Version classiqueVersion mobile

Human and Machine Consciousness

 | 
David Gamez

5. From Correlates to Theories of Consciousness

Texte intégral

5.1 Measurement of the Physical World

  • 1 This is the current definition of a metre.
  • 2 This is our conscious experience of measurement. I could also describe how Randy’s height is measu (...)

1Randy is an elephant who lives at the bottom of my garden. Six blind men often come round to feel Randy. Sometimes I like to measure Randy. To measure his height, I compare my conscious experience of Randy with my conscious experience of a stick that has been calibrated against the distance light travels in a vacuum during 1⁄299,792,458 seconds.1 The ratio between Randy and the stick is his height in metres. Randy is three sticks (three metres) high (see Figure 5.1).2

2The science of consciousness studies the relationship between consciousness and the physical world. It measures consciousness, measures spatiotemporal structures in the physical world and attempts to identify minimal sets of spatiotemporal structures (CC sets) that are linked to conscious states.

3Physical objects do not directly appear in our bubbles of experience. We do not directly perceive their mass, chemical composition or size. To measure a property of a physical object I cause it to interact with another physical object that has been calibrated in some way and observe this interaction in my bubble of experience. This typically results in a number. Eddington describes this process:

  • 3 Eddington (1928, pp. 2512).-

Let us then examine the kind of knowledge which is handled by exact science. If we search the examination papers in physics and natural philosophy for the more intelligible questions we may come across one beginning something like this: “An elephant slides down a grassy hillside…” The experienced candidate knows that he need not pay much attention to this; it is only put in to give an impression of realism. He reads on: “The mass of the elephant is two tons.” Now we are getting down to business; the elephant fades out of the problem and a mass of two tons takes its place. What exactly is this two tons, the real subject matter of the problem? It refers to some property or condition which we vaguely describe as “ponderosity” occurring in a particular region of the external world. But we shall not get much further that way; the nature of the external world is inscrutable, and we shall only plunge into a quagmire of indescribable. Never mind what two tons refers to; what is it? How has it actually entered in so definite a way into our experience? Two tons is the reading of the pointer when the elephant was placed on a weighing-machine. Let us pass on. “The slope of the hill is 60°.” Now the hillside fades out of the problem and an angle of 60° takes its place. What is 60°? There is no need to struggle with mystical conceptions of direction; 60° is the reading of the plumb-line against the divisions of a protractor. Similarly for the other data of the problem.3

Figure 5.1. The measurement of an elephant’s height in a scientist’s bubble of experience. The scientist compares Randy with a stick that has been calibrated against the distance light travels in a vacuum during 1⁄299,792,458 seconds. The ratio between Randy and the calibrated stick is his height in metres. Randy is three sticks (three metres) high. Image © David Gamez, CC BY 4.0.

4When I monitor Randy’s brain activity using electrodes, my equipment compares the effect of his brain on each electrode with the effect of a standard voltage: the ratio between these effects is the electrode’s voltage. A computer converts the electrode voltages into an attractive image of brain activity. Randy’s colourless physical brain does not directly appear to me when I am measuring it. Within my bubble of experience I am conscious of black wires emerging from a pinkish-grey brain and a 3D display of brain activity on a computer screen.

5The physical world can be measured automatically without the instruments or measured objects appearing in a bubble of experience. A robot could measure Randy with a stick and write down the result on a piece of paper.

6Measurements can be processed into numbers that correspond to different properties of an object. Electrode voltages can be processed into neuron firing events, which can be processed into firing frequencies, synchronization patterns, and so on.

  • 4 This definition of measurement is a simplified version of the one put forward by Pedhazur and Schm (...)

7Measurement assigns numbers to aspects of objects, properties or events.4 Objects, properties and events are typically described in natural language. When I measured Randy, 3 metres was the height of an elephant; 30 mV was the membrane potential of a neuron.

8Objects, properties and events are tightly defined in physics and chemistry. For example, we have clear definitions of quarks and carbon. Physicists and chemists can state exactly what it means for a physical object to contain quarks or carbon; their instruments can reliably detect whether quarks or carbon are present in a physical object.

9Context plays an important role in the description of biological objects, properties and events. Suppose I want to measure the membrane potential of a neuron. I do not use an abstract definition of a neuron to identify physical objects that are neurons—I look for a particular type of cell in the brain of an animal. The definition of a neuron only has to be precise enough to distinguish neurons from other cells in the brain. The context of a neuron (in the brain of an animal) is part of its definition.

10A neuron is well defined inside a brain—but what exactly is a neuron? Does a neuron continue to be a neuron if I remove its nucleus, give it a chrome cytoskeleton and change its resting potential to 100 V? Synthetic biologists could construct a series of intermediate cases between neurons and liver cells—it would be difficult to classify the intermediate cases. Neurons are defined in a specific biological context; no formal definition exists that can unambiguously decide whether an arbitrary physical object is a neuron.

