Sunday, August 21, 2016
From Consciousness to Intelligence
“The brain is already a sleight of hand, a massive, operationalist shell game. It designs and runs Turing Tests on its own constructs every time it ratifies a sensation or reifies an idea. Experience is a Turing Test – phenomena passing themselves off as perception’s functional equivalents.”
~Richard Powers, Galatea 2.2
Is intelligence distinct from the unconscious and non-intentional processes that underlie cognitive function? When we administer IQ tests, do we actually measure intelligence? Is intelligence “liftable,” to borrow a term from Douglas Hofstadter – can it be “lifted” from one substrate and installed into another – or is it system-specific, contingent upon the parameters dictated by a formal system? Is intelligence a particular quality of an organism or system, isolatable to the conventional embodied structure of a given formal system, or is it an effect of the organism/system’s relationship to its environment – of its own recursive structure? Is intelligence the ability to operate logically within a given set of axioms, or is it the capacity to produce novel axioms via the pursuit of isomorphic patterns with the external world – through tools, instruments, media, the interfacial film that reflects its existence back to it…?
These are the ridiculous and probably ill-conceived questions that I grapple with, but the more I explore these ideas in the appropriate literature (fiction and nonfiction) the more I’m convinced that I’m not entirely crazy.
In short, I’m interested in the following questions: when we talk about intelligence in humans, do we conflate intelligence with consciousness? Are consciousness and intelligence related – and if so, how? Does intelligence mean consciousness; or, alternatively, does consciousness mean intelligence? These questions require so many qualifications that it’s impossible to give straightforward yes or no answers, but the inquiry into these indeterminacies is as rewarding (if not more so) than any definitive answer we could hope to give. So, here is a very preliminary, and very amateur, attempt to foreground some of these questions. Fair warning: I’m a literature PhD, I have no formal training in neuroscience, cognitive science, computer science, artificial intelligence, et al. But I like to think I’ve encountered a fair amount of critical examination of these questions in my studies – enough at least to warrant a random blog post on the internet. So, here goes.
First, a disclaimer: if there’s one compelling notion that cognitive studies has illuminated, it is that the postmodern anxiety over logical grounds may not be so ill-founded after all. Logical ground entails a premise, formula, or axiom that requires no prior proof. In the wake of Gödel’s incompleteness theorems, the promise of logical grounds seems asymptotically distant – barely visible on the horizon, and never quite in reach. Yet we have to begin somewhere, inaugurating a perpetual series of rejoinders from our critics, who cry in dismay, “But you’re assuming ‘x’!” And they’re right. I am making assumptions. I’m making assumptions all over the place; but then, I’m fine with being an ass.
My primary assumption is that consciousness is, quite obviously, not a centrally organized or isolatable phenomenon. Consciousness is an effect, rather, of brain processes far too complex to rationalize via consciousness itself. In other words, consciousness is an escape mechanism: a way of avoiding the supreme complexity of what exactly is going on inside our heads. To give you an idea of the kind of complexity I’m talking about, I’ll refer to one of the wizards of modern sociobiology, Edward O. Wilson. In his now famous book, Consilience: The Unity of Knowledge (1998), Wilson offers the following description of the processes that underscore conscious thought:
Consciousness consists of the parallel processing of vast numbers of […] coding networks. Many are linked by the synchronized firing of the nerve cells at forty cycles per second, allowing the simultaneous internal mapping of multiple sensory impressions. Some of the impressions are real, fed by ongoing stimulation from outside the nervous system, while others are recalled from the memory banks of the cortex. All together they create scenarios that flow realistically back and forth through time. The scenarios are a virtual reality. They can either closely match pieces of the external world or depart indefinitely far from it. They re-create the past and cast up alternative futures that serve as choices for future thought and bodily action. (119-120)
In other words, consciousness provides the human subject with a virtual experience of the material world; it is an interface between interior qualia and exterior phenomena. It is the internal form that our relation to the world takes: “Conscious experience,” as Thomas Metzinger says, “is an internal affair” (21). It is the way we imaginarily fashion the nonhuman world.
