Faustian Dreams
01-06-2006, 03:56 PM
I understand that posting one's academic material is generally considered a form of misbehavior, and, oddly enough, even punishable by law (if I recall correctly). But the dissemination of information is man's greatest tool...so enjoy!
On the Improbability of Instantiating Consciousness through Artificial Means
With the exponential advances being made in technology, it would appear that the realization of artificial intelligence is not so unreasonable of an aspiration, nor too distant into the future at that. Sophisticated systems technologies that are able to interact with their environment have already been built and robots are able to simulate human movement to increasingly more sensitive degrees. However, AI researchers have become wary of declaring any existing machine to be artificially intelligent or conscious—a machine has yet to pass the much-lauded Turing test, whereby a computer program would convince an interrogator that it is a human being. In fact, there still remains the question of how perception becomes intelligence, which becomes consciousness, in the human brain itself; one might then ask how it could be possible to first create a program, which would in turn explain the mechanisms that drive the former. Subsequently, we are limited to theorization as to how consciousness could possibly arise, and to the possibility that it can be successfully replicated outside of the human mind. In this essay, I shall seek to invalidate such claims, made by Turing and Jacquette in particular, that artificial intelligence and consciousness could be created, as they are embedded in faulty philosophy and logic, and which, as current research in neuroscience has shown, would require an unfathomable amount of complexity and sensitivity.
In his essay, “Computing Machinery and Intelligence,” Alan Turing proposes that the question “Can machines think?” is too vague, instead suggesting a test whereby a behaviorist qualification for artificial intelligence could be made. The Turing test involves two subjects, a human being and a computing machine, and a human interrogator, whom they must convince of their own humanity. One can deduce that not only would this require an extensive database of response sets, but such persuasive capabilities that would suggest the machine can present itself as something more “human” than the other. Assuming that this test would suffice in determining intelligence, this would already be a daunting task. Unfortunately, the Turing test is too insensitive to the demands made upon organisms in their conscious experience; there is no consider of the sensory perception which ultimately provides the being in question with its tools for intelligence. Turing never affirms the possibility of creating a “thinking” machine, let alone one that could reproduce consciousness, his basis remains this test, which ultimately appears to merely try and simulate human behavior, human thought .
This is exactly Searle’s contention with the Turing test; while he has no objection to what he calls ‘weak’ AI, which simply aims to model the human mind, Searle argues that the notion of creating ‘strong’ AI—or computers which can actually be considered “conscious”—is unrealistic. In reference to Roger Schank’s program which purportedly understands stories and can infer various things from the given facts, Searle offers his “Chinese room” analogy, which criticizes the functionalist concept that the means by which processing of input creates an output is arbitrary. His reasoning is that simply because such processing may appear to produce a successful and coherent output, one cannot deduce that there is actual understanding of what transpires . Understanding is synonymous with consciousness, which, Searle argues, presupposes that the subject will act with intentionality:
[T]he main point of the present argument is that no purely formal model will ever be sufficient by itself for intentionality because the formal properties are not by themselves constitutive of intentionality, and they have by themselves no causal powers except the power, when instantiated, to produce the next stage of the formalism when the machine is running.
In addition to the fact that formal properties are not sufficient for intentionality, it is commonly held in the field of neuroscience that consciousness in an organism is characterized by ever-changing conscious states. There are a few errors in the consistency of his argument, however; while he concludes that intentional states are “defined in terms of their content, not their form,” earlier in his essay he remarks that it is because of a distinct biological structure that he is able to partake in intentional phenomena .
While Jacquette does not include this contradiction in his counterargument to Searle, he does bring into question the effectiveness of the “Chinese room” analogy on the basis of whether it operates on the same level of isomorphism as a computer program would. That is to say, if Searle’s analogy is not compatible with the mechanisms by which computer programs operate, then in turn, there is still a possibility of synthesizing an understanding machine. Jacquette asserts that the analogy does not function on the same level as a computer program would, and thus, since Searle envisions himself as a homuncular agent within the room, he incorrectly assumes that there is no understanding within the system as a whole. Additionally, he believes that intentionality is an abstract, a Universal, thus not physically bound to such “wet circuitry,” which is merely an accident. He concludes that Searle’s claim that consciousness, or rather intentionality, is irreducible, analogously to wetness and elasticity, is unjustified unless “he can offer a clearcut case of a decidedly nonintentional property caused by and realized in but not reducible to a material microphysical structure.” In disproving Jacquette’s remarks, although the validity of Searle’s argument in its entirety will not be restored, it will allow for his more cogent remarks to be evaluated from an empirical perspective.
The pervading theme of functional isomorphism on a micro- and macro-level accurately describes the dilemma presented by contrasting theories of artificial intelligence. It is a pertinent question; what must be replicated in order to successfully create consciousness in a machine? Unfortunately, Jacquette’s own reasoning sheds light on the fallacy of functionalism. In addition to the fact that functionalism assumes a dualistic perspective, that function exists as something distinct from form, as Searle points out (although functionalists would deny this in stating that the philosophy actually aims to bind mental states to functional states of physical system ) it fails to take into account the micro-level processing from which consciousness precipitates. In alignment with the linearity of the functionalist, input-process-output concept, I will offer an analogy: Suppose that two individuals plan on making a journey across the country; one by foot, one by automobile. What functionalism fails to take into account are the vastly different means by which the journey is made—so long as the two individuals are present at their destination, they would declare that the two journeys are acceptable isomorphs. There can be no disagreement about the compatibility in terms of their proper levels, as the concept is merely the A to B process.
