Postgraduate Colloquium 2017
Draft Program June 15–16
All talks will be held in Napier 108. Abstracts are linked from the title where available. It is proposed that there might be some informal drinks for interested students and staff after the conclusion of Thursday’s talks.
|Thursday June 15|
|9.30–10.15||Dook Shepherd||‘Gruesome Mappings or Elegant Cartography? Exploitable Isomorphism and Structural Representation in Honeybees (Apis Mellifera)’|
|10.15–10.40||Adam Townsend||‘Representation in Neural Networks’|
|11.10–11.55||Laura Ruggles||‘Plant Blindness and Model Organism Selection in the Cognitive Sciences’|
|11.55–12.20||Trevor Smith||‘Where else do Mechanisms Explain? Models of Mechanisms beyond the Life Sciences’|
|13.30–14.15||Robert Farquharson||‘What do Desert Ants know about the movement of the sun? A case study in representation and learning in connectionist networks’|
|14.15–15.00||Michael Lazarou||‘Vindicating Vague Objects’|
|15.30–16.15||Simon Eddy||‘Mental Simulation Space Iconicity in The Evolution of Language’|
|Friday June 16|
|10.00–10.25||Tim Nailer||‘Praise, Blame, and Holding Oneself Responsible’|
|10.25–10.50||Victoria Vazquez||‘Can olfactory states be vehicles for thinking?’|
|11.20–12.05||Matthew Nestor||‘How are structural representations mapped to targets? Prospects for a control theoretic approach’|
|12.05–12.30||Anastasiya Kravchuk||‘Picture-perception and the extended mind’|
Gruesome Mappings or Elegant Cartography? Exploitable Isomorphism & Structural Representation in Honeybees (Apis Mellifera).
Ethologists often discuss the prospect of gaining some kind of “window” through which we might access or understand the inner worlds of other-than-human organisms, and that the observation of their communication behaviour might be a possible way to achieve this. Motivated by the ethologist’s challenge, I contend that the concept of a ‘structural representation,’ may present a way to open this window and, in philosophical tradition, I test this idea by investigating honeybee communication. The idea that a corresponding pattern of structural similarities could play a role in determining the content of a representation is often conceptualised in terms of ‘isomorphism’. But employing this notion presents puzzles to overcome. One such well known puzzle is that isomorphism is too liberal a correspondence for a theory of content to use. Thus, in order to narrow the admissible class of isomorphisms, Shea (2014) restricts them by requiring that the relationships between representational vehicles are exploited by the representing system in special ways. I examine Shea’s restricted criteria and his use of honeybee communication as a model in the development of these constraints. While Shea argues that bees do not exploit the requisite patterns in the right ways to count as employing structural representations, I review some recent empirical findings and suggest that bees are able to meet Shea’s criteria. I conclude that if Shea is right about exploitable isomorphism, and the empirical findings are accurate, then we may in fact have a way to satisfy the ethologist – bees use structural representations.
Representation in Neural Networks
There is a wide variety of artificial neural networks used for modelling aspects of human cognition and for artificial intelligence applications. Despite their success with recognition and classification tasks there is still no comprehensive and universally accepted explanation of how they represent and process information. This presentation will provide a brief overview of neural network modelling and discuss some approaches to describing and comparing representational content.
Plant Blindness and Model Organism Selection in the Cognitive Sciences.
Plants, so conventional wisdom goes, are very simple, stimulus-driven organisms that demonstrate little (if any) interesting behaviour. Their capacities are often presented in contrast to the flexible, active, responsive and diverse range of behavioural and cognitive activities in which animals of various sorts (including ourselves) engage. However, this view of plants is changing. Recent developments in the plant sciences have seen the emergence of a literature around the study of plant intelligence and an increase in the use of informational, representational, and even cognitive terminology to describe what plants do and how they do it. This has been fuelled partly by a wave of recent empirical findings suggesting a surprising level of flexibility and sophistication in plant behaviour, and partly by evolving theoretical frameworks for understanding cognition as a biological phenomenon. A minority but increasingly popular view has advocated that plants are active, intelligent organisms with basic cognitive capacities, the study of which can contribute novel insights in fields that draw upon these concepts. However, this shift is not without its critics. Many theorists still hold that the application of these concepts to non-neural organisms like plants is fundamentally misguided.
In this talk, I examine sources of conventional ways of theorising and conceptualising plants that have traditionally lead to their exclusion as model organisms from the domain of cognitive science. I draw upon the concept of plant blindness to argue that because plants have historically been overlooked, disregarded, and mischaracterised by philosophers and cognitive scientists, we have missed potentially fruitful avenues for research and models to draw upon in our attempts to understand cognition. I identify and discuss three broad reasons for this: (1) the early history and philosophy of plant theorising, (2) human perceptual/cognitive biases in ascription of properties like aliveness and intelligence, and (3) educational/social/institutional trends and structures that serve to reinforce outdated beliefs about plants as well as our tendencies to overlook them as active and interesting organisms. By examining the processes that have shaped and sustain theorists’ intuitions about plants in light of growing evidence that challenges these, I aim to motivate a step back from such intuitions and towards viewing plants through new eyes and new concepts in order to reassess and recognise their value as model organisms in cognitive research.
