Posthumanism and Instrumental Eliminativism
Abstract:
I distinguish transhumanism from posthumanism in both speculative and critical forms. I interpret posthumanism as a thesis regarding the possibility of the emergence of non-human life as a result of human technical and scientific activity. I take the term ‘posthuman’ to refer to hypothetical descendants of current humans who are no longer human in consequence of some history of technical augmentation and argue that this might occur via technicallyinduced changes in the structure of human cognition. In the remainder of this article I sketch one such scenario for posthuman technogenesis. It assumes what I call the ‘linguistic constitutivity thesis’: that language is a cognitive tool that is necessary for the possession of structured
propositional
attitudes
and
their
associated
concepts.
Given
linguistic
constitutivity, I claim that an augmentation which obviated the need for public language by replacing it with a more powerful non-symbolic medium would ipso facto remove the preconditions for propositional and conceptual thought. Lacking these preconditions, human minds would cease to exist. They would be replaced by posthuman minds with characteristic repertoire of non-propositional attitudes exploiting non-linguaformal media for mental representation.
Keywords:
Transhumanism, posthumanism, propositional attitudes, the extended mind, eliminativism, continuous computation.
1
1) Posthumanism, Transhumanism, Eliminativism
Contemporary transhumanists argue that human nature is an unsatisfactory 'work in progress' that should be technically modified where the instrumental benefits for individuals outweigh the technological risks. This technoprogressive ethic is premised on prospective developments in Nanotechnology, Biotechnology, Information Technology and Cognitive Science – the so-called 'NBIC' suite. Pharmacological agents like amphetamines, modafinil and dopamine agonists are already used to increase the efficiency of learning and working memory (Bostrom and Sandberg 2006). Existing technologies such as genetic engineering and transcranial magnetic stimulation are also liable to be used in increasingly targeted ways to increase the efficiency of cognitive processes. More speculatively, micro-electric neuroprostheses’ might eventually be used to interface the brain directly with non-biological cognitive or robotic systems (Kurzweil 2005, 317) 1. Such developments might bring forward the day in which all humans will be more intellectually capable due to enhancements in their native biological machinery or through seamless interfacing with supplemental cognitive technologies.
Just how unrestricted and capable transhuman minds and bodies can become is hotly contested since the scope for enhancement depends both on hypothetical technologies which may turn out to be impossible or unrealistically dangerous – e.g. artificial general intelligence or wide ranging nanotech manufacturing – and upon unresolved metaphysical issues concerning the nature of mind, embodiment and the physical world.
1
For a rather less sanguine commentary on the state of the art in non-invasive scanning see Jones 2009. 2
The conceivability of a advanced post-biological intelligence – whether implemented entirely on non-biological platforms or by a hybrid of biological and artificial system - indicates to some that a convergence of NBIC technologies will not only enhance intelligence within the normal human range but, beyond a critical point, contribute to a discontinuously rapid change in the level of mentation on this planet. Virnor Vinge refers to this point as 'the technological singularity' (Vinge 1993; Bostrom 2005, 8). Vinge, along with Ray Kurzweil and Hans Moravec, argues that were a single super-intelligent machine created it could create still more intelligent machines, resulting in a recursively generated growth in cognitive capacity to levels that (lacking this capacity) we cannot even imagine. We could no more expect to understand a post-singularity entity than a rat – lacking the capacity for refined propositional attitudes - could be expected to understand human conceptions like justice, number theory or public transportation.
The idea of a post-singularity intelligence provides a limit case of the posthuman, as opposed to the transhuman. Whereas transhumans are humans that (by current standards) are exceptionally gifted in virtue of their technical augmentations, a posthuman is a being whose augmentation history renders it inhuman. Whereas some humans already exhibit cognitive accomplishments of a kind that transhuman enhancements might open up to many – e.g. memorizing very long lists of random data - the distinguishing accomplishments of the posthuman would have no place in the current physical or cognitive horizon of humanity. The rat-human analogy helps us understand the extent to which posthuman cognition might transcend our intellectual capacities – though not a means of accessing posthuman thoughts or experiences.
3
The idea of the singularity suggests that posthumanism should be viewed as a speculative metaphysical position distinct from the normative thesis of transhumanism. One can be a transhumanist while denying the ontological possibility of posthuman transcendence. Similarly, speculative posthumanism is consistent with the rejection of transhumanism. One could hold that a posthuman divergence is a significant possibility but not a desirable one.
