consciousness per se. The PCI only give indirect support of the theory and can not be used as direct evidence in favour of the IIT. There is a promising development of Φ called Φ* (Oizumi et al., 2015). Φ* is presumed to be a practical measurement of integrated information. The math of Φ* supposedly follows the theoretical requirements of the IIT but introduces mismatched decoding (Merhav, Kaplan, Lapidoth, & Shitz, 1994) and Gaussian assumption (Barrett & Seth, 2011) which makes the measure practically applicable. Tsuchiya (2015) and colleagues are working on empirically testing the IIT directly. They are using physiological data collected with multichannel electrocorticography recording from the fusiform face area (FFA) in the human brain. The data represents the neural activity of the different states when being consciously aware of a face, and not being consciously aware of a face. Continuous flash suppression (CFS) (Tsuchiya & Koch, 2005) and backwards masking were used to dissociate stimulus inputs from subjective perception of a face. In CFS, a face was presented to one eye and the Mondrian-like noise pattern to the other eye. By adjusting the contrast of the face, they induced trials with or without visibility of faces with identical stimulus input (N. Tsuchiya, personal communication, August 22, 2015). In backwards masking, stimuli were presented at durations both above and below the subliminal threshold, i.e. the time length it takes to become consciously aware of the content in a picture. The idea of the experiment is to look for the qualia of seeing a face in the FFA. If the measured patterns of generated integrated information, Φ, correlates with the stimuli it is belied to give direct support to the IIT. Tsuchiya expresses his reservations “We don’t know if this interaction is corresponding to seeing a face itself. It is plausible.” There is today still no perfect experiment to test the validity of the IIT, but the research is advancing and getting closer. If we will reach there in a foreseeable time remains AN EVALUATION OF THE IIT !50 an open question. At least the IIT seems to in principle be testable. The future will have to show whether the theory will be kept or rejected. There is though some indirect evidence, e.g. the PCI (Casali et al., 2013), which gives promising support to the theory. As long as the theory is not rejected, I would argue, that it may be used to test for consciousness. Preferably in combination with some other method of choice. Φ and its derivatives are, to my knowledge, the only objective measures of phenomenal consciousness which do not rely on subjective report. To measure Φ, or any closely related measure derived from it may be as a good assessment of consciousness as any other method in clinical use today, such as the Glasgow coma scale. Still though with the reservation that the theory may be false. Discussion I have argued in this thesis that the IIT holds satisfying explanations to the central problems of consciousness: the mind-body problem, the hard problem, the explanatory gap, the binding problem, and the problem of detecting consciousness. Except the latter they have not previously, to my knowledge, been discussed at length in the published literature on the IIT. As long as the theory is not falsified, I believe, it may be used to create artifacts that can be attributed phenomenal consciousness, at least due to the IIT. I need though raise awareness about differentiating viewpoints that may disagree with my opinion on the subject. There is the question if the IIT is a theory of phenomenal consciousness. Block (1995, p. 236) gives numerous examples of theories of consciousness that sets out to be theories of phenomenal consciousness but turn out to really be theories of access consciousness, selfconsciousness, monitoring-consciousness, etc. Has the same mistake been made with IIT? Bitbol (2014) for example is of the opinion that the IIT “provides absolutely no clue about AN EVALUATION OF THE IIT !51 the origin of phenomenal consciousness. They have explained the functions of consciousness” (p. 266). I believe that he is in direct conflict with the IIT. The IIT say very little about the functions of consciousness, but has a focus on explaining the origin of phenomenal conscious. Central to the explanations is my acceptance of perspectival dualism. I have argued in this thesis that consciousness exists within, and that matter and mind are the same, but seen from the objective third-person extrinsic point of view and the subjective first-person intrinsic point of view. If