11This vague definition of biological structures is a problem for consciousness science. We want to use what we know about consciousness in the brain to make inferences about the consciousness of non-biological systems. Suppose we identify neural correlates of consciousness and want to make inferences about the consciousness of synthetic neurons. This cannot be done without an unambiguous context-free definition of a neuron.

12To address this problem we need formal ways of describing the spatiotemporal structures that form CC sets. These will be referred to as p-descriptions:

D7. A p-description is a formal description of a spatiotemporal structure in the physical world. A p-description unambiguously determines whether a spatiotemporal structure is present in a sequence of physical states.

  • 5 For example, if we cannot devise a way of p-describing neurons, then it will be difficult to make (...)

13When a spatiotemporal structure can be completely described by physics or chemistry (for example, an electromagnetic field or a molecule), the p-description is identical to the standard scientific description. We will have to find more formal context-free ways of describing biological structures that can resolve ambiguous cases. For example, we need a p-description that can determine whether an arbitrary part of the physical world contains neurons. This should not rely on the fact that neurons are found in biological creatures, and it should provide definite classifications of synthetic neurons, which lack some of the attributes of biological neurons. If the members of a CC set cannot be captured in a p-description, then we will only be able to make inferences about the consciousness of systems that are similar to platinum standards.5

5.2 Constraints on CC Sets

14There are constraints on the spatiotemporal structures that can form CC sets. These derive from the assumptions that were introduced to measure consciousness (A1-A6) and from the requirements of scientific methodology:

C1. The spatiotemporal structures in a CC set are independent of the observer. My consciousness is a real phenomenon that does not depend on someone else’s subjective interpretation. CC sets must be formed from objective spatiotemporal structures, such as electromagnetic waves and neuron firing patterns.

  • 6 Intrinsic properties are tied to an object’s physical nature. They are held by an object independe (...)

C2. The members of CC sets are intrinsic properties.6 A conscious state supervenes on a CC set (A3a), so each duplicate of a CC set must be associated with an identical conscious state, regardless of the spatial and temporal context in which the duplicate appears.

  • 7 It is conceivable that some CC sets could be 60% correlated with conscious states. Experimental wo (...)

C3. A non-conscious system does not contain a CC set that is 100% correlated with a conscious state.7 If A and B are 100% correlated, then A cannot occur without B. If a CC set is 100% correlated with a conscious state, then all brains that contain that CC set will be conscious.

  • 8 Elsewhere I distinguished between type A and type B correlates of consciousness (Gamez 2014c). Typ (...)

C4. CC sets e-cause c-reports during consciousness experiments (A6).8 It is not necessary for every member of a CC set to e-cause c-reports. But some parts or aspects of the CC set must e-cause them. So when I say ‘I am conscious of a green tomato’, this c-report can be traced back to the CC set that e-caused it, which is functionally connected to a bubble of experience in which there is a green tomato.

15A set of spatiotemporal structures that does not conform to these constraints cannot be a correlate of a conscious state.

5.3 Pilot Studies on the Correlates of Consciousness

16We are in a beastly state of ignorance about the relationship between consciousness and the physical world. We have no idea which spatiotemporal structures form CC sets. Our intuitions are useless. We have to start with the assumption that everything in a platinum standard system that conforms to the constraints is a potential member of a CC set.

  • 9 This discussion assumes that there is a 1:1 ratio between CC sets and conscious states.

17How can we reduce our ignorance? We can carry out pilot studies. We can attempt to identify the CC set that is associated with a particular conscious state.9

18Briony is an adult human with a normally functioning brain. I strap her into a chair and connect electrodes to her temples. At intervals I display a red square at the centre of her visual field, play a loud 500Hz tone and deliver an electric shock. Briony’s attention is completely consumed by these stimuli. They are so compelling that the same conscious state can be induced on multiple occasions. This is conscious state c3.

19Each time c3 is induced I ask Briony to c-report one aspect of it. On subsequent occasions she c-reports the size of the square, the colour of the tone and the shocking sensations in her body. Over time I build up a c-description of c3 that has a tolerable amount of detail.

  • 10 The technologies that are available for measuring the brain are covered in Section 12.2.

20I want to identify the minimal set of spatiotemporal structures that is correlated with c3. When c3 is induced I measure the neuron activity in Briony’s brain as well as the electromagnetic waves, blood movements, glia activity, and so on.10 Some of these spatiotemporal structures could form the CC set by themselves—a pattern of neuron activity might be the sole correlate of c3. Or a combination of spatiotemporal structures might form the CC set that is correlated with c3. For example, a pattern of neuron activity might only be associated with consciousness when it is immersed in blood—the same neuron activity without blood would not be linked to consciousness.