To bolster this analogy, one can even remove the individual from the automobile and have a computer program navigate the vehicle to its proper destination. Thus, we have two distinct entities, and two distinct processes. Still, in spite of any attempts which would be made on behalf of the functionalist argument, it is clear that two very different phenomena are underway. The combustion of the gasoline to propel the pistons that power the engine, which in turn set the wheels rotating on their axis—it is quite different from the digestion of nutrients of the body which creates the adenosine triphosphate (ATP) that powers the cells, which allow from the deployment of acetylcholine to the muscular junctions that allow for locomotion. Clearly, there are processes within processes that determine the specific mechanisms by which the observable behavior is executed, and it is with this same sensitivity that artificial consciousness and intelligence ought to be approached. However, before one is able to create an artificially conscious machine, the neurophysiological correlates of consciousness must be defined and acknowledged. From this evidence, the mere possibility of AI will be assessed—and as the following sources will demonstrate, the likelihood is not very strong!
The most commonly accepted theory for explaining the emergence of consciousness from mental states would be the “dynamic core” hypothesis, proposed by Gerald Edelman and Giulio Tononi, which states that consciousness has three physiological requirements: differentiation, integration, and reentry . A highly differentiated system implies that its substructures all serve unique purposes, which, when integrated, allow for the coherent interpretation of perceptual stimuli. The means by which such a system can be both highly differentiated and integrated at once is reentry; which Edelman describes as an “ongoing, recursive interchange of parallel signals between reciprocally connected areas of the brain, an interchange that continually coordinates the activities of these areas’ maps to each other in space and time.” He explains that reentry is unlike feedback in that there are many parallel paths, and is developed and reinforced selectively, instead of through unidirectional alteration. This plasticity of the brain, which is determined by selection and allows for consciousness to arise, is emphasized by Edelman and Tononi; the brain changes on a daily basis—in fact, conscious experience affects the systems of our brains (which would refute epiphenomenalism). Neurons constantly die and synapses are always being rearranged.
The mind is not based on systems of logic but selectional systems, just as a child is not born with a sense of logic, but instead learns through an experiential trial-and-error process. But before confronting the issue of the acquisition of intelligence, let us continue to formally define consciousness. Zeman attributes the following characteristics to conscious experience: that it is personal; its contents are only stable for short periods, lasting up to a few seconds at most; its contents are unified and continuous over time; it is selective, with a limited capacity . Lastly, he mentions that conscious states are, in fact, intentional—intentionality being a phenomenon which Jacquette criticized for being too abstract and baseless, a decade prior to such research. Among the unique events within the human brain which are involved in consciousness, Zeman describes the parts of the brain, such as V4 and V5, which process color and movement, respectively . The particularity of function of neurons in the occipital lobe suggests that there is some truth to qualia, in that the brain does in fact create a complete picture out of the many properties sensory experience provides. Studies have shown that melody and rhythm are likewise processed in different parts of the brain.
Another unique mechanism is the ability for the subject to recreate images, events, and locations mentally, in the “mind’s eye.” Zeman cites studies that have found that mental rotation tasks and envisioning a route to a destination are both proportional to the time that would be required to enact such tasks in reality . A final example from Zeman is his reference to the decrease in engagement and global brain activity in the acquisition of new skills . All of these examples demonstrate how the parts of the brain work together, through integration, as a master ensemble in order to facilitate consciousness. Giulio Tononi extensively describes how integration is a necessary component for the complex states that are the constituents of consciousness; establishing his hypothesis as follows:
If consciousness corresponds to the capacity to integrate information, then a physical system should be able to generate consciousness to the extent that it has a large repertoire of available states (information), yet it cannot be decomposed into a collection of causally independent subsystems (integration).
He subsequently explains that in order for a system to be conscious, it must be composed of complexes, or rather, consist of a main complex which is necessary and sufficient for the integration of information—nothing outside of the complex is capable of being integrated. That is not to say, however, that we are conscious of everything that occurs within the complex; the many aforementioned processes, the qualia, are not things we can be consciously aware of. We cannot focus on any particular aspect of our perceptual field—a reality which quite well demonstrates a “nonintentional property caused by and realized in but not reducible to a material microphysical structure” that Jacquette wished for Searle to provide, which has been done in his stead; sensory perception is by no means intentional, we do not actively engage in perceiving the world about us, and reducing it past what can be consciously perceived results in an explanation of physical processes—but no distinct locus.
Yet all this refutation of Jacquette’s statement has established is that intentionality exists. Should intentionality be necessarily attributed to consciousness? Or rather, can consciousness exist independent from intentionality? This issue is of no concern in the question of artificial consciousness, and for this reason alone: the only way in which AI researchers could absolutely determine that they have artificially created a conscious machine is by comparing it to the only known substrate which has facilitated consciousness in a form which can be accurately accounted for—the human being. While the exact relationship between intentionality and consciousness remains to be seen, if one admits the fact that the former is a necessary component of the latter, it should follow that in creating a conscious machine, intentionality must be observable in its conscious states as well . Thus, it is not sufficient for the trend in modern research to attempt in creating mere consciousness, without aim, but to consider that these artificially intelligent machines must originate from a familiar mechanism or process that can be causally reduced to identifiable components—they must be instantiated from a consciousness like ours.
