Which Morality: Moral Cognition in Clark and Churchland
In the following essay I explore a discussion between Andy Clark and Paul Churchland published in the Canadian Journal of Philosophy. This discussion centres around the concept of moral cognition as it relates to a connectionist theory of mind. Clark’s and Churchland’s positions share a lot in common but their seemingly minor differences result in vastly different conclusions. In comparing Clark’s and Churchland’s positions I argue that the important question to be asked is not who’s position is correct, but how either of the two positions relate to moral theory.
Both Clark and Churchland argue for their different positions on moral cognition from a theory of mind known as connectionism. I’ll discuss where Clark’s and Churchland’s positions differ below but to understand the import of their positions, and the nuances of the differences, a basic understanding of connectionism, and how moral cognition relates to connectionism, is required.
Consider the following diagram1.
Each of the circles in the diagram above are called nodes and each line between these nodes is given the rather innovative title of connection. Connectionism gets its name from the connections in the connectionist network (Clark 2001, p. 62). Another important fact about the above diagram is that, as the alternate title suggests, it attempts to empirically model the brain; each node is a neuron and the connections between the nodes are synapses (Clark 2001, p. 62). The way that the network functions is that an input is delivered into the input node, this progresses through to the hidden node, which passes it to the output node2. Were this to be described using the biological structure of the brain we would say that an input is delivered to a sensory neuron, which passes it through other neurons before reaching the motor neurons (Garson 2007).
Not just any input is passed along to other nodes in the network however. Each node on the input layer has a weighted value which is passed along the connection. The next node in line determines its weighted value based upon the inputs it receives from the nodes prior to it in the network. If this weight is of an appropriate value it will pass the information along to the next node in line in continuum (Clark 2001, pp. 64 – 66). All information in a connectionist system, from the cold chill on a winter morning that causes the goose flesh to rise on your forearms, to your recollection of a childhood memory, is thus stored in these connection values or vector algorithms.
The way moral cognition relates to a connectionist theory of mind is primarily concerned with two things. One, how the moral information gets in to the connectionist system – that is how moral information enters the input nodes and becomes ’stored’ in connectionist vector algorithms – and two, how the moral information gets out of the connectionist system -that is how the moral information encoded in the connectionist vector algorithms gets processed by the output nodes -. To be clear, moral cognition is not about how one behaves, acts or thinks; it is not about what one does at all. Moral cognition is solely concerned with how we cognise, process, or incorporate morality into our very being.
In Word and Action: Reconciling Rules and Know-How in Moral Cognition (2000a, pp. 1 – 33) Clark distinguishes between two areas in which moral cognition is said to occur, these are exemplar-based learning and non-exemplar-based learning. Exemplar-based means to learn via experience whereas non-exemplar-based means to learn via rules. Clark’s main thesis in this paper is to persuade the reader that both methods of incorporating morality into a connectionist system are important and should not be overlooked. Clark observes that exemplar-based learning, at least in relation to moral cognition, has been favoured in the past within connectionist theories of mind, particularly the theories of Churchland (Clark 2000a, p. 3). As mentioned above, connectionism has all of our mental information, including morality, stored in vector algorithms. The bias that Clark highlights is founded on the assumption that a set of linguistically derived and articulated moral rules or maxims can never do justice to the deep content in a connectionist network; moral cognition based upon rules will never equal the depth of moral cognition based upon learning (Clark 2000a, pp. 4 -5).
Clark’s position doesn’t deny the depth of the learning and storage capacity present in a connectionist system but he doubts that morality can be conducted in an individual mental world. For Clark, moral reasoning, and hence moral cognition, is a ‘communal and collaborative affair’ where moral decisions are made in groups, perhaps via debate, and this process is largely mediated via language (Clark 2000a, pp. 9 – 10). This is the primary area of difference between Clark and Churchland for Clark (2000a, p. 10) asserts that language is not just a tool that we utilise in resolving moral dilemmas but that it is through language that the ‘moral realm “comes into view” as an object of human cognitive endeavour’. Language for Churchland, as I mentioned above, is an insufficient tool to articulate the contents of the connectionist mind; language for Clark, however, is the very substance of the moral world.