  • 11 I have discussed elsewhere how the correlates of consciousness can be experimentally separated out (...)

21I must systematically consider all possible combinations of spatiotemporal structures that could form the CC set. This will enable me to identify the spatiotemporal structures that only occur when c3 is present. Suppose I want to demonstrate that a pattern of neuron activity, p2, is the sole member of the CC set. I will need to measure c3 when p2 is present and blood is present, measure c3 when p2 is present and blood is absent, measure c3 when just blood is present, and measure c3 when neither p2 nor blood are present. Further experiments will be required to distinguish p2 from glia activity, cerebrospinal fluid, and so on. This methodology is illustrated in Table 5.1.11

Table 5.1. Simple example of correlations that could exist between spatiotemporal structures in a physical system and two conscious states. It is assumed that conscious states c1 and c2 can occur simultaneously. The physical structures A, B, C and D could be dopamine, haemoglobin, neural synchronization, electromagnetic waves, etc. These are assumed to be the only possible features of the system. ‘1’ indicates that a feature is present; ‘0’ indicates that it is absent. In this example D is not a correlate of consciousness because it does not systematically co-vary with either of the conscious states. {A, B} is a set of spatiotemporal structures that correlates with conscious state c1. {C} is a set of spatiotemporal structures that correlates with conscious state c2.

  • 12 Rees et al. (2002), Tononi and Koch (2008), Dehaene (2014) and Koch et al. (2016) describe some of (...)

22Many pilot studies have been carried out on the correlates of consciousness. They have identified areas of the brain and features of neuron activity (for example, recurrent connections) that are potential members of CC sets.12 Most of these pilot studies have focused on neural patterns that might form CC sets. No attempt has been made to show that neuron activity patterns form CC sets by themselves, or to demonstrate that glia, electromagnetic waves and haemoglobin are not members of CC sets.

5.4 Natural and Unnatural Experiments

23Consciousness experiments are carried out on platinum standard systems. Assumptions A1-A6 enable us to measure the consciousness of platinum standard systems during these experiments.

24Normally functioning adult human brains are our only platinum standard systems (A4). They change as they interact with the world and learn from their experiences. Most of these changes are part of their normal behaviour—they do not affect their status as platinum standards.

25Brian’s skull contains a normally functioning adult human brain (a platinum standard system). I ask him to raise his right arm. He raises his right arm. I shave off his hair and slowly smear chocolate sauce on his face. These modifications do not affect the normal functioning of his adult human brain.

26I inject Brian with 5 mg of LSD. After a brief spell of bliss he goes wild, bangs his head against the wall, yells in an uncontrollable manner and claws at his face. His brain is not functioning normally. I shoot him in the head. He lies still on the laboratory floor. Blood pours out of his head. His brain is no longer a platinum standard system.

27A platinum standard system must remain a platinum standard system throughout an experiment on consciousness. If it ceases to be a platinum standard system, then assumptions A1-A6 no longer hold and it becomes an open question whether we can interpret its external behaviour as a c-report of its consciousness.

28Some experiments preserve a system’s status as a platinum standard; other experiments transform a system into something that is not a platinum standard. This distinction will be expressed as follows:

D8. In a natural experiment the test system preserves its status as a platinum standard. Assumptions A1-A6 remain valid and consciousness can be measured throughout the experiment.

D9. In an unnatural experiment the test system is transformed into something that is not a platinum standard. A1-A6 cease to apply and we lose our ability to measure the system’s consciousness.

  • 13 Any kind of ‘passive’ monitoring or measurement involves the passage of physically conserved quant (...)

29Natural experiments preserve a system’s physical integrity and normal behaviour. The system can be monitored using passive techniques, such as fMRI, EEG and electrodes.13 These manipulations do not affect our belief that it can c-report its consciousness.

30Unnatural experiments alter the physical constitution of a platinum standard system. They remove material, add unusual chemicals or replace brain parts with functionally equivalent chips. Unnatural experiments undermine our ability to measure a system’s consciousness. They cannot be used to identify CC sets or to test theories of consciousness.

  • 14 This experiment has been extensively discussed—for example, by Moor (1988), Chalmers (1996a), Van (...)

31Suppose we replace part of a subject’s brain with a functionally equivalent chip. This would not affect their behaviour—they would continue to make the same reports as before. This experiment has been put forward as a way of testing the hypothesis that functions or computations in the brain are linked to consciousness, rather than patterns in biological materials.14

  • 15 We have an intuition that we would notice if, for example, the implantation of the chip removed ha (...)
  • 16 It might be argued that neurons die all the time, so surely replacing one neuron with silicon shou (...)