one denies perspectival dualism and does not buy into substance monism the mind-body problem, the hard problem and the explanatory gap will all firmly remain (Parthemore, 2011). By understanding that consciousness, as the IIT describes it, is intrinsic and emerges within the conscious mechanism, I find the IIT to give a convincing explanation of phenomenal consciousness. My guess is that Bitbol has not understood the fundamental implications of the theory, that integrated information is necessarily subjective and meaningful in its nature, i.e. the same as phenomenal consciousness. Some may argue that the phenomenal axioms which the theory builds on are wrong and not self-evident truths that “cannot be doubted and do not need proof” (Oizumi et al., 2014a, p. 2; Tononi & Koch, 2015). It could then be argued that the postulates derived from the axioms also must be wrong. Implying that the mechanism the IIT predicts is wrong and does not give rise to subjective experience. To my knowledge, there is yet no competent attempt to dismantle and disproof the axioms without falling in the trap of describing functional consciousness instead of phenomenal consciousness (Aleksander & Dunmall, 2003; Clowes & Seth, 2008). A closer look at the axioms would, though, be a good place to start at to further evaluating the theory. AN EVALUATION OF THE IIT !52 There are some phenomenal aspects of consciousness which I cannot find any explanations to in the literature of the IIT. One such aspect is attention and the selection process of attention. Why are some things included within the conscious experience while others stay conveniently outside? How may one, for example, attain a focus on the task at hand while not being conscious of the sensations of the clothes on the skin? And how can one selectively ignore the itch on the skin but not ignore one’s vision? I do not find the IIT literature to give any explanations to this selection process. My guess is though that the Φvalue of different concepts, qualia, may play a role. A stronger Φ over the mechanism of a concept may brighten up and bring the concept into focus. As well as being a theory of consciousness, I find the IIT also to be a theory of subjective information, extending traditional objective information theory by adding an explanation of meaning. This should be of interest to artificial intelligence (AI) research as well as consciousness science. A major problem with AI is to get computers to comprehend and add meaning to the data they are processing. Perhaps the IIT as an information theory may bring some light on this problem. Even if the IIT turns out to be the theory of consciousness explaining phenomenal consciousness, there is still one hinder why humans may not attribute consciousness to an artifact in which the theory is implemented. The first conscious artifacts will be very different from humans. Still, humans seem to think that a conscious artefact should experience the world the same way as humans do. Human consciousness has unfortunately become the platinum standard of consciousness (Gamez, 2010), perhaps especially unfortunate for machines. My guess is that humans will mistakingly attribute consciousness to nonconscious machines that behave humanlike, and stubbornly refuse to attribute consciousness to a grey box with a rich inner life of its own. We humans simply forget that we cannot imagine what it AN EVALUATION OF THE IIT !53 is like to be a bat (Nagel, 1974). How can we then imagine what it is like to be a grey box, hardwired to subjective experience? Conclusion The integrated information theory of consciousness does provide, in my opinion, somewhat satisfying explanations to the central problems of consciousness: the mind-body problem, the hard problem, the explanatory gap, the binding problem, and the problem of objectively detecting consciousness. The explanations are, though, questionable, and no final conclusion has been possible to reach. There is, to the author known, no conclusive evidence against the theory, but there is some promising indirect evidence that give the theory support. It is, nonetheless, possible to falsify the theory by showing a negative correlation between Φ and corresponding subjective experience. 
 AN EVALUATION OF THE IIT!54 ReferencesAleksander, I., & Dunmall, B. (2003). Axioms and tests for the presence of minimalconsciousness in agents. Journal of Consciousness Studies, 10(4-5), 7–18.Baars, B. J. (2005). Global workspace theory of consciousness: Toward a cognitiveneuroscience of human experience? Progress in Brain Research, 150, 45–53.