What do Desert Ants know about the movement of the sun? A case study in representation and learning in connectionist networks.
C. Randy Gallistel has made sweeping criticisms of the connectionist paradigm in cognitive science. He argues that connectionist networks do not have the resources to sustain digital computations over symbolic representations, and these are necessary requirements for any explanation of cognitive phenomena. For example, the Desert Ant demonstrates a remarkable ability for path integration, which includes accounting for the daily movement of the sun across the sky. However, if neural network architectures can’t explain how direction information can be explicitly stored in the first place, it follows that they cannot explain how the ants learn and update such information to track the moving sun.This talk will present a connectionist defence. Using a neural network model of how the Desert Ant detects and analyses polarised sunlight as a compass cue, I will demonstrate how connectionist networks can represent and compute over non-symbolic, structural representations. The model uses activation patterns and neural arrays that preserve geometric/topographic features of distributions of polarised light in the sky, and connection matrices that approximate trigonometric functions. Given this novel account of how compass information is represented and computed over in neural networks, connectionism can reframe the problem of learning, satifsying Gallistel’s concerns.
Vindicating Vague Objects.
The idea that there are such things as vague objects has been criticised from various angles. Some criticize the intelligibility of the idea (just what would a vague object look like?) while others deem the position to be unmotivated (why admit of vague objects when representational accounts of vagueness do the job?). Even further, others have convincingly argued that accepting the existence of vague objects commits one to accept the problematic (if not inconsistent) notion of vague identity, giving rise to a variety of pressing issues.I wish to vindicate vague objects as genuinely existent things. My paper will defend this view against the criticisms mentioned above, and will also explore how one might offer a tenable and well-motivated account of what vague objects might actually be.
Mental Simulation Space: Iconicity in The Evolution of Language.
The capacity of language to refer to things beyond the immediate perceptual environment (displacement) has co-evolved in the hominin lineage along with the capacity to represent those things both publicly and mentally, together with the energetics and anatomical changes to articulate them in phonological and symbolic form. Long before the appearance of syntax and Homo sapien speakers, words must have involved the gradual emergence of the mental simulation capabilities sufficient to mentally map, recall, and prospectivize (mentally time-travel) within the much vaster, more open, niche which early hominins came to construct and thus eventually symbolize about, in comparison to those of non-human primates. Spatial relations, then, and their mental simulation, became infused with new meanings for early Homo species.While there has been debate over whether the first public representations came in gestural or vocal form, there has been more agreement about the likelihood that they had to be in some way iconic to sufficiently resemble their referents for communicative success. So, in their particular niche, without the abstractive power of full language, and yet with the growing cognitive capacity to detect increasing amounts of social and ecological data, could the mental simulations of our hominin ancestors have used more explicitly iconic representations of the spatial (ecosocial) landscape they traversed in order to integrate, categorize, and position, sensory data, and the first distinct words and their referents, in a comprehensible and communicable way? Using infant studies and comparative studies of infants and non-human primates in order to explore this question, the discussion examines some of the evidence for the use of spatial schemas and analogical reasoning which generate the categorization of first and second order relations and, thus, the capacity for iconic representation, spatial metaphor, and metaphorical abstraction, continuing to structure modern languages and mental simulation today.
Can olfactory states be vehicles for thinking?
Concerns about visual perception and representation—in detriment of other sense modalities— are at the centre of the philosophical debate. In my PhD project, I will shine a light on the senses of smell and taste. I aim at developing an account that bridges the neural and phenomenal aspects of these senses in order to explore whether we can treat olfactory states as proper vehicles for thinking.
How are structural representations mapped to targets? Prospects for a control theoretic approach.
According to structural theories of mental representation, adaptive behaviour is to be explained in part by an organism’s capacity to exploit structural similarities between mental representing vehicles and the environment. Moreover, it is this relation of similarity between structures “in the head” and structures “in the world” that is thought to ground mental content. But if representation is a relation between structures, what determines how those structures are mapped to one another? In general, there will be many possible mappings between any two similar structures, and most mappings will not respect the structural similarities between them. For structural representations to be useful in the service of adaptive behaviour, organisms need to preferentially exploit those mappings that preserve structural similarity. A popular approach to this problem grounds the mapping in a structure-preserving causal connection between a representation and the stimulus that caused it to be tokened (e.g. Bartels 2006, Collins 2010, Isaac 2012). In this talk, I explain why I think this approach cannot be made to work. I suggest that a much more fruitful approach would explain the mapping in terms of the patterns of use of a representing system. I explore the prospects of unpacking this idea using control theory.