2
I prefer to call this position ‘speculative posthumanism’ in contradistinction to the ‘critical posthumanism’ currently in vogue in the humanities. Critical posthumanists such as Katherine Hayles and Rosi Braidotti claim that current technoscientific change ‘deconstructs’ the philosophical centrality of the human subject in epistemology, ethics and politics (Hayles 1999). However, if it is logically or metaphysically possible for our augmented descendants to transcend our cognitive or phenomenological horizons, there must be some set of capacities distinctive of human beings – a ‘human nature’ if you will - to transcend - which the mere fact of our changing technical and scientific horizons leaves pretty much in place.
The idea of the posthuman to which speculative posthumanists can be summarized as follows: Posthumans are hypothetical ‘descendants’ of current humans that are no longer
human in consequence of some augmentation history . This schema captures the idea that posthumans would not be like superior aliens, gods or angels. They would have ‘wide human lineage’ in that their emergence would be a result of formerly human technological or cultural activity.
Given the acknowledged difficulty of predicting the qualitative nature of future technological change over relatively short time-spans, any prognostic accounts of the emergence of 2
Although some hold that the singularity is ‘beyond good or evil’, one might hold that certain posthumans would be worse off than even the most miserable human; a possibility that could warrant anti-transhumanist policies such as technological relinquishment or pre-emptive species suicide. 4
posthumans need to be approached with incredible caution. However, if speculative posthumanism qua metaphysics is even plausible, it makes theoretical and political sense to explore whatever areas of the possibility space for the emergence of posthumans are within our ken so that we can better understand the implications of our current technoscientific activities. In this paper I wish to consider a speculative scenario for the emergence of the posthuman. This hinges on the claim that our cognitive nature is accurately described by the claim that typical humans have structured propositional attitudes (beliefs, desires, wishes, hopes, thoughts, etc.) and that our capacity for the attitudes depends upon the linguistic acculturation made possible by species typical capacities for language acquisition. The idea, then, is that language is a cognitive tool and that this tool equips us with distinctively human minds. If this is right, then were our technologically advanced descendants to produce cognitive tools that resulted in the natural language becoming obsolete or vestigial, their minds would cease to be human and would become posthuman in the sense outlined.
The linguistic constitutivity thesis (as I shall refer to it) makes possible the instrumental
elimination of propositional attitudes or at least linguaformal conceptual thinking. Instrumental eliminativism, of course, must be sharply contrasted to the theoretical eliminativism championed by Paul Churchland and others. It presupposes that folk psychology is actually a good and true theory of our minds as they are currently constituted but assumes that this constitution supervenes on our using culturally available tools that might cease to exist at some point in the future.
2) Linguistic Constitutivity
5
Following assorted 'linguistic turns', philosophers and social scientists from Marx to Heidegger, Habermas, Dennett and Davidson have claimed that capacities we take to distinguish humans from non-human animals are predicated our command of natural languages. For example, it has been claimed that the ability to use language is essential for thinking with abstract concepts, the possession of propositional attitudes, developing the 'folk psychological competence' to attribute propositional attitudes to others, serially ordered consciousness, the understanding of truth, objectivity or Bei ng.
Any version of the 'linguistic constitutivity thesis' entails that a global loss of language capacities would efface the transmissible cultural forms that equip our minds to be 3
distinctively human. A conflagration that caused humans to lose the capacity to speak, parse or interpret the speech of others, or to engage in verbal thought, would result in a population whose mental life was no longer recognizably human. These would not be posthumans according to the criterion advanced in the last section since no augmentation history would have been involved. However an augmentation history that eventuated in a post-linguistic intelligence for which language either plays no cognitive role or, at best, a vestigial one, would result in the technogenesis of a posthuman life form.
To put bones on this technogenesis scenario, it would help to sketch out a philosophical framework for linguistic constitutivity. I have chosen to articulate it from within the Extended Mind Thesis (EMT) - a position that combines wide acceptance by a range of pragmaticallyoriented cognitive scientists, psychologists and philosophers with a peculiar ease of 3
Some object that language cannot be synchronically constitutive of human thought since many aphasics can perform a range of high-level cognitive tasks such as passing false-belief tests. Much depends on the details and strengths and claims made in any particular version of the linguistic-constitutivity thesis. For example, we could defend a central diachronic role for language in the development and sculpting of cognition and brain structure without subscribing to the synchronic claim that language actually vehiculates high level thought. In what follows, I shall set out a position within which language has both synchronic and (more generally accepted) diachronic roles, though I acknowledge the problems with this view in passing (Carruthers 2002).