doi:10.1016/S0079-6123(05)50004-9Barrett, A. B., & Seth, A. K. (2011). Practical measures of integrated information for time-series data. PLoS Comput Biol, 7(1), e1001052.Bateson, G. (1972). Steps to an ecology of mind: Collected essays in anthropology,psychiatry, evolution and epistemology. Northvale, NJ: Jason Aronson Inc.Bitbol, M. (2014). On the primary nature of consciousness. In F. Capra & P.-L. Luisi (Ed.)(2014), The systems view of life: a unifying vision, (pp. 266–268), Cambridge,England: Cambridge University Press.Blackmore, S. (2003). Consciousness in meme machines. In O. Holland (Ed.), Machineconsciousness, (pp. 47–66). Exeter, England: Imprint Academic.Block, N. (1995). On a confusion about a function of consciousness. Behavioral and BrainSciences 18(2), 227–287. doi:10.1017/S0140525X00038188Casali, A. G., Gosseries, O., Rosanova, M., Boly, M., Sarasso, S., Casali, K. R., ... &Massimini, M. (2013). A theoretically based index of consciousness independent ofsensory processing and behavior. Science Translational Medicine, 5(198ra105), 1–10.doi:10.1126/scitranslmed.3006294Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of ConsciousnessStudies, 2, 200–219.Chalmers, D. J. (1996). The conscious mind. New York, NY: Oxford University Press. AN EVALUATION OF THE IIT!55 Chalmers, D. J. (2000). What is a neural correlate of consciousness? In T. Metzinger (Ed.),Neural correlates of consciousness: empirical and conceptual questions, (pp. 17–40).Cambridge, MA: MIT Press.Chandroo, K., Yue, S., & Moccia, R. (2004). An evaluation of current perspectives onconsciousness and pain in fish. Fish and Fisheries, 5, 281–295. 
doi:10.1111/j.1467-2679.2004.00163.xCleeremans, A. (2005). Computational correlates of consciousness. Progress in BrainResearch, 150, 81–89. doi:10.1016/S0079-6123(05)50007-4Clowes, R. W., & Seth, A. K. (2008). Axioms, properties and criteria: Roles for synthesis inthe science of consciousness. Artificial Intelligence in Medicine, 44, 91–104.
doi:10.1016/j.artmed.2008.07.009Cohen, M. A., & Dennett, D. C. (2011). Consciousness cannot be separated from function.Trends in Cognitive Sciences, 15, 358–364. doi:10.1016/j.tics.2011.06.008Cotterill, R. M. J. (2003). CyberChild: A simulation test-bed for consciousness studies.Journal of Consciousness Studies, 10(4-5), 31–45.Crick, F. (1994). The astonishing hypothesis: The scientific search for the soul. New York,NY: Touchstone.Crick, F., & Koch, C. (1990). Towards a neurobiological theory of consciousness. InSeminars in the neurosciences (Vol. 2, pp. 263–275). Saunders Scientific Publications.Dehaene, S. (2014). Consciousness and the brain: Deciphering how the brain codes ourthoughts. New York, NY: Viking Penguin.Dennett, D. C. (1991). Consciousness explained. New York, NY: Penguin Books.Dennett, D. C. (2003). Explaining the “magic” of consciousness. Journal of Cultural andEvolutionary Psychology, 1, 7–19. doi:10.1556/JCEP.1.2003.1.2 AN EVALUATION OF THE IIT!56 Dennett, D. C. (2013). Intuition pumps and other tools for thinking. New York, NY: W. W.Norton & Company.Descartes, R. (2008). Meditations on first philosophy: With selections from the objections andreplies: Translated with an introduction and notes by Michael Moriarty [Electronicresource]. Oxford, England: Oxford University Press. Retrieved at http://search.ebscohost.com/login.aspx?direct=true&scope;=site&db;=nlebk&db;=nlabk&AN;=257845 (Original workpublished 1641)Dualism. (n.d.). In The American Heritage® New Dictionary of Cultural Literacy, ThirdEdition. Retrieved from http://dictionary.reference.com/browse/dualismFerrarelli, F., Massimini, M., Sarasso, S., Casali, A., Riedner, B. A., Angelini, G., ... &Pearce, R. A. (2010). Breakdown in cortical effective connectivity during midazolam-induced loss of consciousness. Proceedings of the National Academy of Sciences, 107,2681-2686. doi:10.1073/pnas.0913008107Franklin, S. (2003). IDA: a conscious artefact? In O. Holland (Ed.), Machine consciousness(pp. 47–66). Exeter, England: Imprint Academic.Gamez, D. (2008a). Progress in machine consciousness. Consciousness and Cognition, 17,887-910. doi:10.1016/j.concog.2007.04.005Gamez, D. (2008b). The development and analysis of conscious machines (Doctoraldissertation, University of Essex). Retrieved from http://www.davidgamez.eu/mc-thesis/thesis/Gamez08_MC-Thesis.pdfGamez, D. (2010). Information integration based predictions about the conscious states of aspiking neural network. Consciousness and Cognition, 19(1), 294–310. 