One might offer that there is no possible means of determining whether a machine is conscious, let alone that it is conscious in the same way humans are conscious, but such solipsistic mentality can be dismissed primarily because it would be as absurd as asking whether one can be entirely sure that any other human being apart from us is subject to certain, unifying experiential qualities. Additionally, this “consciousness like ours” position may be met with claims that such a high standard for judgment puts AI research up against a nearly impossible task. There is a particular reason for this; for those who would argue that a conscious machine does not need to resemble any existing anatomical form, they must be reminded that consciousness includes self-consciousness, which can only be facilitated through a strong sense of spatial boundaries . It must be aware of where the self ends and the rest of the world begins. Furthermore, it must be able to interact with the world in many different ways, and to gather information from this world through the five senses. Once again, such a statement would be met with contention, such as: “Why must a conscious machine have to have all the senses? Aren’t there people who lack one sense or another, yet are still able to be identified as a conscious being?” To this I would reply: Isn’t the goal of artificial intelligence to come to understand human consciousness and acquisition in its entirety? Even if there are human beings who may lack the sense of smell, or taste, or touch, what relevance does this have to the study which aims to determine the mechanisms that are underway within a healthy, fully functional individual? There is no sense in building a handicapped robot. Thus, in addition to the “consciousness like ours” argument, we have raised another qualification for AI research; that it must be a complete form of consciousness, equipped with the same functions that evolution has provided for organisms.
From a structural perspective, referring back to the concepts of neural plasticity and the dynamic core, it is clear that this conscious machine would have to be built in a manner such that, in addition to having a vast option for forming networks or pathways, it would be able to continually alter these networks as experience dictates within the brain itself. Consequently, a simple geometric layout would not suffice in offering the greatest amount of information integration possible, since the bypass systems which would be required to allow for a unit of information to travel from microprocessor α to travel to microprocessor γ, without interference from microprocessor β (which, perhaps in this scenario, forwards directly to ω, eliminating the subsequent algorithms instantiated from γ, resulting in an altogether different effect), would require yet another framework, an inhibitory one, to be superimposed upon, or built into the standard processing unit. More broadly speaking, there can be nothing which records and permanently stores all information, particularly visual perception, as consciousness is a selective process . Thus, there must also be some way in which the machine could direct its focus. Of the currently existing robots which are able to successfully process images, their responses are limited to whatever stimuli they have been programmed to respond to, either colors or particular faces.
Here it is important to emphasize the importance that a truly conscious machine cannot be preprogrammed—as consciousness is something entirely personal, I would go so far as to state that the only inbuilt processes a conscious machine may have are the senses. If a machine is programmed, that implies that it can be reproduced by the programmer who created it. Unfortunately, there is a clear distinction between the manual programming that results in artificial intelligence (by every sense of the word “artificial”) and the ‘passive’ programming a child experiences during their early development, which occurs through interaction with their environment, so the former would not be representative of personal experience. This raises another limitation for AI research: a conscious machine will have to be provided with the mechanisms required for obtaining information, but by no means may it be equipped with any “adult” skills; the robot must be a child. What this implies is that the robot must be able to establish some sort of value system, whereby its experiences are categorically stored based on some form of threshold- or frequency-sensitive basis. For example, in order for a significant event to be stored under the category of “traumatic experience,” certain qualifications must be met, which may be allowed to have been preprogrammed. This subsequently raises the question of how human beings themselves experience emotions, and how memories are influenced and reinforced by our emotions.
Consciousness has already shown to have emerged from biological processes, such as natural selection and evolution, as well as the principles that dictate the structure of the brain itself, such as neuronal selection. Evolutionarily speaking, then, it is clear that the advantages of consciousness as a social mechanism—that is, a way of forming strong relationships with other individuals—play an equally important role in determining the requirements for this conscious machine . If the goal of self-consciousness, in addition to the survival of the organism, is to be able to comprehend the mental states of other individuals, then the “consciousness like ours” argument becomes more reliable. But how is it possible to create a conscious machine that can identify with human emotions? Or is it even possible?
Let us review the qualifications made for the construction of a truly conscious machine: that it should have a consciousness like ours; that its consciousness be fully functional and comparable to that of a healthy human being; and that it should be equipped with mechanisms that enable it to acquire information, but no preprogrammed systems for the semantics and syntax of language, logical reasoning, etc. It appears to me that such requirements not only substantially limit the possibility of creating artificial intelligence, but in all probability completely eliminate such a possibility. The Turing test is in complete antithesis to such claims, despite Turing’s own assertion that a thinking machine would have to be constructed like a child—there is no basis for determining consciousness in his test. If, however, one was to maintain the Turing test as the standard for AI research, then I would suggest that, in addition to being able to fool the interrogator into believing it is a human, it should perform equally well as the interrogator itself in Turing’s test! Consciousness involves making judgments, which is entirely missing from Turing’s thesis; this corollary aims to incorporate this obvious necessity so as to further qualify the requirements for a conscious machine.
Furthermore, in having asserted that functionalism and its correlated philosophies (computationalism, connectivism, and other forms of materialism ) are errant in incorrectly assuming the mechanisms that drive our own consciousness can be reproduced outside of the brain, it is clear that artificial consciousness is impossible. Every particular substance and process has its own unique constituents, and while one may come close to replicating its properties, it will always be simulation, and nothing more. Additionally, emphasizing that consciousness is personal; to build a conscious machine would imply that there is a particular center which can theoretically be transplanted to other robots or even mass-produced—there is nothing personal about that! To be sure, neuroscience has not even come close to determining exactly how consciousness operates; yet AI researchers maintain that through a bottom-up approach this question will be solved. The only way to approach the study of consciousness is from the top down, as it is currently being done by cognitive neuroscientists, and this should be obvious, considering the way in which consciousness functions as a whole—that is, within a group of sublevel processes and interactions between neuronal groups. While human beings are programmed biologically to be human, and are predisposed to certain behaviors, this by no means can be considered to be isomorphic to the programming of machines. There is some aspect of personality which remains to be understood, yet which ultimately distinguishes man from machine. It is best to assume that while tinkering with technology may produce interesting novelties, determining the bases for human behavior is best left for human examples.
References:
Aleksander, Igor. “A Neurocomputational View of Consciousness.” From Brains to Consciousness? Essays on the New Sciences of the Mind. Ed. Steven Rose. Princeton: Princeton University Press, 1999. 180-199.