32Prior to the experiment the subject’s brain was a platinum standard system and we interpreted its speech as a c-report of its conscious states. The implantation of the chip transforms the subject’s brain into a freak neuro-silicon hybrid that is not a platinum standard system. We have no idea whether brains with implanted chips are associated with consciousness. Assumptions A1-A6 do not apply—we have not assumed that the external behaviour of this type of system is a c-report that can be used to measure consciousness. Similar problems occur with other unnatural experiments, such as the replacement of haemoglobin with an artificial blood substitute, the removal of glia, and so on.15,16

33We could add brains with implanted chips to our list of platinum standard systems. This would transform an unnatural chip implantation experiment into a natural experiment. Both the original system and the transformed system would be platinum standards, so we could measure consciousness throughout the experiment.

  • 17 The assumption that brains with implanted chips are conscious is equivalent to the assumption that (...)

34New assumptions about platinum standard systems should not be made lightly. Pilot studies look for CC sets in the systems that are assumed to be platinum standards. A science of consciousness that studied brains with implanted chips would be very different from our current science of consciousness.17

35It will be difficult or impossible to identify all the members of a CC set using natural experiments. The members of a CC set can only be identified by systematically varying the physical world to test the link between each combination of candidate structures and consciousness (see Table 5.1). When a combination does not occur naturally it is impossible to test its link with consciousness. So natural experiments cannot test the connection between consciousness and biological neurons, because we cannot remove biological neurons from the brain without compromising its status as a platinum standard system.

5.5 Theories of Consciousness (C-Theories)

  • 18 Popper (2002, pp. 279-80).

It is possible to interpret the ways of science more prosaically. One might say that progress can ‘… come about in only two ways: by gathering new perceptual experiences, and by better organizing those which are available already’. But this description of scientific progress, although not actually wrong, seems to miss the point. It is too reminiscent of Bacon’s induction: too suggestive of his industrious gathering of the ‘countless grapes, ripe and in season’, from which he expects the wine of science to flow: of his myth of a scientific method that starts from observation and experiment and then proceeds to theories […] The advance of science is not due to the fact that more and more perceptual experiences accumulate in the course of time. […] Bold ideas, unjustified anticipations, and speculative thought, are our only means for interpreting nature: our only organon, our only instrument, for grasping her.
Karl Popper, The Logic of Scientific Discovery18

  • 19 Tononi (2008, p. 217.) I also could have quoted Dehaene: ‘Only mathematical theory can explain how (...)

What’s the matter with consciousness, then, and how should we proceed? Early on, I came to the conclusion that a genuine understanding of consciousness is possible only if empirical studies are complemented by a theoretical analysis. […] This state of affairs is not unlike the one faced by biologists when, knowing a great deal about similarities and differences between species, fossil remains, and breeding practices, they still lacked a theory of how evolution might occur. What was needed, then as now, were not just more facts, but a theoretical framework that could make sense of them.
Giulio Tononi, Consciousness as Integrated Information: A Provisional Manifesto19

36When people studied the heavens they were not seeking an infinitely long list of the planets’ positions. They wanted a compact theory that could calculate the positions of the planets at an arbitrary point in time. Ptolemy developed a model based on deferents and epicycles. This was superseded by Newton’s and Einstein’s equations.

37Pilot studies might identify the correlates of some conscious states. This would be a major scientific achievement. It would help us to find the correlates of other conscious states. It would tell us something about the consciousness of non-platinum standard systems, such as coma patients, bats and robots.

  • 20 I have discussed the need for c-theories elsewhere (Gamez 2012b).

38The wine of a science of consciousness will not flow from industrious gathering of data about the correlates of individual conscious states. There are an effectively infinite number of conscious states—we cannot identify the CC sets associated with each one. Instead we need a compact mathematical theory that can map physical states onto conscious states and vice versa. This will be referred to as a c-theory:20

  • 21 The search for c-theories is closely related to the attempt to discover the relationship between b (...)

D10. A c-theory is a compact expression of the relationship between consciousness and the physical world. A c-theory can generate a c-description from a p-description, and generate a p-description from a c-description.21

39The role of c-theories is illustrated in Figure 5.2.

Figure 5.2. Theory of consciousness (c-theory). On the left, a measurement of the invisible physical world is converted into a formal p-description of a physical state. On the right, consciousness is measured with a c-report, which is converted into a formal c-description of a conscious state. The c-theory maps between the p-description and the c-description. Image © David Gamez, CC BY 4.0.

40C-theories specify which types of spatiotemporal structures form CC sets. For example, neuron activity patterns, information patterns or computations might be members of CC sets. The most popular types of c-theory are covered in the next three chapters.

  • 22 ‘Mathematics’ should be interpreted in a broad sense that includes computer algorithms.