6
integration within the technogenesis scenario. EMT holds that that mental process don’t just happen in brains, but in the environments of embodied thinkers . Proponents of the extended mind thesis like Andy Clark and David Chalmers argue from a principle of 'parity' between processes that go on in the head and any functionally equivalent process in the world 4
beyond. The parity principle implies that mental processes need not occur only in biological nervous systems but in the environments and tools of embodied thinkers. If I have to make marks on paper to keep in mind the steps of a lengthy logical proof, the PP states that my mental activity is constituted by these inscriptional events, as by the knowledge and habits reposing in my acculturated neural networks.
EMT is more congenial to the technogenesis of new human (and prospectively posthuman) forms of cognition because it allows that I can augment my cognitive structure by altering affordances in my extra-bodily environment to produce new cognitive resources (e.g. new type of representation).
According to the EMT-perspective, public languages have structural properties which furnish a shared 'workspace' for thinking, allowing symbol users to shrug off their dependence on the idiomatic features of different neural connectivities. Non-linguistic representation play a part in this model: e.g. implementing symbol-using capacities or furnishing a raft of non-symbolic recognitional powers (See section 3). Nonetheless, we ‘Natural born cyborgs’ – to cite Clark’s book of the same name – are dealers in hybrid mental representations which exploit both a linguistically mapped environment and our multifariously talented brains (Clark 2003).
4
'
Parity Principle. If, as we confront some task, a part of the world functions as a process which, were it to go on in the head, we would have no hesitation in accepting as part of the cognitive process, then that part of the world is (for that time) part of the cognitive process'.(from Clark and Chalmers (1998) p.XX)
7
In order to better appreciate how language might furnish extended cognition and the implications of obviating it though cognitive enhancement, we need to specify in sufficiently broad terms the key features of language which might plausibly warrant attributing a humanconstitutive role to it:
1) Linguistic behaviour consists in the production of symbols. Symbols are articulated (or finitely differentiated). If it is not possible to determine which symbol type a given mark belongs to, it can’t be a symbol.
2) Just as importantly, a given symbol can be employed by different users in different times and places for different purposes. Some philosophers analyze this in terms of a type-token ontology, where tokens are particular instances of symbols and types are the syntactic or semantic essence instantiated on each use. This analysis is not uncontroversial (MacClelland 2002). However, it is worth emphasizing that the 'ideality' of symbols – their repeatability in different contents - is essential both to the function conventional linguistic signs or the hypothetical 'symbols in the head' postulated by proponents of classical computationalist accounts of mental processing.
3) Linguistic symbols also exhibit compositional structure. Complex symbols like sentences or predicate expressions are composed according to the grammatical rules of the language. Rules governing logical connectives or embedded clauses, for example, allow symbols of arbitrary complexity to be composed recursively by reapplying them to their own output.
Compositional structure with recursion intuitively furnishes a huge representational gain over sign systems with minimal constituency – particularly those lacking recursion. Not only does
8
it allow the rule governed generation of arbitrarily complex grammatical strings, it allows inference rules to be defined in terms of the syntactic structures of those strings.
Classical computationalists like Jerry Fodor propose that comprehending public language involves translation into a language of thought (‘mentalese’). Thinking is the manipulation of mentalese symbols according to structure sensitive algorithms which reflect the brain’s physical dispositions analogously to the way in which software loaded in a digital computer reflects its physical state at compile time.5 For CC, mentalese rather than public language is the medium of thought, whereas public or natural languages are media for their communication (Fodor 1987, 1990).
However some argue that public language is, if not a language of thought per se , a prerequisite of richly articulated propositional thinking; not merely a public medium for its expression. A psychological plausible – if contentious - way of establishing this position is to tie the capacity for propositional or conceptual thinking to the capacity for representing propositional attitudes like beliefs or desires, then to identify language as the medium of meta-representation. Propositional attitude psychology and the full-dress conceptual thinking that goes with can then be attributed to the 'linguistic hybridity' of human cognition.
Donald Davidson's work in semantics provides a powerful schema for this strong version of the linguistic constitutivity thesis. Davidson argues that the ability to have beliefs requires a grasp of what belief is and the relationship of belief to other propositional attitudes (e.g. thoughts, intentions and desires). Understanding belief requires that one 'understands the 5
Where these rules are rational or truth-preserving so are the thoughts they implement. CC thus reconciles physicalism with the claim that psychological explanations rationalise or justify. It explains the apparent 'systematicity' of conceptual thought by identifying concepts with predicate expressions in the mentalese lexicon. If one has a concept in one’s mental lexicon, then it can enter into any grammatical combination with other parts of one's lexicon. If I can think that John loves Mary and Mary killed Ned, I can think that Ned loves John or Ned killed Mary. 9
possibility of being mistaken' and a recognition that others might have true or false beliefs about its topic. Belief is an attitude of 'holding true' that which is believed and cannot be intelligibly adopted without a grasp of what this implies for the evaluation of one's own others' mental states (Ibid., p. 170).
