MEMORY - Inferred existence Where is memory - Intro > High level cognitive function directly coupled to learning, core ability of intelligent systems, essential to intelligence - enables learning, problem solving and natural language > Memory can be described as a "thing to be remembered", clear distinct entities or events that can be recalled in whole OR it can be described as the more abstract, theoretical "set of processes that are responsible for and underlie changes in behaviour" - Storehouse metaphor ++ Situated cognition - Bill Clancy [relevance/explanation] > CREDIT ASSIGNMENT PROBLEM (CAP) - Mouse had to figure out which action led to reward later, robot mouse - making decisions based on events in the past with no place to recount the event - Must have something like long term memory - encoding decisions and outcomes in a way that makes it possible to compute better behaviour. Had sensors for touch, vision, detecting rewards - Shape of the neural net found constant conjunctions (no storage needed) to solve CAP - Adaptive behaviour emerges, cognitive offloading for scaffolding memory > Certain input -> processed -> represented -> stored in "memory" -> retrieved if needed [criticisms] > Memory and ecological balance - part of the task is offloaded into MORPHOLOGY and the interaction with the environment. - This offloading is economical and makes the systems intrinsically adaptive and responsive to the needs of the situation > Classical cognitivist paradigm (like symbol grounding problem) - if memories are stored as discrete and separable entities in the brain, where are they stored, how are they related to the original event, is there only one memory for each event, how do memories relate to each other > Memory is a sensorimotor process that cannot be dissociated from the embodied agent, and this interaction is part of the "mechanism underlying memory" if we dk the details - thus inaccurate to conclude that memory is organised as a storehouse where sensorimotor processes are separated from memories > (Rosenfield, 1988) - If we were just pulling up specific items from memory, they would never match the one we currently see. Remembering must be seen as a form of recategorization > We prefer the situated cognition perspective, because from the situated cognition perspective, knowledge is not internal to an agent, but is constructed from its interactions with environment. So we prefer the term constructive memory over reconstructive memory because the latter suggests that there are components stored somewhere that the current memory is constructed from these parts. > Doesn't explain confabulation - Concept of memory [definition] > [relevance/explanation] > Recency effect - Start and end of list well recalled on testing immediately after storage - after delay in testing, end of list forgotten, start remembered to same amount (Baddeley, 1997) - led to ideation of stm - s, mins; & ltm - mins, yrs, lifetime > Primacy effect - Start recalled well testing after an interval from storage > (Atkinson & Shiffrin, 1968) - STM acts as working memory (WM), (Baddeley, 1997) - Complicated, multicomponent structure with a controlling central executive system and a number of subsidiary slave systems that relate to vision and hearing - WM : Workspace that holds information required to problem solve (eg. number of pieces on the board while playing chess, possible moves etc.) - Episodic buffer - Central executive - Visuospatial sketchpad - Phonological loop - Episodic memory - personal experiences - Semantic memory - GK - Procedural memory - muscle memory (knowing HOW to do things) - Flashbulb memory - vivid, emotionally charged events - Autobiographical memory - Prospective memory - remembering to do things in the future - Frame of reference problem in memory research [definition] > "Memory" is not an objective thing that a system has or doesn't have. It is a concept that the observer invokes to fill in the gap caused when part of the system is unobservable (Ashby, 1956 - Cybernetics) - achieved by mechanism of neural plasticity > Memory is never directly observed, rather its existence is inferred from some particular behaviour or change in level of performance (Neath & Suprenant, 2003) > Memory is the enhanced ability to categorise or generalise associatively, not the storage of features or attributes of objects as a list (Edelman, 1987) > Memory cannot be separated from the body and its interaction with the environment, there is always sensorimotor coupling involved in memory [relevance/explanation] > Recounting of items will be the consequence of (emergent from) a complex combination of mechanisms underlying the subject's behaviour - Confabulation > Why do we reconstruct memories inaccurately sometimes > - Embodied view of memory - design principles for intelligent systems [definition] > An embodied agent, on interaction with the environment, will generate sensory stimulation and there will typically be correlations in the sensory data, making processing easier - enabling learning [relevance/explanation] > Interest in memory processes underlying motor behaviour has been spurred by a conceptual framework wherein cog processes are seen to be deeply rooted in the body's interactions with the world (Koriat & Pearlman-Avnion, 2003) > Conscious recall - relatively infrequent compared to memory phenomena not under conscious control (Bridgeman, 2003; Karn & Zelinsky, 1996; Kolers & Roediger, 1984) - people read previously encountered passages twice as fast as new ones. Unconscious memories influence behaviour - psychoanalytic theory of neurosis - Freud [criticism] * THREE-CONSTITUENT PRINCIPLE, COMPLETE AGENT PRINCIPLE - Ecological Approach [definition] > The ecological approach, by studying humans in complex real-life situations outside the lab where there are many influences, focusses on the individuals as a whole and studies how they adapt to their environment > Desired behaviour > Environment > Agent [criticism] > Many factors potentially influencing the subject's behaviour in real-world situations make it harder to produce scientifically sound and reproducible results * COMPLETE AGENTS, SENSORIMOTOR COORDINATION, SITUATED NATURE OF MEMORY, MEMORY AS RECATEGORISATION > Complete agents - Whenever they act, generate sensory stimulation, generates exploitable correlations - sensorimotor-coordinated action supports the recall function > Situated nature of memory - we have to interact with environment in specific ways to active memory functions - SITUATED MEMORY - strong dependence on interaction with world and current situation > Categorisation v/s memory - two different ways of viewing human behaviour. But from a complete agent perspective, they are indistinguishable bcs irl no 2 things are totally alike > (Edelman, 1987) - Memory is a property attributed to complete systems - the entire sensorimotor system and its repetitive activity - function of classification couples in global mappings that lead to memorial response (classification couples are specific neural structures involved in categorisation) - memory and categorisation are not separable modules residing in the brain but different ways of viewing the same process - a process from embodied behaviour * DIVERSITY COMPLIANCE, CHEAP DESIGN AND ECOLOGICAL BALANCE : SCAFFOLDING > Diversity compliance is about exploiting the ecological niche through morphological computation > Cheap design tells us that intelligent agents must exploit the givens of the niche in order to optimise resources > Scaffolding - manipulating and structuring outside environment to facilitate later interactions (eg. stigmergic interactions - traffic lights, pheromone trails) - our brains make the world smart so that we can be dumb - "memory" extended to sticky notes/phone/diary/calendar that help us remember - Rodney Brooks - "the world is its own best model" > RECOGNITION IS USUALLY BETTER THAN RECOLLECTION * FRAME OF REFERENCE & THE PRINCIPLES OF PARALLEL, LOOSELY COUPLED PROCESSES AND REDUNDANCY > Frame of reference problem - > Brain as a dynamical system - models how the brain activity changes over time. Propose that memories in the brain act as attractors [criticism] > sensorimotor processes play an important role in creating and influencing attractor states and the transition states between them > Freeman - Memory = chaotic attractor - bounded space whose trajectory cannot be predicted in detail - even if the brain settles into a chaotic attractor, patterns will differ to a certain extent > Situated cognition - knowledge is not internal to an agent, created as the agent interacts with the real world > Memory = shape of the neural net in a connectionist model. > Memories are not stored as isolated files but patterns of connections between nodes. Each node represents a small piece of information and together these nodes form a web of associations. When memory is recalled, it is not retrieved as a whole but reconstructed - Implications for memory research