doi:10.1016/j.concog.2009.11.001 AN EVALUATION OF THE IIT!57 Gamez, D. (2014a). Are information or data patterns correlated with consciousness? Topoi,1-15. doi:10.1007/s11245-014-9246-7Gamez, D. (2014b). Can we prove that there are computational correlates of consciousness inthe brain? Journal of Cognitive Science, 15(2), 149–186.Gardelle, V. D., Kouider, S. (2009). Cognitive theories of consciousness. In W. P. Banks(Ed.), Encyclopedia of consciousness, volume 1, A-L, (pp. 135–146). Oxford,England: Academic Press.Gärdenfors, P., & Prawitz, D. (n.d.). Axiom. In Nationalencyklopedin. Retrieved from http://www.ne.se/uppslagsverk/encyklopedi/lång/axiomGazzaniga, M. S. (2005). Forty-five years of split-brain research and still going strong.Nature Reviews Neuroscience, 6, 653-659. doi:10.1038/nrn1723Glickstein, M. (2007). What does the cerebellum really do? Current Biology, 17, R824–R827.doi:10.1016/j.cub.2007.08.009Grainger, J., & Whitney, C. (2004). Does the huamn mnid raed wrods as a wlohe? Trends inCognitive Sciences, 8(2), 58-59. doi: 10.1016/j.tics.2003.11.006Graziano, M. S. A., & Webb, T. W. (2014). A mechanistic theory of consciousness.International Journal of Machine Consciousness, 6(2), 1–14. 
doi:10.1142/S1793843014001316Herculano-Houzel, S. (2009). The human brain in numbers: A linearly scaled-up primatebrain. Frontiers in Human Neuroscience, 3(31), 1–11. 
doi:10.3389/neuro.09.031.2009Hermerén, G. (n.d.). Dualism. In Nationalencyklopedin. Retrieved from http://www.ne.se/uppslagsverk/encyklopedi/lång/dualism-(1-filosofi) AN EVALUATION OF THE IIT!58 Huntingford, F., Adams, C., Braithwaite, V. A., Kadri, S., Pottinger, T. G., Sandøe, P., &Turnbull, J. F. (2006). Current issues in fish welfare. Journal of fish biology, 68(2),332–372. doi:10.1111/j.0022-1112.2006.001046.xKant, I. (2009). Critique of pure reason [electronic resource] translated by J.M.D.Meiklejohn. Waiheke Island: Floating Press. Retrieved from http://search.ebscohost.com/login.aspx?direct=true&scope;=site&db;=nlebk&db;=nlabk&AN;=313757 (Original workpublished 1781)Koch, C. (2012). Consciousness: Confessions of a romantic reductionist. Cambridge, MA:The MIT Press.Koch, C., & Tononi, G. (2008). Can machines be conscious? Spectrum, IEEE, 45(6), 55–59.doi:10.1109/MSPEC.2008.4531463Koch, C., & Tononi, G. (2011). A test for consciousness. Scientific American, 304(6), 44–47.doi:10.1038/scientificamerican0611-44Koch, C., & Tononi, G. (2013). Can a photodiode be conscious? The New York Review ofBooks. Retrieved from http://www.nybooks.com/articles/archives/2013/mar/07/can-photodiode-be-conscious/Kurzweil, R. (2012). How to create a mind: The secret of human thought revealed. London,England: Penguin books.Levine, J. (1983). Materialism and qualia: The explanatory gap. Pacific PhilosophicalQuarterly, 64, 354–361. Retrieved from http://www.uoguelph.ca/~abailey/Resources/levine.pdfMalle, B. F. (2009). Folk theories of consciousness. In W. P. Banks (Ed.), Encyclopedia ofConsciousness, Volume 1, A-L, (pp. 251–263). Oxford, England: Academic Press. AN EVALUATION OF THE IIT!59 Massimini, M., Boly, M., Casali, A., Rosanova, M., & Tononi, G. (2009). A perturbationalapproach for evaluating the brain’s capacity for consciousness. Progress in BrainResearch, 177, 201–214. doi:10.1016/S0079-6123(09)17714-2Massimini, M., Ferrarelli, F., Huber, R., Esser, S. K., Singh, H., & Tononi, G. (2005).Breakdown of cortical effective connectivity during sleep. Science, 309, 2228–2232.doi:10.1126/science.1117256Merhav, N., Kaplan, G., Lapidoth, A., & Shitz, S. S. (1994). On information rates formismatched decoders. Information Theory, IEEE Transactions on, 40(6), 1953-1967.Mudrik, L., Faivre, N., & Koch, C. (2014). Information integration without awareness.Trends in Cognitive Sciences, 18(9), 488–496. doi:10.1016/j.tics.2014.04.009Mudrik, L., & Koch, C. (2013). Differential processing of invisible congruent andincongruent scenes: a case for unconscious integration. Journal of Vision, 13(13):24,1–14. doi:10.1167/13.13.24Muehlemann, T., Haensse, D., & Wolf, M. (2008). Wireless miniaturized in-vivo nearinfrared imaging. Optics Express, 16, 10323-10330. doi: 10.1364/OE.16.010323Nagel, T. (1974). What is it like to be a bat? The Philosophical Review, 83, 435–450.