Suppose that we know Fred to have an obsession with tiger hunting and are aware that a large dog is prowling the vicinity. Seeing Fred stalking with a sense of nervous anticipation around his garden while holding a rifle might lead us to attribute to his anomalous behaviour to the (false) belief that a tiger is in the area, plus the (true) belief that there is a predator in the area. Distinctions such as between the belief that tiger is around the the belief that predator is around require a grasp of the truth conditions and inferential consequences of said beliefs. For Davidson, only a semantic understanding that relates arbitrary sentences to their truth conditions on the basis of some interpretation of their sub-sentential components (which could reoccur in other sentences about tigers, dogs, hunting or predators, or snow) can furnish this since only speech behaviour provides the intensional distinctions relevant to the explanatory description of action:
Our manner of attributing attitudes ensures that all the expressive power of language can be used to make such distinctions. One can believe that Scott is not the author of
Waverley while not doubting that Scott is Scott; one can want to be the discoverer of a creature with a heart without wanting to be the discoverer of a creature with a kidney. One can intend to bite into the apple in the hand without intending to bite into the only apple with a worm in it; and so forth. The intensionality we make so much of in the attribution of thoughts is very hard to make much of when speech is not present. The dog, we say, knows that its master is home. But does it know that Mr Smith (who is
10
his master), or that the president of the bank (who is that same master), is home? We have no real idea how to settle, or make sense of, these questions (Ibid. 163).
Compositionality and the repeatability of linguistic symbols also play a role in the familiar linguistic device of embedding sentences as grammatical complements of sentences attributing beliefs, desires or indirect speech. Davidson calls this repeatability 'the autonomy of meaning'. Linguistic symbols are 'autonomous' with regard to their usage insofar as they retain an identity across divergent uses. Thus while the sentence 'There is a tiger in the woods' can be used in a blunt assertion, it can also be embedded in an utterance that attributes the belief that there is a tiger in the woods to Fred or used sarcastically or ironically (Ibid, pp. 6
164-165). So the autonomy (or ideal repeatability) of linguistic symbols provides the metarepresentational device by which we evaluate our own or others' utterances and attitudes in the course of interpreting them. Given the assumed tie between evaluating and having attitudes, then, we could not qualify as true believes without the reflexivity afforded by complex linguistic structures.
One might object that even granting the constitutive tie between the capacity for evaluating attitudes and for having them, it is prima facie as plausible to suppose that these are metarepresented in mentalese or some other neural medium of representation (Bermudez 2003, p. 158). This issue is, in part, an empirical one and has received considerable attention from psychologists concerned with the role of language in children's development of folk psychological competences such as the ability to infer behaviour from false-belief attributions
6
Davidson's paratactic analysis of direct and indirect discourse actually denies that s in attributive sentences like 'Fred believes that s' is a semantic constituent. Davidson prefers to treating 'that' as an indexical device. However, this is still consistent with compositionality so the details of his theory of direct and indirect speech need not concern us. 11
7
(Pyers and Senghas 2009) . However, linguistic hybridity arguably confers benefits on thinkers that a purely 'inner' medium such as mentalese or non-linguaformal 'neuralese' cannot afford. In saying or writing some text we turn it into a relatively persistent articulate structure which we return to for further evaluation and reflection (Clark 1996, 177). Jose Bermudez argues that this renders structured thoughts accessible for conscious or personal recapitulation and reflection, a feature which is arguably absent in the case of mentalese or Churchland-style higher-dimensional brainstates, or any other hypothesized inner medium (Bermudez 2003). If so, then language appears to be in a unique position to supply what Clark refers to as 'second order cognitive dynamics' (Clark 1996, 177): a suit of selfmonitoring capacities which, as he avers, furnish the 'distinctly human capacity' to think about our thoughts.