Aydede, Murat and Guven Guzeldere. “Consciousness, Intentionality, And Intelligence: Some Foundational Issues for Artificial Intelligence.”
Journal of Experimental & Theoretical Artificial Intelligence.
Baars, Bernard J. "The Conscious Access Hypothesis: Origins and Recent Evidence." TRENDS in Cognitive Sciences 6.1 (2002): 47-52.
Crick, Francis and Christof Koch. "The Unconscious Homunculus." Neural Correlates of Consciousness. Ed. Thomas Metzinger. Cambridge: Bradford/MIT Press, 2000. 103-110.
Edelman, Gerald M., and Giulio Tononi. A Universe of Consciousness: How Matter Becomes Imagination. New York: Basic Books, 2001.
Gray, Jeffrey A. Consciousness: Creeping Up on the Hard Problem. New York: Oxford University Press, 2004.
Harnad, Steve. "The Symbol Grounding Problem." Physica D 42 (1990): 335-346.
Jack, Anthony I., and Time Shallice. “Introspective Physicalism” The Cognitive Neuroscience of Consciousness. Ed. Stanislas Dehaene, Cambridge: Bradford/MIT Press, 2001. pp. 161-196.
Jacquette, Dale. "Adventures in the Chinese Room." Philosophy and Phenomenological Research 1989. Retrieved 12 Nov 2005 <http://ezproxy.tcnj.edu:2065/cgi-bin/jstor/printpage/00318205/di975025/97p0136m/0.pdf?backcontext=page&dowhat=Acrobat&config=jstor&userID=9f5b0fbe@tcnj.edu/01cc993341251fa1078a6caed5&0.pdf>.
Kugel, Peter. "How Long Will Computers Stay "Dumb"?" MIS Quarterly 12.1 (1988): 1-3. 285-306.
Metzinger, Thomas. “The Subjectivity of Subjective Experience” Neural Correlates of Consciousness. Ed. Thomas Metzinger. Cambridge: Bradford/MIT Press, 2000.
Moody, Todd C. “Distinguishing Consciousness.” Philosophy and Phenomenological Research 1986. Retrieved 13 Nov 2005 http://ezproxy.tcnj.edu:2065/view/00318205/di975015/97p05232/0?currentResult=00318205%2bdi975015%2b97p05232%2b0%2cFF&searchUrl=http%3A%2F%2Fwww.jstor.org%2Fsearch%2FBasicResults%3Fhp%3D25%26si%3D1%26Query%3Ddistinguishing%2Bconsciousness
Parsons, Lawrence M. “Exploring the Functional Neuroanatomy of Music Performance, Perception, and Comprehension.” Annals of the New York Academy of Sciences 930 (2001): 211-231.
Revonsuo, Antti. "Prospects for a Scientific Research Program on Consciousness." Neural Correlates of Consciousness. Ed. Thomas Metzinger. Cambridge: Bradford/MIT Press, 2000. 57-76.
Sampson, Geoffrey. "In Defence of Turing." Mind 82.328 (1973): 592-594.
Searle, John R. "Minds, Brains, and Programs." The Behavioral and Brain Sciences 1980. 13 Nov 2005 <http://members.aol.com/NeoNoetics/MindsBrainsPrograms.html>.
Seth, Anil K., and Bernard J. Baars. "Neural Darwinism and Consciousness." Consciousness and Cognition 14 (2005): 140-168.
Tononi, Giulio. "An Information Integration Theory of Consciousness." BMC Neuroscience 5. Retrieved 11 Oct 2005 <http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=543470>.
Torrance, Steve. “Producing Mind.” Journal of Experimental & Theoretical Artificial Intelligence 2000 http://weblinks1.epnet.com/externalframe.asp?tb=1&_ua=bo+B%5F+shn+1+db+buhjnh+bt+TD++%22B9N%22+7053&_ug=sid+7E35B0BA%2D898A%2D4FF3%2D86BE%2D59DC92B7720B%40sessionmgr2+dbs+buh+cp+1+7C63&_us=hd+False+fcl+Aut+or+Date+frn+1+sm+KS+sl+%2D1+dstb+KS+ri+KAAACBZD00005557+2364&_uh=btn+N+6C9C&_uso=st%5B0+%2DJN++%22Journal++of++Experimental++%26++Theoretical++Artificial++Intelligence%22++and++DT++20000701+tg%5B0+%2D+db%5B0+%2Dbuh+op%5B0+%2D+hd+False+2064&fi=buh_4106878_AN&lpdf=true&pdfs=261K&bk=R&tn=8&tp=CP&es=cs%5Fclient%2Easp%3FT%3DP%26P%3DAN%26K%3D4106878%26rn%3D8%26db%3Dbuh%26is%3D0952813X%26sc%3DR%26S%3DR%26D%3Dbuh%26title%3DJournal%2Bof%2BExperimental%2B%2526%2BTheoretical%2BArtificial%2BIntelligence%26year%3D2000%26bk%3D&fn=1&rn=8
Tunney, Richard J., and David R. Shanks. “Does opposition logic provide evidence for conscious and unconscious processes in artificial grammar learning?” Consciousness and Cognition 2003. Retrieved 13 Nov 2005 http://www.sciencedirect.com/science?_ob=MImg&_imagekey=B6WD0-48KVPDV-3-S&_cdi=6752&_user=1086025&_orig=browse&_coverDate=06%2F30%2F2003&_sk=999879997&view=c&wchp=dGLbVtz-zSkzV&md5=dc2c4fea599371dcab533573d3c8edda&ie=/sdarticle.pdf
Turing, Alan M. "Computing machinery and intelligence." Mind 1950. Retrieved 13 Nov 2005 <http://www.loebner.net/Prizef/TuringArticle.html>.