41C-theories should be based on mathematics.22 This is the most compact way of linking p-descriptions to c-descriptions. Philosophical theories of consciousness might inspire c-theories. But the relationship between c-descriptions and p-descriptions cannot be expressed in natural language. Natural language is too weak and vague—it cannot make strong testable predictions.

  • 23 This example and its intensity values are purely illustrative. More work needs to be done on the c (...)

42Suppose we discover a neuron whose firing rate is correlated with a bubble of experience in which there is a single point of red light. We develop a c-theory that uses the equation ln(r)=2i to connect the neuron’s firing rate, r, with the intensity of the conscious red light, i. This theory predicts that when the neuron fires at 7 Hz it will be associated with conscious red light that has intensity 0.97. It also predicts that the neuron will fire at 20 Hz when conscious red light occurs with intensity 1.5.23

43C-theories map between conscious states and sets of spatiotemporal structures in the physical world. These spatiotemporal structures must be valid members of CC sets. So the constraints on the members of CC sets (Section 5.2) are constraints on c-theories. C-theories must generate p-descriptions of valid CC sets from c-descriptions, and they must generate c-descriptions from p-descriptions of valid CC sets. C-theories that do not conform to the constraints should be excluded from the science of consciousness.

  • 24 This is an unashamedly Popperian approach to the science of consciousness. Some would argue that P (...)

44C-theories become scientifically credible when their predictions pass experimental tests. It is not enough for c-theories to match data gathered during pilot studies—they have to generate strong predictions that can be experimentally confirmed. The most compact and accurate c-theory will be considered to be a correct description of the relationship between consciousness and the physical world.24

  • 25 C-theories describing brute regularities have some similarity with Chalmers’ psychophysical laws: (...)

45Our final c-theories will describe brute regularities in the relationship between consciousness and the physical world (see Section 3.5). As the science of consciousness progresses we are likely to develop c-theories that describe regularities which can be further decomposed into more basic relationships. It might be impossible to tell whether a c-theory describes a genuine brute regularity.25

  • 26 For example, Tononi’s (2008) information integration theory is based on his first-person observati (...)

46Some people base c-theories on their conscious experiences.26 But the source of inspiration of a c-theory is irrelevant to its success. We cannot directly imagine the relationship between consciousness and the invisible physical world (see Section 3.3), so the intuitive plausibility of a c-theory has no bearing on whether it is correct. C-theories stand or fall on their ability to make falsifiable predictions that pass experimental tests.

47C-theories are not likely to provide intuitively satisfying explanations of the relationship between consciousness and the physical world. Since we cannot imagine the physical world, a mathematical c-theory cannot help us to make an imaginative transition from the invisible physical world to consciousness. At most a c-theory could help us to make an imaginative transition from a conscious experience of brain activity to another conscious experience (see Section 3.4).

48People might use particles, forces or novel aspects of the physical world to explain why a particular relationship between c-descriptions and p-descriptions holds (see Section 6.3). Such explanations might be scientifically fruitful—they might help us to develop new mathematical c-theories. But they are unlikely to make a c-theory more intuitively plausible. Newton could not imagine how gravity acted at a distance. We cannot explain why brute regularities exist between consciousness and the physical world.

5.6 The Computational Discovery of Theories of Consciousness

  • 27 Humphreys (2004, p. 90).

[…] the era of simple mathematics effectively modelling parts of the world is drawing to a close. It is possible that new areas of investigation will lend themselves to simple models, but the evidence is that within existing areas of investigation, the domain of simple models has been extensively mined to the point where the rewards are slim.
Paul Humphreys, Extending Ourselves27

49Traditional science is based on the idea that people identify regularities in the physical world. It is a working assumption that physical regularities are simple enough to be found by humans. When a human finds a regularity s/he might use a novel property to explain it. A mathematical description of the regularity can be experimentally tested. Humans find this satisfying—they like solving puzzles. But it is not necessarily the most effective approach.

  • 28 There is also a more general question about whether one human brain can fully understand another—o (...)

50Humans are biased and stupid. They have small working memories and little imagination. They cannot process large data sets. These limitations will prove fatal to consciousness science if there are complex relationships between c-descriptions and p-descriptions.28

51We have little or no idea about the spatiotemporal structures that form CC sets. The mathematical relationships between c-descriptions and p-descriptions are unknown. They might be simple—a few differential equations. Or the mathematical complexity of these regularities could be way out of reach of human capabilities. Macro-scale laws of the brain could extend to thousands of pages of differential equations.

52We should drop the assumption that there are simple relationships between p-descriptions and c-descriptions. We have no reason to believe that this is the case. If we persist with the assumption that there are simple relationships, we could spend large amounts of time and money on a fruitless quest for something that does not exist. It is better to assume that the relationships between c-descriptions and p-descriptions are potentially complex, and develop a methodology that can identify simple and complex relationships (or prove that no relationships exist).