Proceeding from this initial premise, we can arrive at a quite obvious conclusion: that consciousness does not necessarily imply intelligence, at least as intelligence is defined by IQ tests. If consciousness implied intelligence, then there would be no need to test for intelligence among obviously conscious human subjects. Rather, IQ tests would reflect disparities in intelligence between humans and other species – apes, birds, insects, etc., but not disparities in intelligence among humans, unless we are willing to admit that some human beings are not conscious. These human beings would be, according to philosophical tradition, zombies: “a human who exhibits perfectly natural, alert, loquacious, vivacious behavior but is in fact not conscious at all, but rather some sort of automaton” (Dennett, Consciousness Explained 73). According to this definition, however, a philosophical zombie would yield intelligent results despite being a non-conscious entity. It would exhibit intelligent behavior. How are we to square this? If an IQ test assumes consciousness on the part of its examinees, how can a non-conscious entity exhibit intelligent behavior? Are we forced to admit that such behavior is not actually intelligent, but merely superficially intelligent – like a group of monkeys who happen to write The Tragedy of Hamlet, Prince of Denmark? Such a claim constructs intelligence as a substance, something that can be genuinely isolated and identified, and can be opposed to false intelligence – seemingly intelligent behavior that actually lacks the rudiments of intelligence itself, intelligence tout court. But we cannot do this – all we have to judge intelligence by is behavior, not by any privileged access to an interiority that exposes its immediate authenticity.
IQ tests operate on the assumption that intelligence is a genuine and measurable quality, that it conforms to specific aspects of material existence. But in fact, many of these aspects of existence cannot be verified by IQ tests, but are only applied in retrospect. That is, IQ tests assume that their examinees are conscious agents, and apply consciousness to their examinees, but IQ tests cannot test for consciousness, as made clear by the philosophical zombie scenario – they can only test, purportedly, for intelligence. An advanced computer can pass an IQ test. All of this would seem to suggest that intelligence is not a quality per se, not something that inheres substantially within a conscious subject, but rather an effect of behavior. So when we examine human subjects for intelligence, and quantify their performance with a number, what are we actually saying? If we’re not isolating and substantiating some quality of conscious experience, then what are we doing?
Let’s take a moment and assess what examinees do when they take an intelligence test: they are presented with problems, composed of specific elements of information, and are asked to parse these problems. This is not an official definition, nor am I citing any source. This is my own definition: IQ tests present examinees with particular elements of information, ask them to analyze this information, and produce new equivalences of this information. Clearly, I’m stacking the deck; for if intelligence is simply constructing equivalences between informational elements, then intelligence actually has little to do with any interior capabilities or substances. It has to do only with the relational structure that emerges between bits of information – that is, it emerges in the pattern, or isomorphism, that mediates information. Intelligence is not quality, nor substance, but effect. Human subjects are not “intelligent” in any possessive sense, but only in a behavioral sense; and in this regard, even the most clueless human beings can hypothetically fake their way through intelligence tests. In fact, we can take this one step further, beyond the limits of the human: even non-consciouscomputer programs can pass IQ tests with flying colors.
So when Alan Turing proposed his perpetually fascinating inquiry into machine intelligence – “Can machines think?” – what does he mean? What would it mean for a machine to think? According to our discussion above, it would not mean for a machine to possess consciousness (necessarily); rather, for a machine to “think” it would merely have to exhibit evidence of thought. The implications that build up around this proposal are unsettling – for if a machine merely pretends to think, is it not actually thinking…? And if this distinction no longer holds, then are we – us human subjects, strung along this mortal coil – also merely pretending to think? What is the self that only pretends to think? Is it even a “self” at all?
One of the most serious objections to the proposal of machine intelligence that Turing entertained was the “argument from consciousness,” which Hofstadter helpfully articulates in his masterful work, Gödel, Escher, Bach: an Eternal Golden Braid (1979): “Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain” (597). Fair enough, objectors – perhaps intelligence cannot be dissociated from the capacity to create art. Yet as we have seen, and as recent developments in computer science (more recent than 1979) have shown, machines can compose poetry. They can reorganize words in novel ways, in ways that may even engender emotions among their human readers. Furthermore, we know that computers may be able to perform such tasks without even understanding much of the semantic content of the words themselves – John Searle suggests this very point in his Chinese Room scenario, in which he argues that an artificial system can be programmed to engage in conversation without actually understanding the content of the conversation. Certainly, and as Searle declares, this means that the program does not possess understanding, or consciousness… but these are resolutely human conceptions of being, and have little bearing on models of intelligence. For if, as we’ve already shown, intelligence adheres in behavior, not in substance, then a Chinese Room may not be conscious, but it can damn well be intelligent.