This opens up two routes to linguistic constitutivity. Firstly, as emphasized by Bermudez, it allows us to represent complex concepts via the articulate structures of public language. If Bermudez is right, then Clark is surely on the right track in proposing that we treat representations of natural language strings (whether neural or textual) as proper parts of hybrid mental representations, for these representations derive their capacity for refined logical articulation from a juxtaposition public-discursive representational resources with native discriminative powers (Clark 2006). This implies that a considerable portion of human 8
conceptual-space is essentially of a hybrid nature. Secondly, if these self-monitoring tools are essential for the evaluation of propositional attitudes and (following Davidson) evaluating
attitudes is a precondition for their possession , then language use must also be a precondition 7
Rules governing sentential embedded clauses allow language users to form sentences about linguistic or propositional objects such as 'Joan believed that Bill is the culprit' or 'Nick said “Bill is falsely accused”' (). These rules can be applied recursively (as in 'Nick hoped that Joan's belief that Bill is the culprit was mistaken'). 8 Working within a modular approach to cognitive architecture, Peter Carruthers has argued that competence in natural language furnishes the means to integrate the output of domain-specific modules (e.g. those for tracking color or geometric relationships) using abstract representations with no appurtenance to a cognitive or sensory modality (Carruthers 2002). 12
for a distinctively propositional attitude psychology. The second, Davidsonian, position is arguably the stronger since one might envisage beings possessing a similar propositional attitude psychology to humans but which – due to the dearth of language – are incapable of accessing the space of human culture opened up by the external representation of conceptual and discursive relationships.
3) The Technogenesis of the Posthuman
Both strong and weak forms of linguistic constitutivity make significant areas of human cognition dependent on the existence of public language. It should be emphasized that even the stronger version does not entail that the attitudes are a mere abstracta posited for the sake of prediction or interpretation. Nor, as emphasized in Section 1, does it entail that they posited in a moribund folk theory. However, it does opens up a prospect of instrumental eliminativism. If linguistic hybridity is constitutive of propositional thought (strong constitutivity) or refined conceptual thought (weak constitutivity) we don’t have to theoretically defeat folk psychology to eliminate the mental. We can eliminate the mental by eliminating (the need for) language.
Could this occur as a consequence of some or other augmentation technology?
Well, for any such technology to count as an augmentation it would have to render language vestigial while providing at least equivalent scope for sophisticated cognition. For example, we would expect our hypothetical posthumans to be capable of non-symbolic analogs of the 'second order cognitive dynamics' that Clark attributes to linguistically-mediated cognition: e.g. sophisticated conceptual thinking, reflection and interpretation.
13
In his seminal ‘Eliminativism Materialism and the Propositional Attitudes’, Paul Churchland discusses an inter-cranial commissure thought on analogy with the thick trunk of nerve fibers which links our right and left cerebral hemispheres. Churchland supposes that a high bandwidth microwave connection between brains might allow two or more people to ‘coordinate their behaviour with the same intimacy and virtuosity displayed by your own cerebral hemispheres’. Churchland speculates that were such a commissure to become widely available: ‘language of any kind might well disappear completely, a victim of the “why crawl when you can fly?” principle. Libraries become filled not with books, but with long recordings of exemplary bouts of neural activity’ (Churchland 1981).
However, while this vision of microwaved bliss was a nice rhetorical flourish; there are technical problems integrating it into the vector-coding model that Paul and Patricia Churchland later adopted. If the vectors corresponding to activation states in neural networks are the primary determinants of content identity or similarity, then two networks can only be in states with the same content if their state spaces are at least mathematically equivalent (Churchland 1998; Garzon 2000). If they have different numbers of neurons, for example, any total state will have a vector of different length and thus a different content. As Jerry Fodor and Ernest Lepore argue, given reasonable assumptions about human neural idiosyncrasy, this proposal has the counter-intuitive consequence that no two individuals could share the same concepts (Fodor and Lepore 1992; Garzon). Churchland later argued that the dimensionality problem could be obviated by a similarity measure which the content of activation states/prototypes in terms of a) correlations between their Euclidean distances from other states within the hidden layer of the network and b) the causal relationships
14
between hidden layer activation states and the sensory ‘input’ layer (Churchland 1998). This would still present a problem for conceptualizing direct neural communication, however.
Firstly, sufficiently richly connected brains would intuitively constitute a single network. Even allowing for a detailed knowledge of the brains and their interactions, what principles preclude treating two people linked by the commissure as a single coupled system with some representations distributed between crania?
More pertinently, if linguistic constitutivity is right, coupled brains might also lack the capacity for second order cognitive dynamics identified by Clark. Their bouts of neural activity would lack the embedding in a transcultural system of implication and meaning required for refined conceptual thought.