Zeman, Adam. "Consciousness." Brain 124 (2001): 1263-1289.
On the Improbability of Instantiating Consciousness through Artificial Means
With the exponential advances being made in technology, it would appear that the realization of artificial intelligence is not so unreasonable of an aspiration, nor too distant into the future at that. Sophisticated systems technologies that are able to interact with their environment have already been built and robots are able to simulate human movement to increasingly more sensitive degrees. However, AI researchers have become wary of declaring any existing machine to be artificially intelligent or conscious—a machine has yet to pass the much-lauded Turing test, whereby a computer program would convince an interrogator that it is a human being. In fact, there still remains the question of how perception becomes intelligence, which becomes consciousness, in the human brain itself; one might then ask how it could be possible to first create a program, which would in turn explain the mechanisms that drive the former. Subsequently, we are limited to theorization as to how consciousness could possibly arise, and to the possibility that it can be successfully replicated outside of the human mind. In this essay, I shall seek to invalidate such claims, made by Turing and Jacquette in particular, that artificial intelligence and consciousness could be created, as they are embedded in faulty philosophy and logic, and which, as current research in neuroscience has shown, would require an unfathomable amount of complexity and sensitivity.
In his essay, “Computing Machinery and Intelligence,” Alan Turing proposes that the question “Can machines think?” is too vague, instead suggesting a test whereby a behaviorist qualification for artificial intelligence could be made. The Turing test involves two subjects, a human being and a computing machine, and a human interrogator, whom they must convince of their own humanity. One can deduce that not only would this require an extensive database of response sets, but such persuasive capabilities that would suggest the machine can present itself as something more “human” than the other. Assuming that this test would suffice in determining intelligence, this would already be a daunting task. Unfortunately, the Turing test is too insensitive to the demands made upon organisms in their conscious experience; there is no consider of the sensory perception which ultimately provides the being in question with its tools for intelligence. Turing never affirms the possibility of creating a “thinking” machine, let alone one that could reproduce consciousness, his basis remains this test, which ultimately appears to merely try and simulate human behavior, human thought .
This is exactly Searle’s contention with the Turing test; while he has no objection to what he calls ‘weak’ AI, which simply aims to model the human mind, Searle argues that the notion of creating ‘strong’ AI—or computers which can actually be considered “conscious”—is unrealistic. In reference to Roger Schank’s program which purportedly understands stories and can infer various things from the given facts, Searle offers his “Chinese room” analogy, which criticizes the functionalist concept that the means by which processing of input creates an output is arbitrary. His reasoning is that simply because such processing may appear to produce a successful and coherent output, one cannot deduce that there is actual understanding of what transpires . Understanding is synonymous with consciousness, which, Searle argues, presupposes that the subject will act with intentionality:
[T]he main point of the present argument is that no purely formal model will ever be sufficient by itself for intentionality because the formal properties are not by themselves constitutive of intentionality, and they have by themselves no causal powers except the power, when instantiated, to produce the next stage of the formalism when the machine is running.
In addition to the fact that formal properties are not sufficient for intentionality, it is commonly held in the field of neuroscience that consciousness in an organism is characterized by ever-changing conscious states. There are a few errors in the consistency of his argument, however; while he concludes that intentional states are “defined in terms of their content, not their form,” earlier in his essay he remarks that it is because of a distinct biological structure that he is able to partake in intentional phenomena .
While Jacquette does not include this contradiction in his counterargument to Searle, he does bring into question the effectiveness of the “Chinese room” analogy on the basis of whether it operates on the same level of isomorphism as a computer program would. That is to say, if Searle’s analogy is not compatible with the mechanisms by which computer programs operate, then in turn, there is still a possibility of synthesizing an understanding machine. Jacquette asserts that the analogy does not function on the same level as a computer program would, and thus, since Searle envisions himself as a homuncular agent within the room, he incorrectly assumes that there is no understanding within the system as a whole. Additionally, he believes that intentionality is an abstract, a Universal, thus not physically bound to such “wet circuitry,” which is merely an accident. He concludes that Searle’s claim that consciousness, or rather intentionality, is irreducible, analogously to wetness and elasticity, is unjustified unless “he can offer a clearcut case of a decidedly nonintentional property caused by and realized in but not reducible to a material microphysical structure.” In disproving Jacquette’s remarks, although the validity of Searle’s argument in its entirety will not be restored, it will allow for his more cogent remarks to be evaluated from an empirical perspective.
The pervading theme of functional isomorphism on a micro- and macro-level accurately describes the dilemma presented by contrasting theories of artificial intelligence. It is a pertinent question; what must be replicated in order to successfully create consciousness in a machine? Unfortunately, Jacquette’s own reasoning sheds light on the fallacy of functionalism. In addition to the fact that functionalism assumes a dualistic perspective, that function exists as something distinct from form, as Searle points out (although functionalists would deny this in stating that the philosophy actually aims to bind mental states to functional states of physical system ) it fails to take into account the micro-level processing from which consciousness precipitates. In alignment with the linearity of the functionalist, input-process-output concept, I will offer an analogy: Suppose that two individuals plan on making a journey across the country; one by foot, one by automobile. What functionalism fails to take into account are the vastly different means by which the journey is made—so long as the two individuals are present at their destination, they would declare that the two journeys are acceptable isomorphs. There can be no disagreement about the compatibility in terms of their proper levels, as the concept is merely the A to B process.