“But wait!” the objectors proclaim, “there’s more!” Fine, let’s hear what they have to say. Mr. Hofstadter, if you please…? Of course, as he says, not until a machine writes a sonnet can we agree that it has achieved the level of “brain,” but there’s another element to the theorem:
“that is, not only write it but know that it has written it” (597; emphasis mine).
Ah, yes… clever. As Hofstadter relates, Turing’s objectors insist that a machine cannot be intelligent not only unless it composes a sonnet, but unless it knows that it composes a sonnet. For the time being, let’s grant that this is a fair objection. So now, I ask: how do we know that it knows? Why, let’s ask it:
Objectors: “Machine, do you know that you’ve just composed a sonnet?”
Let’s imagine that the machine replies: “Yes, I do.”
Assuming the machine answered thus, our concerned objectors would presumably acquiesce. The machine admits that it knows it composed a sonnet – it has become a brain. But how do we know the machine is telling the truth? After all, we’ve already witnessed that the machine has the capacity to compose believably plausible statements, to even craft poetic texts. Why do we assume that it can’t lie to us? Why are we convinced that its intelligence ends at pretense – that it cannot compose deception? Oscar Wilde associated art with the supreme aesthetic practice of lying; if a machine can compose a sonnet, then why can’t it lie to us?
Now, let’s not be unfair to our objectors. Let’s also imagine that the machine answers to the question of whether it knows it has produced a sonnet: “No, I do not.” But what does this mean? By saying it does not know that it has produced a sonnet, is it saying that it doesn’t know what a sonnet is? Or is it saying that it knows what a sonnet is and that it has not produced one? Assuming that the poem composed by the machine conforms to the rules of a sonnet, we have a couple options available to us: either it is unaware of what a sonnet actually is, or it has made a profound aesthetic judgment regarding the qualities of a sonnet. Either way, its response is ambiguous, and provides no definitive evidence as to its ignorance of the sonnet form. In an even more radical sense, perhaps the machine is making an uncanny comment not on the aesthetic qualities of a sonnet, but on the limitations of knowledge itself.
So, once again – what the hell is intelligence? Where do we locate it? Can we locate it? And, pressing our inquiry a bit further, what exactly does intelligence mean? In Gödel, Escher, Bach, Hofstadter provides the following speculative assessment of intelligence:
Thus we are left with two basic problems in the unraveling of thought processes, as they take place in the brain. One is to explain how the low-level traffic of neuron firings gives rise to the high-level traffic of symbol activations. The other is to explain the high-level traffic of symbol activation in its own terms – to make a theory which does not talk about the low-level neuronal events. If this latter is possible – and it is a key assumption at the basis of all present research into Artificial Intelligence – then intelligence can be realized in other types of hardware than the brains. Then intelligence will have been shown to be a property than can be “lifted” right out of the hardware in which it resides – or in other words, intelligence will be a software property. (358)
I know –awesome, right? If intelligence can be isolated from its intra-cranial hardware, can be distinguished and abstracted, and transposed into other systems… then suddenly we have a vision of intelligence that need not yield to consciousness, to that sacred and protected form of human experience. We have an intelligence, in other words, that exceeds consciousness.
Yet there is something deeply troubling about this proposal; for even if we limit intelligence in humans to behavioral patterns – that is, to how humans act – this behavior still must be associated with the dense, gray matter of the three-pound organ balancing above our shoulders. Unless we want to reify intelligence into an abstraction, a formal system that subsists beyond its material components… we have to take into account its substructure. Suggesting that we can “lift” intelligence out of our brains and implant it (or, perhaps, perceive it) in other hardware systems assumes a certain limitability of intelligence – that it comprises an entire system in itself, a formal arrangement of operations. In contrast to this hypostatizing tendency, I want to argue that intelligence is a trans-formal mediation, an isomorphic pattern that dictates the interaction between systems or organisms. It is not the quality of a system, capable of being lifted and transplanted, but rather an interfacial effect, and therefore always in flux. Intelligence cannot be quantified or delimited, but only perceived. It is for this reason that we can perceive patterns of intelligence in human behavior, but we cannot isolate them as characteristics of an individual human subject. Isolating intelligence involves a conflation of intelligence with consciousness; indeed, going beyond that, it involves intelligence’s capitulation to consciousness. As is the case of the philosophical zombie, however, an individual subject may in fact be entirely devoid of consciousness and yet still exhibit intelligent behavior. This is because intelligence presents itself not in the illusory self-presence of a rational, conscious subject, but in the conformity between a subject’s behavior and its external environment.