Still, it seems to me that we can propose a cognitive prosthesis in the context of the EMT that would have a similarly eviscerating effect on libraries. Rather than describing how brains might talk to one another, we can envisage a cognitive augmentation which could be conceivably adopted by humans or augmented transhumans which adds sufficient functionality to be considered an augmentation but which inclines its users to employ nonsymbolic vehicles for thinking, communicating, interpreting, etc. in preference to symbolic vehicles such as linguistic entities.
To ensure that these are non-symbolic we can specify initially that they should have a nonsymbolic syntax. One of the syntactic requirements of a symbol system such as a language is that it is possible in principle to determine unambiguously which disjoint symbol a given token belongs to (finite differentiation). Contrastingly, we can specify that our proposed
15
augmentation should use vehicles which do not have to belong to definite representational types and which are subject to a far wider class of computational processes than symbols.
The dropping of the articulation requirement is the syntactical basis for Brian MacLennan's distinction between a Discrete Formal System (DFS) and a Continuous Formal System (CFS) or 'simulacrum' and this provides a model for understanding what a reflective system of non propositional thinking might be like. MacLennan calls the representational vehicles of CFS's 'images'. Even where some kind of representational typology exists for a CFS, an image's membership of a correlative type is a matter of degree and subject to continuous variation (MacLennan 1995, p. 2). It follows that there may be cases where a given image belongs to one or more types to varying degrees. Similarly, whereas interpretability is a binary matter in an DFS's (only grammatically well-formed strings can be interpreted) the interpretation function for an image in a simulacrum can map onto superpositions of continuously varying semantic values within a vector space defined by the semantic dimensions along which the image can be evaluated. An image can be a slightly more true than false, but also a 'bit' undefined compared with a truer or less indefinite image (Ibid. p.6). Similarly, two images might refer to an object to greater or lesser degrees.
A material instantiation of a CFS would be syntactically dense – more like pictorial than symbolic representation - since any change in the morphology of an image could be accompanied by a change in meaning. Similarly, computation in simulacra could involve the continuous evolution of images rather than the discontinuous transitions between states involved in classical or discrete conceptions of computation. Importantly, MacLennan shows how the idea of a program can be generalized so that they can be represented by further images. For example, the method of gradient descent (employed in variants of the Delta Rule
16
in the theory of Neural Networks) can be conceptualized as the use of an error surface (the 'program' image) to define the rate and direction of change in some continuous variable corresponding to the computed image (Ibid, 9-10). A similar principle can be applied to a generalized grammar where well-formedness is a matter of degree and maximally wellformed images correspond to some sink region on a guiding surface (Ibid, 11-13).
MacLennan's theory of simulacra allows us to envisage a representational format which is a) non-symbolic and b) has computational resources unavailable to symbolic systems and c) is capable of representing its own computational procedures and grammatical structures in terms of its own imagistic resources. Moreover, since the semantic values of DFS's can be treated as points along the continua calculated by the interpretation functions and grammars of simulacra, it is possible for an computational system described by some CFS to approximate any system describable by an DFS to an arbitrary degree of accuracy (Ibid. 14).
Let us suppose, then, that a radical cognitive enhancement becomes available that implements or closely approximates styles of continuous computation associated with simulacra, which includes not only imagistic representations of non-images (entities in the world outside the system) but images of programs and images which 'parse' other images by representing their grammatical derivation. Presumably, the this non-symbolic workspace (NSW) would have to be represented for its users in some form or other - perhaps through some immersive virtual reality technology in which images of varying orders could be manipulated and shared (we need not concern ourselves with precisely how it does this). Is there any reason to expect that the NSW could instrumentally eliminate typically human forms of propositional and conceptual thinking?
17
For this to occur, the kind of thinking which typifies human propositional/conceptual thinking: a) must be or must closely approximate computational processes where state transitions mediate between cognitive states that are constitutively symbolic; b) must be eliminable through the use of the NSW.
Condition b) might be satisfied if this discrete style of cognitive dynamics is linguistically constituted. For if the NSW were to replace the dominant symbolic medium of thought and communication with a non-symbolic medium, current humans or their descendants might cease to use language altogether or employ it as a vestigial instrument (e.g. for speaking to unaugmented humans, if any are still around). If language constitutes distinctive human cognition in this way, our augmented successors would cease to have distinctively human cognitive states and would (given linguistic constitutivity) have ceased to be human. Since
this becoming-non-human would have derived from a technological augmentation, these hypothetical non-symbolic cognizers would satisfy the requirement for posthumanity introduced above .