doi:10.2307/2183914Oizumi, M., Albantakis, L., & Tononi, G. (2014a). From the phenomenology to themechanisms of consciousness: Integrated information theory 3.0. PLoSComputational Biology, 10(5), 1–25. doi:10.1371/journal.pcbi.1003588Oizumi, M., Albantakis, L., & Tononi, G. (2014b). Main differences between IIT 3.0 andearlier versions. PLoS Computational Biology, 10(5), 1–2. 
doi:10.1371/journal.pcbi.1003588.s002 AN EVALUATION OF THE IIT!60 Oizumi, M., Albantakis, L., & Tononi, G. (2014c). Some differences between integratedinformation and Shannon information. PLoS Computational Biology, 10(5), 1–2. 
doi:10.1371/journal.pcbi.1003588.s004Oizumi, M., Amari, S. I., Yanagawa, T., Fujii, N., & Tsuchiya, N. (2015). Measuringintegrated information from the decoding perspective. arXiv preprint arXiv:1505.04368.Owen, A. M., Coleman, M. R., Boly, M., Davis, M. H., Laureys, S., & Pickard, J. D. (2006).Detecting awareness in the vegetive state. Science, 313(5792), 1402. 
doi:10.1126/science.1130197Parthemore, J. (2011). The limits of concepts and conceptual abilities (Ch. 5). In Parthemore,J., Concepts enacted: confronting the obstacles and paradoxes inherent in pursuing ascientific understanding of the building blocks of human thought, unpublisheddoctoral thesis, Falmer, Brighton, UK: University of Sussex. Retrieved from http://sro.sussex.ac.uk/6954/1/Parthemore,_Joel.pdfParthemore, J., & Whitby, B. (2014). Moral agency, moral responsibility, and artifacts: Whatexisting artifacts fail to achieve (and why), and why they, nevertheless, can (and do!)make moral claims upon us. International Journal of Machine Consciousness, 6, 141–161. doi:10.1142/S1793843014400162Pinker, S. (1997). How the mind works. London: Penguin Books.Protheroe, S. M. (1991). Congenital insensitivity to pain. Journal of the Royal Society ofMedicine, 84, 558–559. Retrieved from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1293421/ AN EVALUATION OF THE IIT!61 Revonsuo, A. (2009). Altered and exceptional states of consciousness. In W. P. Banks (Ed.),Encyclopedia of Consciousness, Volume 1, A-L, (pp. 9–22). Oxford, England:Academic Press.Revonsuo, A. (2010). Consciousness: The science of subjectivity. Hove, England: PsychologyPress.Revonsuo, A., & Newman, J. (1999). Binding and consciousness. Consciousness andcognition, 8, 123–127. doi: 10.1006/ccog.1999.0393Robinson, H. (2011). Dualism. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy(Winter 2012 ed.). Retrieved from http://plato.stanford.edu/archives/win2012/entries/dualismSearle, J. R. (1990). Who is computing with the brain? Behavioral and BrainSciences, 13, 632–642. doi:10.1017/S0140525X00080663Searle, J. R. (2002). Why I am not a property dualist. Journal of Consciousness Studies,9(12), 57–64. Retrieved from http://imprint.co.uk/pdf/searle-final.pdfSearle, J. R. (2013, January 10). Can information theory explain consciousness? The NewYork Review of Books. Retrieved from http://www.nybooks.com/articles/archives/2013/jan/10/can-information-theory-explain-consciousness/Shannon, C. E. (1948). A mathematical theory of communication. The Bell System TechnicalJournal, 27, 379–423. doi:10.1145/584091.584093Stubenberg, L. (2010). Neutral monism. In