  • 29 A substantial amount of research has been carried out on the use of computers for scientific disco (...)

53We could use computers to identify the relationships between p-descriptions and c-descriptions. This would require a large amount of data, spanning multiple levels of the brain. C-descriptions and p-descriptions would have to be recorded for many different conscious states. This data could be gathered by human scientists. Or robots could capture it automatically.29

54Machine-learning techniques could be applied to this data. The patterns that were found could be used to make predictions, which could be tested in further experiments. C-theories could be automatically tested against new data as it came in.

  • 30 For example, Billeh et al. (2014) have developed a way of identifying functional circuits in recor (...)

55C-theories that are discovered by computers should be expressed in a format that can be read by both humans and machines (they should not be stored as weights in a complex neural network). This would enable them to be partly viewed and verified by humans—but there would be no expectation that an individual human scientist could check or comprehend them in their entirety. Sets of differential equations would be a good choice of output format—there is a long tradition of using differential equations to describe complex relationships in the physical world. Or perhaps we could use graph theory to describe the relationships between c-descriptions and p-descriptions.30

56This computational approach to the science of consciousness could identify simple relationships between c-descriptions and p-descriptions. It could find regularities that are too complex to be identified by humans. Or it could prove that no simple or complex laws exist in the current data.

  • 31 The Blue Brain Project has developed detailed models of a cortical column (Markram 2006) and this (...)
  • 32 The ‘c-reports’ of a simulated brain could not be used to measure its consciousness because a neur (...)
  • 33 Simulations are very different from real brains, so this would primarily be a test of the methodol (...)
  • 34 This methodology could also be used to solve the more general problem of the relationship between (...)

57This approach could be prototyped on a simulated human brain.31 This would generate reports ‘about consciousness’ that are similar to c-reports and could be converted into c-descriptions.32 The computer could search for relationships between these c-descriptions and different aspects of the neural model. It could control the simulation (rewinding it, rewiring it, changing its parameters) to robustly test its hypotheses.33 The relationships between c-descriptions and p-descriptions that were identified by this method could be tested on platinum standard systems.34

5.7 Summary

58This chapter has described how we measure the physical world. Physical measurements have to be expressed in a formal way (a p-description), so that we can use our knowledge about CC sets in humans to make inferences about the consciousness of non-biological systems.

59Pilot studies could identify the CC sets that are linked to individual conscious states. These must use natural experimental methods, which preserve our ability to measure consciousness in platinum standard systems.

60In the longer term we need to develop mathematical c-theories that map between p-descriptions and c-descriptions (see Figure 5.2). These c-theories must conform to the constraints on CC sets (C1-C4).

61Humans might be incapable of discovering complex mathematical relationships between p-descriptions and c-descriptions. To avoid this potential problem, computers should be used to discover c-theories.

Notes

1 This is the current definition of a metre.

2 This is our conscious experience of measurement. I could also describe how Randy’s height is measured by the physical brain of the scientist.

3 Eddington (1928, pp. 2512).-

4 This definition of measurement is a simplified version of the one put forward by Pedhazur and Schmelkin (1991), who take it from Stevens (1968.) According to Stevens, most measurement involves ‘the assignment of numbers to aspects of objects or events according to one or another rule or convention.’ (Stevens 1968, p. 850). Pedhazur and Schmelkin stress that numbers are assigned to aspects of objects, not to objects themselves. We measure the height, width and colour of a box, not the box itself.

5 For example, if we cannot devise a way of p-describing neurons, then it will be difficult to make inferences about the consciousness of animals with larger neurons, such as snails and insects, and we will not be able to say anything about the consciousness of artificial systems.

6 Intrinsic properties are tied to an object’s physical nature. They are held by an object independently of its spatial and temporal context. Extrinsic properties depend on an object’s relationships with other parts of the world. The chemical composition of a neuron is an intrinsic property. The distance of a neuron from the North Pole is an extrinsic property, which would change if the North Pole changed location.

7 It is conceivable that some CC sets could be 60% correlated with conscious states. Experimental work could determine whether this is the case. C3 will not apply if there are inhibitors of consciousness (see Chapter 4, Footnote 32).

8 Elsewhere I distinguished between type A and type B correlates of consciousness (Gamez 2014c). Type A correlates can e-cause c-reports and are compatible with C4. Type B correlates are not compatible with C4 because they cannot e-cause c-reports.