A number of objections can be leveled against this scenario for the technogenesis of posthumans:
1)
Why suppose that an NSW could be so powerful and comprehensive a thinking tool
as to prompt a transition from linguaformal to non-linguaformal modes of cognition? 2)
Why assume the linguistic constitutivity thesis?
3)
Even assuming linguistic constitutivity, do we know that human cognition is not
already better thought of as implementing a CFS than a DFS? (How do we know that condition a) applies?).
18
Assumptions about the power and generality of an NSW are supported by the mathematical claim that DFS's can be approximated by CFS's. If this is right, then for any discrete computer there ought to be a possible continuous computer that is at least as capable . This does not suffice to show that our posthuman descendants will have hybrid thoughts or attitudes mediated by a NSW that are strictly unavailable for humans. However, given that a NSW would be at least as flexible as any current system, it seems reasonable to allow that they might.
The linguistic constitutivity thesis I simply take as a working assumption for developing the technogenesis scenario. I have argued that it is independently plausible. However, variants on this scenario are possible which do not assume it. Supposing that our thinking is performed in mentalese and that this is an innate and not culturally acquired feature of our neural organization. It remains possible that our dependence on mentalese could be obviated through the use of an NSW. One could postulate an imaginary technology that reformats the brains of its users so as to render them capable of using only its proprietary non-symbolic scheme for mental representation. The capacity for propositionally structured thought would ‘be overwritten’. Such a technology hard to envisage in the absence of a convincing story about how nervous systems are shaped to think linguaformally. It would arguably presuppose the near-posthuman capacity to turn brains of one kind into brains of a completely different kind by directly manipulating synaptic connections.
It is even harder to see why anyone would wish to use it. If you were told that after fixing some gadget to your head, all your beliefs, desires would be erased and you would lose the capacity to understand fellow humans, you might be unimpressed by the hazy prospect of
19
being able to engage in some form of non-symbolic cognition beyond the ken of flat-footed symbol users. You might be forgiven for regarding its effects as a form of wholesale cognitive mutilation rather than enhancement.
Alternatively, one might envisage the NSW as added connectivity, providing a non propositional option for sophisticated mental representation. This product would offer people a choice of coding schemes for complex thinking where none currently exist. As in the first option, though, any shift from propositional to non-propositional coding would require altering the structure of human brains in ways that no culture or technology has done to date. For example, assuming some kind of modular architect ure, every mentalese output or input to a module would need to have a non-symbolic duplicate. By contrast, assuming linguistic constitutivity has the advantage of allowing us to visualize the technogenesis of posthumans as a cultural process involving no substantial alterations in human biology beyond that required for the use of the NSW.
Objection 3 is perhaps the most serious. Many cognitive scientists and philosophers deny that human cognition is most usefully conceptualized in terms of structure-sensitive algorithms applied compositionally structured representations. One of MacLennan's motivations for formulating the notion of a CFS is that the rules and structured representations approach does not seem likely to capture the context-sensitivity of much human and animal thinking, which is liable to resist any summation within a finite rule set (See MacLennan 2002, pp. 7-8). As is well known, other philosophers and cognitive scientists have argued that the classical computationalist model of structure sensitive algorithms applied to syntactically structured representations cannot explain phenomena such as the holistic nature of belief-fixation, the temporal structure of experience or our capacity for relevance-addressable memory. In its
20
place and have proposed a range of more or less breaks with the classical computationalism such as vector-coding models or entirely representation-free dynamical models. If any of these non-classical alternatives are right – or significantly near right – then, the objection goes, it is likely that humans are already implementations of CFS's. Thus the implementation of a CFS via a non-symbolic shared workspace would simply be augmenting a cognitive architecture that is already non-algorithmic in terms of the state transitions it implements or non-symbolic in its representational vehicles, or both (See for example, Horgan and Tienson 1994 and 1999).
The hybrid conception of representation developed in the EMT model already acknowledges that some recognitional and inferential capacities supporting the intelligent use of public symbols are not best modeled as operations on syntactically structured strings. Jeffrey Elman's work on connectionist language-learning suggests that grammatical distinctions between sentences with different degrees of grammatical embedding could be represented in terms of distances between hyperplanes within the activation space of a recurrent neural network rather than, say, the syntactic structure of linguaformal internal representations (Elman 1995). However, the instrumental elimination scenario only assumes that human conceptual thinking or propositional attitude psychology has an essential symbolic component, not that it is symbolic 'all the way down'.