To bolster this analogy, one can even remove the individual from the automobile and have a computer program navigate the vehicle to its proper destination. Thus, we have two distinct entities, and two distinct processes. Still, in spite of any attempts which would be made on behalf of the functionalist argument, it is clear that two very different phenomena are underway. The combustion of the gasoline to propel the pistons that power the engine, which in turn set the wheels rotating on their axis—it is quite different from the digestion of nutrients of the body which creates the adenosine triphosphate (ATP) that powers the cells, which allow from the deployment of acetylcholine to the muscular junctions that allow for locomotion. Clearly, there are processes within processes that determine the specific mechanisms by which the observable behavior is executed, and it is with this same sensitivity that artificial consciousness and intelligence ought to be approached. However, before one is able to create an artificially conscious machine, the neurophysiological correlates of consciousness must be defined and acknowledged. From this evidence, the mere possibility of AI will be assessed—and as the following sources will demonstrate, the likelihood is not very strong!
The most commonly accepted theory for explaining the emergence of consciousness from mental states would be the “dynamic core” hypothesis, proposed by Gerald Edelman and Giulio Tononi, which states that consciousness has three physiological requirements: differentiation, integration, and reentry . A highly differentiated system implies that its substructures all serve unique purposes, which, when integrated, allow for the coherent interpretation of perceptual stimuli. The means by which such a system can be both highly differentiated and integrated at once is reentry; which Edelman describes as an “ongoing, recursive interchange of parallel signals between reciprocally connected areas of the brain, an interchange that continually coordinates the activities of these areas’ maps to each other in space and time.” He explains that reentry is unlike feedback in that there are many parallel paths, and is developed and reinforced selectively, instead of through unidirectional alteration. This plasticity of the brain, which is determined by selection and allows for consciousness to arise, is emphasized by Edelman and Tononi; the brain changes on a daily basis—in fact, conscious experience affects the systems of our brains (which would refute epiphenomenalism). Neurons constantly die and synapses are always being rearranged.
The mind is not based on systems of logic but selectional systems, just as a child is not born with a sense of logic, but instead learns through an experiential trial-and-error process. But before confronting the issue of the acquisition of intelligence, let us continue to formally define consciousness. Zeman attributes the following characteristics to conscious experience: that it is personal; its contents are only stable for short periods, lasting up to a few seconds at most; its contents are unified and continuous over time; it is selective, with a limited capacity . Lastly, he mentions that conscious states are, in fact, intentional—intentionality being a phenomenon which Jacquette criticized for being too abstract and baseless, a decade prior to such research. Among the unique events within the human brain which are involved in consciousness, Zeman describes the parts of the brain, such as V4 and V5, which process color and movement, respectively . The particularity of function of neurons in the occipital lobe suggests that there is some truth to qualia, in that the brain does in fact create a complete picture out of the many properties sensory experience provides. Studies have shown that melody and rhythm are likewise processed in different parts of the brain.
Another unique mechanism is the ability for the subject to recreate images, events, and locations mentally, in the “mind’s eye.” Zeman cites studies that have found that mental rotation tasks and envisioning a route to a destination are both proportional to the time that would be required to enact such tasks in reality . A final example from Zeman is his reference to the decrease in engagement and global brain activity in the acquisition of new skills . All of these examples demonstrate how the parts of the brain work together, through integration, as a master ensemble in order to facilitate consciousness. Giulio Tononi extensively describes how integration is a necessary component for the complex states that are the constituents of consciousness; establishing his hypothesis as follows:
If consciousness corresponds to the capacity to integrate information, then a physical system should be able to generate consciousness to the extent that it has a large repertoire of available states (information), yet it cannot be decomposed into a collection of causally independent subsystems (integration).
He subsequently explains that in order for a system to be conscious, it must be composed of complexes, or rather, consist of a main complex which is necessary and sufficient for the integration of information—nothing outside of the complex is capable of being integrated. That is not to say, however, that we are conscious of everything that occurs within the complex; the many aforementioned processes, the qualia, are not things we can be consciously aware of. We cannot focus on any particular aspect of our perceptual field—a reality which quite well demonstrates a “nonintentional property caused by and realized in but not reducible to a material microphysical structure” that Jacquette wished for Searle to provide, which has been done in his stead; sensory perception is by no means intentional, we do not actively engage in perceiving the world about us, and reducing it past what can be consciously perceived results in an explanation of physical processes—but no distinct locus.
Yet all this refutation of Jacquette’s statement has established is that intentionality exists. Should intentionality be necessarily attributed to consciousness? Or rather, can consciousness exist independent from intentionality? This issue is of no concern in the question of artificial consciousness, and for this reason alone: the only way in which AI researchers could absolutely determine that they have artificially created a conscious machine is by comparing it to the only known substrate which has facilitated consciousness in a form which can be accurately accounted for—the human being. While the exact relationship between intentionality and consciousness remains to be seen, if one admits the fact that the former is a necessary component of the latter, it should follow that in creating a conscious machine, intentionality must be observable in its conscious states as well . Thus, it is not sufficient for the trend in modern research to attempt in creating mere consciousness, without aim, but to consider that these artificially intelligent machines must originate from a familiar mechanism or process that can be causally reduced to identifiable components—they must be instantiated from a consciousness like ours.
One might offer that there is no possible means of determining whether a machine is conscious, let alone that it is conscious in the same way humans are conscious, but such solipsistic mentality can be dismissed primarily because it would be as absurd as asking whether one can be entirely sure that any other human being apart from us is subject to certain, unifying experiential qualities. Additionally, this “consciousness like ours” position may be met with claims that such a high standard for judgment puts AI research up against a nearly impossible task. There is a particular reason for this; for those who would argue that a conscious machine does not need to resemble any existing anatomical form, they must be reminded that consciousness includes self-consciousness, which can only be facilitated through a strong sense of spatial boundaries . It must be aware of where the self ends and the rest of the world begins. Furthermore, it must be able to interact with the world in many different ways, and to gather information from this world through the five senses. Once again, such a statement would be met with contention, such as: “Why must a conscious machine have to have all the senses? Aren’t there people who lack one sense or another, yet are still able to be identified as a conscious being?” To this I would reply: Isn’t the goal of artificial intelligence to come to understand human consciousness and acquisition in its entirety? Even if there are human beings who may lack the sense of smell, or taste, or touch, what relevance does this have to the study which aims to determine the mechanisms that are underway within a healthy, fully functional individual? There is no sense in building a handicapped robot. Thus, in addition to the “consciousness like ours” argument, we have raised another qualification for AI research; that it must be a complete form of consciousness, equipped with the same functions that evolution has provided for organisms.