9 This discussion assumes that there is a 1:1 ratio between CC sets and conscious states.

10 The technologies that are available for measuring the brain are covered in Section 12.2.

11 I have discussed elsewhere how the correlates of consciousness can be experimentally separated out from their prerequisites and consequences and from sensory and reporting structures (Gamez 2014c.) Pitts et al. (2014) describe experimental work that attempts to separate the correlates of conscious perception from reporting structures. This is also discussed by Koch et al. (2016.)

12 Rees et al. (2002), Tononi and Koch (2008), Dehaene (2014) and Koch et al. (2016) describe some of the research that has been carried out on the neural correlates of consciousness.

13 Any kind of ‘passive’ monitoring or measurement involves the passage of physically conserved quantities from the system to the measuring device. In a natural experiment this is small compared to the exchange of physically conserved quantities within the system, so it does not affect our assumption that the system is a platinum standard.

14 This experiment has been extensively discussed—for example, by Moor (1988), Chalmers (1996a), Van Heuveln et al. (1998) and Prinz (2003). Part of the brain could be replaced by any functionally equivalent system, such as a giant lookup table or the population of China communicating with radios and satellites (Block 2006).

15 We have an intuition that we would notice if, for example, the implantation of the chip removed half of our visual consciousness. But according to the premises of the experiment, our behaviour would be identical, so nothing in our thoughts or speech would reflect this change in consciousness. If the implanted chip did affect our consciousness we would not be cognitively aware of the change and it would not affect our ability to perceive and respond to the world. We would be like people with anosognosia (see Chapter 4, Footnote 27), with the difference that our sight and bodies would be working perfectly, so no external observer could detect the change in our consciousness.

16 It might be argued that neurons die all the time, so surely replacing one neuron with silicon should not affect our assumption that the brain is a platinum standard? And so on with two neurons, three neurons, until the entire brain has been replaced. Chalmers’ (1995a) fading and dancing qualia argument proceeds along these lines. One problem with this argument is that it is based on the invalid assumption that we can imagine the relationship between consciousness and the brain (see Chapter 3). Another problem is that the brain can be sensitive to individual spikes, so the replacement of individual neurons could affect its consciousness. For example, a single neuron could individually encode an abstract concept or make a significant contribution to a population code. If this neuron was part of a CC set, then its replacement with a silicon chip could alter the associated conscious state.

17 The assumption that brains with implanted chips are conscious is equivalent to the assumption that functionalism is true. This brings in all of the problems with computation and information theories of consciousness that are discussed in Chapters 7 and 8.

18 Popper (2002, pp. 279-80).

19 Tononi (2008, p. 217.) I also could have quoted Dehaene: ‘Only mathematical theory can explain how the mental reduces to the neural. Neuroscience needs a series of bridging laws, analogous to the Maxwell-Boltzmann theory of gases, that connect one domain with the other. This is no easy task: the “condensed matter” of the brain is perhaps the most complex object on earth. Unlike the simple structure of a gas, a model of the brain will require many nested levels of explanation. In a dizzying arrangement of Russian dolls, cognition arises from a sophisticated arrangement of mental routines or processors, each implemented by circuits distributed across the brain, themselves made up of dozens of cell types. Even a single neuron, with its tens of thousands of synapses, is a universe of trafficking molecules that will provide modelling work for centuries.’ (Dehaene 2014, p. 163).

20 I have discussed the need for c-theories elsewhere (Gamez 2012b).

21 The search for c-theories is closely related to the attempt to discover the relationship between brain activity and behaviour. Computational methods could also be used to study this relationship (see Section 5.6). However, c-theories might be based on non-neural structures in the brain, such as novel materials, haemoglobin and electromagnetic waves (see Section 6.2 and Section 6.3), that would not be required by theories that describe the relationship between neuron activity and external behaviour.

22 ‘Mathematics’ should be interpreted in a broad sense that includes computer algorithms.

23 This example and its intensity values are purely illustrative. More work needs to be done on the conversion of c-reports into c-descriptions that record the intensity of different aspects of conscious experience. This could draw on previous work in psychophysics—for example, Gescheider (1997).

24 This is an unashamedly Popperian approach to the science of consciousness. Some would argue that Popper (2002) presents an outmoded account of the philosophy of science, which should be replaced by Kuhn (1962) at least, or perhaps Feyerabend (1975) or Latour (1987). Some of these later ‘relativist’ ‘constructivist’ ‘postmodern’ accounts reject the possibility of scientific progress altogether. However, if we are attempting to understand how a science of consciousness can be developed, we need a model of what science is. And I would argue that Popper provides a carefully thought out and convincing account of what good scientific practice should be. Other philosophies of science can be used to interpret the science of consciousness, but many of them are considerably less useful as guiding principles than Popper—how (or why) would we develop a science of consciousness based on Feyerabend or Latour?