This is not a matter of “lifting” intelligence out of its hardware, because its hardware is simply not isolated to the brain, nor to the body, nor to the hard drive. The hardware of intelligence materializes between entities, as the interface that mediates their relations. It is not a quality of entities, but an effect of systems. It is for this reason that we can, and should, talk about what has been popularly and professionally referred to as artificial intelligence – but as I hope is becoming clear, artificial intelligence is actually not “artificial” at all, at least not in the sense of false, or imitative, or untrue. It is, quite simply, intelligence tout court. Perhaps intelligence is not a genetic trait, but a technological phenomenon; it is not limited to the domain of biological life, but manifests between organisms as a technological prosthesis.
I am not the first to suggest that intelligence is technological, not genetic. This notion of intelligence can be detected throughout the tradition of communications theory since World War Two. In a short essay on technology, John Johnston traces the evolution of computer intelligence back to the necessity of new communications technologies and code-breaking during the Second World War:
For the first modern computers, built in the late 1940s and early 1950s, “information” meant numbers (or numerical data) and processing was basically calculation – what we call today “number crunching.” These early computers were designed to replace the human computers (as they were called), who during World War II were mostly women calculating by hand the trajectories of artillery and bombs, laboring to break codes, and performing other computations necessary for highly technical warfare.” (199)
Johnston’s comment highlights a detail of computer history that we all too often forget. When we hear the word “computer,” we typically make a semantic leap whereby we associate the word with a technical instrument; but originally, a computer simply meant someone who computes. Computers were, in their first iteration, human subjects. The shift to technical instruments in the place of human labor did not fundamentally alter the shape of intelligence as it manifested among these ceaselessly calculating women – it simply programmed the symbolic logic by which these original computers operated into the new technical instruments.
Human bodies, like computers, are information processors. “In short,” Johnston writes, “both living creatures and the new machines operate primarily by means of self-control and regulation, which is achieved by means of the communication and feedback of electrochemical or electronic signals now referred to as information” (200). We must recall, however, that Johnston is making a historical claim; that is, he’s attuned to the specificity of the postwar moment in providing the conditions necessary for aligning humans and machines. The war was pivotal in directing science and technology toward a cybernetic paradigm, which in turn allowed for a new perspective on intelligence to emerge. At first, this evolution in intelligence simply took the form of technical Turing machines – simple computers that could produce theorems based on an algorithm, essential glorified calculators – but eventually it transformed, blossoming into the field that we know today as artificial intelligence.
The entire premise of AI, as popularly defined and pursued, is a controversial and even contradictory one: controversial because there is intense disagreement over whether or not artificial intelligence has ever been attained – and if not, then whether or not it is even attainable. AI research is also contradictory, however, and its contradictions touch upon the primary subject of this post. It is contradictory because the achievement of intelligence in technological constructs undermines that very achievement, according to the way its proponents define intelligence: “There is a related ‘Theorem’ about progress in AI,” Hofstadter writes; “once some mental function is programmed, people soon cease to consider it as an essential ingredient of ‘real thinking.’ The ineluctable core of intelligence is always in that next thing which hasn’t yet been programmed. This ‘Theorem’ was first proposed to me by Larry Tesler, so I call it Tesler’s Theorem: ‘AI is whatever hasn’t been done yet’” (601). Hofstadter’s comment is revealing, and it draws the humanization of intelligence back into the discussion. That is, according to Tesler’s Theorem, experiments in AI pose functions performed by the brain as goals for computers; but as soon as computers are able to carry out these functions, they lose their importance for human brain function. They can’t actually be aspects of intelligent behavior, Tesler’s Theorem retorts, if a computer can do them!
The contradiction immediately presents itself – for if we’ve already presumed that real intelligence is behavior that a computer cannot perform, then why do we even bother experimenting with AI at all!? The entire enterprise becomes self-defeating. Tesler’s Theorem tells us that we need to reorient ourselves with respect to intelligence, we need to construct a new perspective on what intelligence actually is – for if we remain convinced that intelligence can only be human, then it’s pointless to even try programming intelligence in nonhuman substrates.
