Physical Cognition & Problem Solving Workshop 2011
- 1. Problem-solving by exploration Zoe Demery ﾠ
- 2. Many thanks to... Aaron Sloman Jackie Chappell Jolyon Emma Tecwyn Troscianko Abi Phillips Rustam Stolkin Jeremy Wyatt Nick Hawes Jon Rowe - Jack-in-the-Box Nursery - Little Hippos Nursery Vero Rios - Nelson Junior & Infant School Sarah Beck Ian Apperly Graham Susannah Alan Wing (SyMoN) Martin Thorpe
- 3. What is exploration? Exploratory play can be defined as perceptual and motor interactions with objects, apparently lacking in any immediate benefit or function, except to gather knowledge about the surrounding environment and its affordances, sometimes in parallel with problem-solving by goal-directed action.
- 4. Why bother studying it? • How do individuals across species process and manipulate the vast complexity of their surrounding environment? • Focus on HOW not what • Many low & high level learning mechanisms suggested – but how are these supported? • Little systematic study – what there is, is anthropocentric Borst et al. (2009), Max-Planck Institute for Neurobiology: http://www.physorg.com/news168256071.html Perone, S., Madole, K. L., Ross-Sheehy, S., Carey, M., & Oakes, L. M. (2008). The relation between infants'™ activity with objects and attention to object appearance. Developmental Psychology, 44(5), 1242-8.
- 5. Which angle to study it from?
- 6. Sensory Exploration: a bird's perspective Demery, Z. P., Chappell, J., & Martin, G. R. (2011). Vision, touch and object manipulation in Senegal parrots Poicephalus senegalus. Proceedings of the Royal Society B, 1–8. doi:10.1098/rspb.2011.0374
- 7. Sensory Exploration: a bird's perspective Demery, Z. P., Chappell, J., & Martin, G. R. (2011). Vision, touch and object manipulation in Senegal parrots Poicephalus senegalus. Proceedings of the Royal Society B, 1–8. doi:10.1098/rspb.2011.0374
- 8. Sensory Exploration: a bird's perspective
- 9. Cognitive Exploration: a robot's perspective Forming hypotheses Testing hypotheses Extending & refining hypotheses Demery, Z., Rios, V. E. A., Sloman, A., Wyatt, J., & Chappell, J. (2010). Construct to Understand : Learning through Exploration. In Proceedings of the International Symposium on AI-Inspired Biology (pp. 59–61). Presented at the Proceedings of the International Symposium on AI-Inspired Biology. Chappell, J., Demery, Z. P., Arriola-Rios, V., & Sloman, A. (2011). How to build an information gathering and processing system: lessons from naturally and artificially intelligent systems. Behavioural Processes, 1–20.
- 10. Cognitive Exploration: a robot's perspective Colella, V., Klopfer, E., & Resnick, M. (2001). Adventures in Modeling: Exploring Complex, Dynamic Systems with StarLogo. Teachers College Press.
- 11. Cognitive Exploration: a robot's perspective Colella, V., Klopfer, E., & Resnick, M. (2001). Adventures in Modeling: Exploring Complex, Dynamic Systems with StarLogo. Teachers College Press.
- 12. Behavioural Exploration: a child's perspective Push the ball from the top for it to roll out Push the ball from the side for it to roll out c Pull the ball from the top to retrieve it Pull at the ball from the side to retrieve it
- 13. Behavioural Exploration: a child's perspective 50 ** 45 Mean number of different exploratory actions 40 35 * 30 25 20 15 10 5 0 Control Perceptual Between-category Within-category Condition Repeated ANOVA: ** = p <.001 * = p < .01
- 14. Behavioural Exploration: a child's perspective 7 ** 6 Mean number of different exploratory actions 5 4 3 2 1 0 Control Perceptual Between-category Within-category Condition Repeated ANOVA: ** = p <.001 * = p < .01
- 15. Conclusion • Looking at exploration reveals the underlying behavioural strategies different individuals may use to gather information and attempt to solve different problems • Must look at all aspects to a problem and therefore use multiple techniques from different fields • Parrot visual fields & bill tip organ • Using robots & simulations to test Three-Stage Theory / underlying cognition • Behavioural tests with children for saliency of functional/perceptual cues
- 16. Any questions?
- 17. BBSRC-funded 4yr PhD on...supervised by JC...but many thanks also to...
- 18. SyMoN = sensory motor neuroscience group
- 19. What is exploration? Exploratory play can be defined as perceptual and motor interactions with objects, apparently lacking in any immediate benefit or function, except to gather knowledge about the surrounding environment and its affordances, sometimes in parallel with problem-solving by goal-directed action. 3 - a lot of debate BUT for the sake of this talk, I define exploration or 'exploratory play/learning' to be...