From a structural perspective, referring back to the concepts of neural plasticity and the dynamic core, it is clear that this conscious machine would have to be built in a manner such that, in addition to having a vast option for forming networks or pathways, it would be able to continually alter these networks as experience dictates within the brain itself. Consequently, a simple geometric layout would not suffice in offering the greatest amount of information integration possible, since the bypass systems which would be required to allow for a unit of information to travel from microprocessor α to travel to microprocessor γ, without interference from microprocessor β (which, perhaps in this scenario, forwards directly to ω, eliminating the subsequent algorithms instantiated from γ, resulting in an altogether different effect), would require yet another framework, an inhibitory one, to be superimposed upon, or built into the standard processing unit. More broadly speaking, there can be nothing which records and permanently stores all information, particularly visual perception, as consciousness is a selective process . Thus, there must also be some way in which the machine could direct its focus. Of the currently existing robots which are able to successfully process images, their responses are limited to whatever stimuli they have been programmed to respond to, either colors or particular faces.
Here it is important to emphasize the importance that a truly conscious machine cannot be preprogrammed—as consciousness is something entirely personal, I would go so far as to state that the only inbuilt processes a conscious machine may have are the senses. If a machine is programmed, that implies that it can be reproduced by the programmer who created it. Unfortunately, there is a clear distinction between the manual programming that results in artificial intelligence (by every sense of the word “artificial”) and the ‘passive’ programming a child experiences during their early development, which occurs through interaction with their environment, so the former would not be representative of personal experience. This raises another limitation for AI research: a conscious machine will have to be provided with the mechanisms required for obtaining information, but by no means may it be equipped with any “adult” skills; the robot must be a child. What this implies is that the robot must be able to establish some sort of value system, whereby its experiences are categorically stored based on some form of threshold- or frequency-sensitive basis. For example, in order for a significant event to be stored under the category of “traumatic experience,” certain qualifications must be met, which may be allowed to have been preprogrammed. This subsequently raises the question of how human beings themselves experience emotions, and how memories are influenced and reinforced by our emotions.
Consciousness has already shown to have emerged from biological processes, such as natural selection and evolution, as well as the principles that dictate the structure of the brain itself, such as neuronal selection. Evolutionarily speaking, then, it is clear that the advantages of consciousness as a social mechanism—that is, a way of forming strong relationships with other individuals—play an equally important role in determining the requirements for this conscious machine . If the goal of self-consciousness, in addition to the survival of the organism, is to be able to comprehend the mental states of other individuals, then the “consciousness like ours” argument becomes more reliable. But how is it possible to create a conscious machine that can identify with human emotions? Or is it even possible?
Let us review the qualifications made for the construction of a truly conscious machine: that it should have a consciousness like ours; that its consciousness be fully functional and comparable to that of a healthy human being; and that it should be equipped with mechanisms that enable it to acquire information, but no preprogrammed systems for the semantics and syntax of language, logical reasoning, etc. It appears to me that such requirements not only substantially limit the possibility of creating artificial intelligence, but in all probability completely eliminate such a possibility. The Turing test is in complete antithesis to such claims, despite Turing’s own assertion that a thinking machine would have to be constructed like a child—there is no basis for determining consciousness in his test. If, however, one was to maintain the Turing test as the standard for AI research, then I would suggest that, in addition to being able to fool the interrogator into believing it is a human, it should perform equally well as the interrogator itself in Turing’s test! Consciousness involves making judgments, which is entirely missing from Turing’s thesis; this corollary aims to incorporate this obvious necessity so as to further qualify the requirements for a conscious machine.
Furthermore, in having asserted that functionalism and its correlated philosophies (computationalism, connectivism, and other forms of materialism ) are errant in incorrectly assuming the mechanisms that drive our own consciousness can be reproduced outside of the brain, it is clear that artificial consciousness is impossible. Every particular substance and process has its own unique constituents, and while one may come close to replicating its properties, it will always be simulation, and nothing more. Additionally, emphasizing that consciousness is personal; to build a conscious machine would imply that there is a particular center which can theoretically be transplanted to other robots or even mass-produced—there is nothing personal about that! To be sure, neuroscience has not even come close to determining exactly how consciousness operates; yet AI researchers maintain that through a bottom-up approach this question will be solved. The only way to approach the study of consciousness is from the top down, as it is currently being done by cognitive neuroscientists, and this should be obvious, considering the way in which consciousness functions as a whole—that is, within a group of sublevel processes and interactions between neuronal groups. While human beings are programmed biologically to be human, and are predisposed to certain behaviors, this by no means can be considered to be isomorphic to the programming of machines. There is some aspect of personality which remains to be understood, yet which ultimately distinguishes man from machine. It is best to assume that while tinkering with technology may produce interesting novelties, determining the bases for human behavior is best left for human examples.
References:
Aleksander, Igor. “A Neurocomputational View of Consciousness.” From Brains to Consciousness? Essays on the New Sciences of the Mind. Ed. Steven Rose. Princeton: Princeton University Press, 1999. 180-199.
Aydede, Murat and Guven Guzeldere. “Consciousness, Intentionality, And Intelligence: Some Foundational Issues for Artificial Intelligence.”