25 C-theories describing brute regularities have some similarity with Chalmers’ psychophysical laws: ‘Where we have new fundamental properties, we also have new fundamental laws. Here the fundamental laws will be psychophysical laws, specifying how phenomenal (or protophenomenal) properties depend on physical properties. These laws will not interfere with physical laws; physical laws already form a closed system. Instead they will be supervenience laws, telling us how experience arises from physical processes. We have seen that the dependence of experience on the physical cannot be derived from physical laws, so any final theory must include laws of this variety.’ (Chalmers 1996a, p. 127.) However, this book suspends judgment about some of the metaphysical substance-based theories, and the relationship between c-descriptions and p-descriptions is symmetrical, not a causal relationship in which consciousness arises from physical processes.

26 For example, Tononi’s (2008) information integration theory is based on his first-person observations about the differentiation and integration of consciousness.

27 Humphreys (2004, p. 90).

28 There is also a more general question about whether one human brain can fully understand another—one might think that a brain could only be understood by a larger and more complicated system. This issue can potentially be addressed by using the world as external memory (Clark 2008; O’Regan 1992). This would only work if our understanding of the brain can be broken down into interrelated modules. For example, we could develop a detailed understanding of how a neuron works, write it down, and then work on a different aspect of the problem, until we had written down everything about the brain. Although the final solution could not be comprehended by a single brain all at once, one or more brains could check the validity of each part and the links between them.

29 A substantial amount of research has been carried out on the use of computers for scientific discovery (Dzeroski and Todorovski 2007). Robotic systems have been developed that can carry out experiments automatically (Sparkes et al. 2010), and there has been research on the automatic discovery of differential equations that describe the behaviour of dynamic physical systems (Schmidt and Lipson 2009). This work suggests how consciousness could be scientifically studied in the future.

30 For example, Billeh et al. (2014) have developed a way of identifying functional circuits in recordings of spiking activity from hundreds of neurons. Using this approach it might be possible to develop a way of describing brain activity in terms of interacting circuits, which could be identified automatically by a computer.

31 The Blue Brain Project has developed detailed models of a cortical column (Markram 2006) and this work is being continued on a larger scale in the Human Brain Project (www.humanbrainproject.eu). Larger, less detailed models have also been built of human and animal brains (Ananthanarayanan et al. 2009; Izhikevich and Edelman 2008). The feasibility of scanning and simulating a human brain is discussed in Chapter 11, Footnote 14. None of the current models generates behaviour that is similar to c-reports and most of them do not include non-neural components of the brain, such as glia.

32 The ‘c-reports’ of a simulated brain could not be used to measure its consciousness because a neural simulation is not a platinum standard system.

33 Simulations are very different from real brains, so this would primarily be a test of the methodology. However, this type of work might lead to c-theories that could be tested on real brains.

34 This methodology could also be used to solve the more general problem of the relationship between an organism’s brain activity and all of its behaviour (both conscious and non-conscious). Once the behaviour had been formally described, computers could be used to discover relationships between the brain activity and behaviour. This approach could be prototyped on a very simple system, such as a simulated C. elegans.

Table des illustrations

Légende Figure 5.1. The measurement of an elephant’s height in a scientist’s bubble of experience. The scientist compares Randy with a stick that has been calibrated against the distance light travels in a vacuum during 1⁄299,792,458 seconds. The ratio between Randy and the calibrated stick is his height in metres. Randy is three sticks (three metres) high. Image © David Gamez, CC BY 4.0.
URL http://books.openedition.org/obp/docannexe/image/5558/img-1.jpg
Fichier image/jpeg, 121k
Légende Table 5.1. Simple example of correlations that could exist between spatiotemporal structures in a physical system and two conscious states. It is assumed that conscious states c1 and c2 can occur simultaneously. The physical structures A, B, C and D could be dopamine, haemoglobin, neural synchronization, electromagnetic waves, etc. These are assumed to be the only possible features of the system. ‘1’ indicates that a feature is present; ‘0’ indicates that it is absent. In this example D is not a correlate of consciousness because it does not systematically co-vary with either of the conscious states. {A, B} is a set of spatiotemporal structures that correlates with conscious state c1. {C} is a set of spatiotemporal structures that correlates with conscious state c2.
URL http://books.openedition.org/obp/docannexe/image/5558/img-2.jpg
Fichier image/jpeg, 218k
Légende Figure 5.2. Theory of consciousness (c-theory). On the left, a measurement of the invisible physical world is converted into a formal p-description of a physical state. On the right, consciousness is measured with a c-report, which is converted into a formal c-description of a conscious state. The c-theory maps between the p-description and the c-description. Image © David Gamez, CC BY 4.0.
URL http://books.openedition.org/obp/docannexe/image/5558/img-3.jpg
Fichier image/jpeg, 88k

Acheter

Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search