Clearly, if the Churchlands’ theoretical eliminativism or radical anti-representationalists like Van Gelder are right and symbol-use is largely extrinsic to the content or dynamics of cognition, then the non-symbolic cognitive augmentation envisaged would be ‘more of the same’ and no discontinuity between human and posthuman cognition could be inferred. However, if language usage is constitutive of our psychological nature in the manner outlined
21
in Section 2, then a non-symbolic cognitive augmentation might conceivably prompt a transition from a human to a posthuman psychology.
Bibliography
Bermudez, Jose L (2003), Thinking without Words , Oxford: OUP. Bostrom, Nick (2005), ‘A History of Transhumanist Thought’, Journal of Evolution and
Technology , 14 (1). Bostrom N, Sandberg A (2006), 'Converging Cognitive Enhancements', Ann. N.Y. Acad. Sci. 1093: 201–227. Carruthers, Peter (2002), 'The cognitive functions of language', Behavioral and Brain
Sciences 25, 657-726. Clark Andy, Chalmers David (1998), ‘The Extended Mind’, Analysis 58(1), pp. 7-19. Clark, Andy (2003), Natural Born Cyborgs , (Oxford OUP). ___’Language, Embodiment and the Cognitive Niche’, Trends in Cognitive Science 10(8), pp. 370-374) ___(1993) Associative Engines, (MIT Bradford). ___(2006) ‘Material Symbols’, Philosophical Psychology Vol. 19, No. 3, June 2006, pp. 291–307 Churchland, Paul (1998), ‘Conceptual Similarity Across Sensory and Neural Diversity: The Fodor/LePore Challenge Answered’, Journal of Philosophy , XCV, No.1, pp. 5-32. _____(1995). The Engine of Reason, The Seat of the Soul (Cambridge Mass.: MIT Press.
22
_____(1989)‘Folk Psychology and the Explanation of Human Behaviour,’ Philosophical
Perspectives 3, pp. 225–241. ____(1981), ‘Eliminative Materialism and the Propositional Attitudes’, Journal of
Philosophy 78(2), 67-90. Davidson, Donald (1984) ‘Thought and Talk’, in Inquiries into Truth and Interpretation (Oxford, Clarendon Press), pp. 155-170. Jeffery Elman (1995), ‘Language as a Dynamical System,’ in R Port and T Van Gelder (eds),
Mind as Motion: Explorations in the Dynamics of Cognition , (MIT: Cambridge Mass.), pp. 208–222. Fodor, Jerry (1987) Psychosemantics (Cambridge Mass.: MIT 1987). ____(1990)Theory of Content and Other Essays , (Cambridge Mass.: MIT 1990). Fodor, Jerry and Lepore, Ernst (1992), Holism: A Shoppers Guide (Oxford, Blackwell). Garzon, Francisco C (2000), ‘State space semantics and conceptual similarity: reply to Churchland’, Philosophical Psychology 13(1), pp. 77-95. Hayles, N Katharine (1999), How We Became Posthuman: Virtual bodies in Cybernetics,
Literature and Informatics (Chicago: University of Chicago Press). Horgan, T, Tienson J (1994), 'A Nonclassical Framework for Cognitive Science' , Synthese, 101:3, p.305-345. ____(1996, Connectionism and the Philosophy of Psychology , (Cambridge, MA: MIT Press). Jones,
Richard
(2009),
‘Brain
Interfacing
with
Kurzweil’,
http://www.softmachines.org/wordpress/?p=450 , Accessed 08.09.2009.
Kurzweil, Ray (2005), The Singularity is Near (New York Viking). MacLennan, B.J. (1995), ‘Continuous Formal Systems: A Unifying Model in Language and Cognition’ in Proceedings of the IEEE Workshop on Architectures for Semiotic Modeling
and
23
Situation
Analysis
in
Large
Complex
Systems ,
Monterey,
CA,
pp.
161–172,
ftp://cs.utk.edu/pub.maclennan/CFS-UMLC.pdf . Cited 20 December 2009.
____ (2002), Transcending Turing Computability, Minds and Machines 13: 3–22. O’Brien G, Opie J (2002), ‘Radical connectionism: thinking with (not in) language’,
Language & Communication , 313-329. Pyers J and Senghas A (2009), 'Language Promotes False-Belief Understanding Evidence From Learners of a New Sign Language', Psychological Science, 20(7), pp. 805-812. Vinge, Vernor (1993), ‘The Coming Technological Singularity: How to Survive in the PostHuman
Era’,
http://www.rohan.sdsu.edu/faculty/vinge/misc/singularity.html .
Accessed
2008.24.04.
24