Journal of Experimental & Theoretical Artificial Intelligence.
Baars, Bernard J. "The Conscious Access Hypothesis: Origins and Recent Evidence." TRENDS in Cognitive Sciences 6.1 (2002): 47-52.
Crick, Francis and Christof Koch. "The Unconscious Homunculus." Neural Correlates of Consciousness. Ed. Thomas Metzinger. Cambridge: Bradford/MIT Press, 2000. 103-110.
Edelman, Gerald M., and Giulio Tononi. A Universe of Consciousness: How Matter Becomes Imagination. New York: Basic Books, 2001.
Gray, Jeffrey A. Consciousness: Creeping Up on the Hard Problem. New York: Oxford University Press, 2004.
Harnad, Steve. "The Symbol Grounding Problem." Physica D 42 (1990): 335-346.
Jack, Anthony I., and Time Shallice. “Introspective Physicalism” The Cognitive Neuroscience of Consciousness. Ed. Stanislas Dehaene, Cambridge: Bradford/MIT Press, 2001. pp. 161-196.
Jacquette, Dale. "Adventures in the Chinese Room." Philosophy and Phenomenological Research 1989. Retrieved 12 Nov 2005 <http://ezproxy.tcnj.edu:2065/cgi-bin/jstor/printpage/00318205/di975025/97p0136m/0.pdf?backcontext=page&dowhat=Acrobat&config=jstor&userID=9f5b0fbe@tcnj.edu/01cc993341251fa1078a6caed5&0.pdf>.
Kugel, Peter. "How Long Will Computers Stay "Dumb"?" MIS Quarterly 12.1 (1988): 1-3. 285-306.
Metzinger, Thomas. “The Subjectivity of Subjective Experience” Neural Correlates of Consciousness. Ed. Thomas Metzinger. Cambridge: Bradford/MIT Press, 2000.
Moody, Todd C. “Distinguishing Consciousness.” Philosophy and Phenomenological Research 1986. Retrieved 13 Nov 2005 http://ezproxy.tcnj.edu:2065/view/00318205/di975015/97p05232/0?currentResult=00318205%2bdi975015%2b97p05232%2b0%2cFF&searchUrl=http%3A%2F%2Fwww.jstor.org%2Fsearch%2FBasicResults%3Fhp%3D25%26si%3D1%26Query%3Ddistinguishing%2Bconsciousness
Parsons, Lawrence M. “Exploring the Functional Neuroanatomy of Music Performance, Perception, and Comprehension.” Annals of the New York Academy of Sciences 930 (2001): 211-231.
Revonsuo, Antti. "Prospects for a Scientific Research Program on Consciousness." Neural Correlates of Consciousness. Ed. Thomas Metzinger. Cambridge: Bradford/MIT Press, 2000. 57-76.
Sampson, Geoffrey. "In Defence of Turing." Mind 82.328 (1973): 592-594.
Searle, John R. "Minds, Brains, and Programs." The Behavioral and Brain Sciences 1980. 13 Nov 2005 <http://members.aol.com/NeoNoetics/MindsBrainsPrograms.html>.
Seth, Anil K., and Bernard J. Baars. "Neural Darwinism and Consciousness." Consciousness and Cognition 14 (2005): 140-168.
Tononi, Giulio. "An Information Integration Theory of Consciousness." BMC Neuroscience 5. Retrieved 11 Oct 2005 <http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=543470>.
Torrance, Steve. “Producing Mind.” Journal of Experimental & Theoretical Artificial Intelligence 2000 http://weblinks1.epnet.com/externalframe.asp?tb=1&_ua=bo+B%5F+shn+1+db+buhjnh+bt+TD++%22B9N%22+7053&_ug=sid+7E35B0BA%2D898A%2D4FF3%2D86BE%2D59DC92B7720B%40sessionmgr2+dbs+buh+cp+1+7C63&_us=hd+False+fcl+Aut+or+Date+frn+1+sm+KS+sl+%2D1+dstb+KS+ri+KAAACBZD00005557+2364&_uh=btn+N+6C9C&_uso=st%5B0+%2DJN++%22Journal++of++Experimental++%26++Theoretical++Artificial++Intelligence%22++and++DT++20000701+tg%5B0+%2D+db%5B0+%2Dbuh+op%5B0+%2D+hd+False+2064&fi=buh_4106878_AN&lpdf=true&pdfs=261K&bk=R&tn=8&tp=CP&es=cs%5Fclient%2Easp%3FT%3DP%26P%3DAN%26K%3D4106878%26rn%3D8%26db%3Dbuh%26is%3D0952813X%26sc%3DR%26S%3DR%26D%3Dbuh%26title%3DJournal%2Bof%2BExperimental%2B%2526%2BTheoretical%2BArtificial%2BIntelligence%26year%3D2000%26bk%3D&fn=1&rn=8
Tunney, Richard J., and David R. Shanks. “Does opposition logic provide evidence for conscious and unconscious processes in artificial grammar learning?” Consciousness and Cognition 2003. Retrieved 13 Nov 2005 http://www.sciencedirect.com/science?_ob=MImg&_imagekey=B6WD0-48KVPDV-3-S&_cdi=6752&_user=1086025&_orig=browse&_coverDate=06%2F30%2F2003&_sk=999879997&view=c&wchp=dGLbVtz-zSkzV&md5=dc2c4fea599371dcab533573d3c8edda&ie=/sdarticle.pdf
Turing, Alan M. "Computing machinery and intelligence." Mind 1950. Retrieved 13 Nov 2005 <http://www.loebner.net/Prizef/TuringArticle.html>.
Zeman, Adam. "Consciousness." Brain 124 (2001): 1263-1289.