This is my point in emphasizing that intelligence isn’t a genetic trait, or quality of a specific structure of being. If we reimagine intelligence as an emergent phenomenon deriving from complex pattern-matching, then we open the door not only to legitimate experiments in AI, but an entirely new conceptualization of intelligence itself. Of course, such a proposal will inevitably invite resistance. Humans don’t like to separate their intelligence from their experience of it. It doesn’t only make us feel as though we’re different from other animals (which we are), but it makes us feel that our experiences are our own, that our intelligence is our own. In other words, it makes us feel that we’re not simply going through our lives pretending to be smart, or imitating intelligence (although, again, we all are – I pretend to be smart for a living). We feel the need to make our intelligence a part of our consciousness, as Dennett suggests: “Consciousness, you say, is what matters, but then you cling to doctrines about consciousness that systematically prevent us from getting any purchase on why it matters. Postulating special inner qualities that are not only private and intrinsically valuable, but also unconfirmable and uninvestigable is just obscurantism” (450). As Dennett convincingly intimates, many of us prefer that consciousness matters because we experience it – but then we succumb to a tautology befitting our desire for presence and immediacy. We claim that consciousness is important because it is mine, I am living it. It’s important because we have it.
But being able to say “I have consciousness” is consciousness. It’s a nasty little Ouroboros we’ve gotten ourselves into here. This, of course, is the hard problem of consciousness: explaining it begins to look uncanny when we really try and orient ourselves to it in a removed fashion. Who is this shambling, talking, possibly thinking thing saying it’s conscious? Why, it’s just me, of course… as far as you know. Which is what Dennett means when he says that consciousness is unconfirmable and uninvestigable. Of course, from my private and privileged perspective, it is absolutely confirmable – but you don’t have access to my consciousness, you can’t prove to yourself that I am conscious. Even if you hired a trained surgeon to remove my brain so you could inspect it, you won’t find my consciousness. It isn’t there. It’s only there for me.
So now, imagine that I’m not actually a conscious person, that I’m what we call a “philosophical zombie.” I can say all the right things, answer your questions in the right way… but I’m not conscious. Does this mean I’m unintelligent? Even if I have no human connection to the words you say, even if I have no intimate knowledge of the semantics of language… I possess the capacity to match patterns to such an extent that I can fake it. I can appear conscious. It’s the same with a specifically advanced computer system, a system that interprets theorems and algorithms and performs specific functions. At a certain point of complexity, the functions it performs might include carrying on a conversation regarding the aesthetic qualities of “Dover Beach,” or the ethical issues surrounding immigration – subjects that we consider fundamental to our humanness. If a machine matches our communicative capabilities, that may not make it conscious… but does this mean it isn’t intelligent?
My position on the matter is probably clear by this point, so I won’t hammer it into the ground anymore. I’ll only leave you with this final thought. You and I aren’t conversing face-to-face. You can’t see me, my human body; you can’t hear my words in the form of speech, you’re just reading them on a computer screen. Yet we assume each other’s presence because of the highly complex degree of our communication, of the pattern-matching and analysis required to interact via an amateur blog post on the topic of intelligence. And we likely feel quite comfortable in our positions, confident in our identification of the person across the interwebz. After all, it would be difficult for something to fake this level of linguistic communication, right? Yet there are programs out there that perform similar functions all across the web. They’re called (fittingly) bots, and some are even used to make posts on internet forums. Granted, it’s highly unlikely that a bot would make a blog post as extended and immaculately crafted as this one (why, thank you – ah, damn, this is quite a self-reflexive bot!); but even if one could, would this post be any less… intelligent? (oh please, you’re too kind)
Anyway, no need to fret. I’m not a bot. As far as you know… ;-)
(at least… I think I’m not)
Dennett, Daniel. Consciousness Explained. New York: Back Bay Books, 1991. Print.
Hofstadter, Douglar R. Gödel, Escher, Bach: an Eternal Golden Braid. 1979. New York: Basic
Books, 1999. Print.
Johnston, John. “Technology.” Critical Terms For Media Studies. Eds. W.J.T. Mitchell and
Mark B.N. Hansen. Chicago: U Chicago P, 2010. 199-214. Print.
Metzinger, Thomas. The Ego Tunnel: The Science of the Mind and the Myth of the Self. 2009.
New York: Basic Books, 2010. Print.
Wilson, Edward O. Consilience: The Unity of Knowledge. New York: Vintage Books, 1999.