I’ve focused quite heavily upon Clark’s original paper but in the same edition of the journal where it is to be found there is a reply from Churchland titled Rules, Know-How, and the Future of Moral Cognition. In Churchland’s introduction he states that, at least on one level, he agrees with Clark; he agrees that the moral world is indeed social and hence should not be limited to the space occupied by the human brain. The human brain is important in this respect as Churchland argues that other animals, that is animals other than humans, exhibit moral behaviour, another point of difference between himself and Clark (Churchland 2000, pp. 297 – 298). Churchland doesn’t agree with Clark’s main thesis however. Language may certainly play a part in the human moral world but non-exemplar-based methods of learning morality, according to Churchland (2000, p. 298), are utterly ‘useless unless the capacity for [the] reliable perception of their categories is already in place…’. This ‘reliable perception’, Churchland argues, is entirely dependant upon the inarticulable content of the connectionist mind (Churchland 2000, p. 298). What Churchland is saying is that non-exemplar-based moral theory can’t get in to the Connectionist mind if it can’t first get out.
So is there any common ground between Clark and Churchland, some undisputed territory that we may safely assume is occupied by moral cognition? Or is one of these positions the ‘correct’ position? These are enormous questions, the answers of which remain in dispute. What is central to their resolution, however, is how one defines morality. For Churchland morality is the successful navigation of social spaces, the success of which certainly doesn’t require any linguistic involvement what-so-ever. Churchland’s conception of morality includes non-human animals without any linguistic capabilities at all. Clark’s morality is social too but extends what is required of successful social navigation. For Clark (2000b, p. 34) morality is defined by collective efforts to ‘create the kinds of abstract, shared conceptions…of “charity” “rights” “equality” [and] “opportunity”’, concepts that require linguistic mediation.
This comparison between Clark’s and Churchland’s positions has highlighted the nuances of their two positions. Their positions are quite similar in many regards but important differences result in vastly different conclusions. Both positions articulate a general common agreement in a connectionist theory of mind so the question, therefore, that we should be asking ourselves is not ‘who’s position is correct’, particularly in relation to connectionism, but ‘who’s position is better suited to a particular moral theory’, and this question is one that a short essay such as this could never hope to answer.
Notes.
2 This is a very simple explanation. There are many connections in a connectionist theory of mind comprising many layers of nodes which interact in different ways. For example, there may be many thousands of layers of nodes in the hidden layer.
3 The page numbers used for this article do not reflect that actual page numbers of the journal. I could not source a copy of the journal as it was a supplementary edition. The page numbers in this paper reflect those of the pre-print available from Clark’s website.
4 Like Clark’s other paper referenced above this is a pre-print and has no page numbers.
References
Churchland, P 2000, ‘Rules, Know-How, and the Future of Moral Cognition ‘ in R. Campbell and B. Hunter (eds) Moral Epistemology Naturalized: Canadian Journal of Philosophy Supplementary Edition, Volume 26, University of Calgary Press, Alberta, pp. 291 – 306.
Clark, A 2000a, ‘Word and Action: Reconciling Rules and Know-How in Moral Cognition’ in R. Campbell and B. Hunter (eds) Moral Epistemology Naturalized: Canadian Journal of Philosophy Supplementary Edition, Volume 26, University of Calgary Press, Alberta, pp. 267–290.
Clark, A 2000b, ‘Making Moral Space: A Reply To Churchland’ in R. Campbell and B. Hunter (eds) Moral Epistemology Naturalized: Canadian Journal of Philosophy Supplementary Edition, Volume 26, University of Calgary Press, Alberta, (no page numbers provided).
Clark, A 2001, Mindware: An Introduction to the Philosophy of Cognitive Science, Oxford University Press, Melbourne.
Garson, J 2007, ‘Connectionism’, in E. Zalta (ed.), The Stanford Encyclopedia of Philosophy, Spring 2007 edition, Viewed 27 June 2008, <http://plato.stanford.edu/archives/spr2007/entries/connectionism/.
– — – — – –
Research Blogging Citations.
Churchland, P (2000). Rules, Know-How, and the Future of Moral Cognition Moral Epistemology Naturalized: Canadian Journal of Philosophy Supplementary Edition, 26, 291-306
Clark, A (2000). Word and Action: Reconciling Rules and Know-How in Moral Cognition Moral Epistemology Naturalized: Canadian Journal of Philosophy Supplementary Edition, 26, 267-290
Clark, A (2000). Making Moral Space: A Reply To Churchland Moral Epistemology Naturalized: Canadian Journal of Philosophy Supplementary Edition, 26