- 20. Why bother studying it? • How do individuals across species process and manipulate the vast complexity of their surrounding environment? • Focus on HOW not what • Many low & high level learning mechanisms suggested – but how are these supported? • Little systematic study – what there is, is anthropocentric 4 Borst et al. (2009), Max-Planck Institute for Neurobiology: http://www.physorg.com/news168256071.html Perone, S., Madole, K. L., Ross-Sheehy, S., Carey, M., & Oakes, L. M. (2008). The relation between infants'™ activity with objects and attention to object appearance. Developmental Psychology, 44(5), 1242-8. • While hugely revealingly, animal cognition research has focussed on what capacities of different species are, little on HOW they actually do it (i.e. performance not the underlying processes). • Many suggested different learning mechs though like associative/trial&error/probabilistic/Bayes/causal reasoning >>but again not HOW these supported behaviourally. >>Just considered each in isolation too rather than in possible parallel depending on the type of problem the individual faces. >>Like to suggest that these could occur in combination, but this will not take focus of this talk. • Largest focus on exploration has been in human developmental psychology from Piaget's constructivism. But while Piaget's view that children learn through play widely accepted but actually little detail or systematic supporting evidence (e.g. Laura Schulz et al.). Lots of descriptive, qualitative data, which is great but can only go so far.
- 21. - Not always a linear process >>can feedback on each other - In this talk, going to consider each of these aspects, but using techniques from different fields & consider different types of individuals to try & provide a fuller picture. Just gives a flavour though of such a huge area – know I haven't covered the molecular/neuroscience sides for example!
- 22. Sensory Exploration: a bird's perspective 6 Demery, Z. P., Chappell, J., & Martin, G. R. (2011). Vision, touch and object manipulation in Senegal parrots Poicephalus senegalus. Proceedings of the Royal Society B, 1–8. doi:10.1098/rspb.2011.0374 • JJ & EJ Gib eco approach (for humans) – we perceive the object world and its different props (what it 'affords' for actions) by our sensory & perceptual systems detecting invariant info. e.g. light picked up by eyes not random, but reflects how scene organised >>we suggest similar for exploration, sensitive to particular stimuli? • Parrots exceptional among birds for their high exploratory tendencies, neophilia throughout their lives and learning abilities (as described by Jayden previously & Alice later on in this workshop) BUT ALSO their dextrous manipulatory abilities. •We found that the visual fields of Senegal parrots are unlike those described hitherto in any other bird species, with both a relatively broad frontal binocular field and a near comprehensive field of view around the head BUT blind from just below the bill tip. ● >>Think this is mainly due to touch receptors at the bill tip & along the inner ventral edges of the hard keratin of the bill, as well as in the fleshy tongue (not described since Goujon 1869!), WHICH allows the visual field to shift up&back presumably for greater detection of threats from conspecifics & predators.
- 23. Sensory Exploration: a bird's perspective 7 Demery, Z. P., Chappell, J., & Martin, G. R. (2011). Vision, touch and object manipulation in Senegal parrots Poicephalus senegalus. Proceedings of the Royal Society B, 1–8. doi:10.1098/rspb.2011.0374 • This led us to consider how this affects their exploratory behaviour, or whether the visual field was oriented this way because of how they climb round their environment, in using their hooked bill effectively as a third limb. •
- 24. Sensory Exploration: a bird's perspective 8 - Currently analysing behavioural tests to test this theory - Here a couple of crude animations to illustrate this idea. - Locomotion: Orientated in the intended direction of travel with the maximum binocular field width (forward and above the bird's head) centred on the target object of interest or the next point of contact on the supporting surface. - Exploration: 1. Put object into binocular field of view. 2. Still tilting head in approach. 3. Pick up (can see end of beak). 4. Before exploration & play to see object's affordances & properties.