E. N. Zalta (Ed.), The Stanford Encyclopedia ofPhilosophy (Fall 2014 Edition). Retrieved from http://plato.stanford.edu/archives/fall2014/entries/neutral-monism/Syagin, A., Cicekli, I., & Akman, V. (2000). Turing test: 50 years later. Minds and Machines,10, 463–518. doi:10.1023/A:1011288000451 AN EVALUATION OF THE IIT!62 Teasdale, G., & Jennett, B. (1974). Assessment of coma and impaired consciousness apractical scale. The Lancet, 304, 81–84. doi:10.1016/S0140-6736(74)91639-0Thagard, P., & Stewart, T. C. (2014). Two theories of consciousness: semantic pointercompetition vs. information integration. Consciousness and Cognition, 30, 73–90.doi:10.1016/j.concog.2014.07.001Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience,5(42), 1–22. doi:10.1186/1471-2202-5-42Tononi, G. (2008). Consciousness as integrated information: a provisional manifesto.Biological Bulletin, 215, 216–242. Retrieved from http://www.biolbull.org/content/215/3/216.shortTononi, G. (2010). Information integration: its relevance to brain function and consciousness.Archives Italiennes de Biologie, 148, 299–322. doi:10.4449/aib.v148i3.1224Tononi, G. (2012a). Integrated information theory of consciousness: An updated account.Archives Italiennes de Biologie, 150, 293–329. Retrieved from http://architalbiol.org/aib/article/view/1388Tononi, G. (2012b). PHI: A voyage from the brain to the soul. New York, NY: Pantheon.Tononi, G., & Edelman, G. (1998). Consciousness and complexity. Science, 282, 1846–1851.doi:10.1126/science.282.5395.1846Tononi, G., & Koch, C. (2015). Consciousness: Here, there and everywhere? PhilosophicalTransactions of the Royal Society B, 370(1668), 1–18. doi:10.1098/rstb.2014.0167Tononi, G., & Sporns, O. (2003). Measuring information integration. BMC Neuroscience,4(31), 1-20. doi:10.1186/1471-2202-4-31 AN EVALUATION OF THE IIT!63 Tononi, G., Sporns, O., & Edelman, G. M. (1992). Reentry and the problem of integratingmultiple cortical areas: simulation of dynamic integration in the visual system.Cerebral Cortex, 310–335. doi:10.1093/cercor/2.4.310Tononi, G., Sporns, O., & Edelman, G. M. (1994). A measure for brain complexity: relatingfunctional segregation and integration in the nervous system. Proceedings of theNational Academy of Sciences of the United States of America, 91, 5033–5037.
doi:10.1073/pnas.91.11.5033Treisman, A. (1996). The binding problem. Current Opinion in Neurobiology, 6, 171–178.doi:10.1016/S0959-4388(96)80070-5Tsuchiya, N. (2015, June 15). Empirical testing of the integrated information theory ofconsciousness. In A. Haag (Moderator), The return of consciousness. Seminarconducted by Axel and Margaret Ax:son Johnson Foundation, Stockholm, Sweden.Tsuchiya, N., & Koch, C. (2005). Continuous flash suppression reduces negative afterimages.Nature Neuroscience, 8, 1096–1101. doi:10.1038/nn1500Turing, A. M. (1950). Computing machinery and intelligence. Mind, 59, 433–460. 
doi:10.1093/mind/LIX.236.433Vimal, R. L. P. (2013). Emergence in dual-aspect monism. In A. Pereira Jr., and D. Lehmann(Eds.), The unity of mind, brain and world, current perspectives on a science ofconsciousness, (pp. 149–190). Cambridge, UK: Cambridge University Press. 
doi:10.1017/CBO9781139207065.006