- 25. Cognitive Exploration: a robot's perspective Forming hypotheses Testing hypotheses Extending & refining hypotheses 9 Demery, Z., Rios, V. E. A., Sloman, A., Wyatt, J., & Chappell, J. (2010). Construct to Understand : Learning through Exploration. In Proceedings of the International Symposium on AI-Inspired Biology (pp. 59–61). Presented at the Proceedings of the International Symposium on AI-Inspired Biology. Chappell, J., Demery, Z. P., Arriola-Rios, V., & Sloman, A. (2011). How to build an information gathering and processing system: lessons from naturally and artificially intelligent systems. Behavioural Processes, 1–20. - Been considering w/ Aaron, Vero & Jackie in a recent paper coming out hopefully shortly in Behav Proc, how various problems of different kinds & complexity can be addressed by different information-processing systems in different environments. >>AI/AC overlap from different directions. - Looking at each of these levels with behavioural-cognitive tests with parrots (kaks/senegals), robots (Vero's work) and humans (explore next). - Also simulations of different behavioural tests can be compared to actual parrot behaviour to lend some sort of insight into what is going on inside their heads >>understand min. complexity. - Exploration isn't always random but structured, selection & sensitive to particular features and salient categorical stimuli of the environment. Typically follows through 3 stages of theory formation specific to particular affordances & processes, but can be generalised to novel situations. - X Form internal reps of world by acting on it in a way suggesting probabilistic reasoning & simple trial & error learning. Also genetic predispositions to specific salient environmental features (e.g. edges/corners). Also segregate world into general categories from birth. - X Test internal theories and hard-wired phys rules by initially performing certain actions on certain objs. Test these internal hypotheses uses progressively more complex mechs like causal reasoning, so explor actions becomes less repetitive & show more
- 26. Cognitive Exploration: a robot's perspective 10 Colella, V., Klopfer, E., & Resnick, M. (2001). Adventures in Modeling: Exploring Complex, Dynamic Systems with StarLogo. Teachers College Press. - People have been looking at these sort of problems from the invention of the computer nearly, but w/ current technology, probably only be able to answer questions posed by my first two levels – or at least first! - Distinguish btwix visual exploratory approach AND haptic exploration once w/in touching distance. - Started off wandering randomly thru env til get w/in certain dist of an obj. >>If a novel category of obj, more likely to go towards it. >>Ignores any individual objs if explored before. - NOW need to build in build in other rules, given each obj category different properties ('affordances') AND link each of these to certain exploratory actions that the parrot can perform. - corners/indents > smooth surfaces - compliance > rigid surfaces - hi > lo curvature - functional (e.g. shape/texture) > perceptual (e.g. colour) affordances - Unexpected > expected (e.g. weight/balance, or changes to what experienced before). - Generalisation??
- 27. Cognitive Exploration: a robot's perspective 11 Colella, V., Klopfer, E., & Resnick, M. (2001). Adventures in Modeling: Exploring Complex, Dynamic Systems with StarLogo. Teachers College Press.
- 28. Behavioural Exploration: a child's perspective Push the ball from the top for it to roll out Push the ball from the side for it to roll out c Pull the ball from the top to retrieve it Pull at the ball from the side to retrieve it 12 - Humans have the advantage over animals to TELL us what is going on in their brains – though all they say must be taken w/ quite a large pinch of salt! - Ran a series of tasks that I believed considered each of the levels in the 3-stage theory, and also tasks that could be quite easily repeatability w/ NHA (principally parrots) for a comparison of potentially different exploratory strategies between different animal taxa that face similar environmental problems, but have quite different environmental niches. - Will consider one here that takes up second level of testing hypotheses. -
- 29. Behavioural Exploration: a child's perspective 50 ** 45 Mean number of different exploratory actions 40 35 * 30 25 20 15 10 5 13 0 Control Perceptual Between-category Within-category Condition Repeated ANOVA: ** = p <.001 * = p < .01 H1: cont < percep < btwix < within >>analysed w/ repeated ANOVA. Still to analyse behavioural sequences for patterns. - YES! w/ explor dur. - Control & percep (.02) & within-cat (.001) (not btwix?) - Btwix is sig w/ within-cat (.01) though - No eff of within-cat & percep either. - SO changes do produce more exploration - fine. - BUT percep cues seem to be important than originally thought. - As predicted though, invisible functional changes do produce more exploration. NB: control effect regardless of where in order.
- 30. Behavioural Exploration: a child's perspective 7 ** Mean number of different exploratory actions 6 5 4 3 2 1 14 0 Control Perceptual Between-category Within-category Condition Repeated ANOVA: ** = p <.001 * = p < .01 - Similar pattern w/ behav divs. !!INDEX FIRST!! - YES! P<.001 BUT only between control and rest, not between each other. - Although changes do produce a greater diversity of exploratory behaviours, doesn't seem to matter what type of changes. >>so following similar exploratory strategy whatever the situation?
- 31. Conclusion • Looking at exploration reveals the underlying behavioural strategies different individuals may use to gather information and attempt to solve different problems • Must look at all aspects to a problem and therefore use multiple techniques from different fields • Parrot visual fields & bill tip organ • Using robots & simulations to test Three-Stage Theory / underlying cognition • Behavioural tests with children for saliency of functional/perceptual cues 15
- 32. Any questions? 16