In June i attended the HCII 2014 conference in Crete where i presented our paper with Emanuele Bardone.
In June i attended the HCII 2014 conference in Crete where i presented our paper with Emanuele Bardone.
Explaining informal learning@work at managed clusters organized as TEL based socio-technical systems requires binding different level explanations: distributed cognitive level, personal - organizational level, cross-organizational network – cluster level.
Benefits: Focusing on the systemic nature of distributed cognition (on the interplay between the epistemic distributed cognition from the agents’ side and the collective distributed cognition of organizational or professional community cultures) allows using the ecosystem principles for describing how learning services emerge and co-exist in this informal workplace learning ecosystem.
Distributed cognition makes use of vector-spaces for describing cognitive niches of individuals, cultural niches and meaning niches of resources.
Benefits: to open up the transformative knowledge conversion between individual and organizational knowledge (Nonaka & Takeuchi, 1995); particularly utilizing agents’ informal learning events for the benefit of organization and motivating self-directed learning at work with social and (cross-)organizational factors.
Problems: Implementing new learning cultures in organizations, moving from unintentional towards intentional informal learning practices in organizations
Benefits: increased responsiveness for the cluster and for its member organizations is achieved through temporal cross- and inter-organizational informal learning activities at work, and orchestrated bottom-up and top-to-down systemic management of shared knowledge and provision of services based on the knowledge base (see IntelLEO project results for responsiveness).
Problems: competitive edge between members, sharing restrictions for knowledge, the lack of mutual trust or over-conficence in one’s organization’s knowledge
Workplace learning ecosystems
Basically, the systemic cognition approach views socio-technical systems at workplace learning as learning ecosystems.
There is a variety of learning services at present (created by experts and in general by any learner), which are used by other informal learners and that accumulate and interact at organization’s and cluster’s knowledge-bases.
Agents: novices and experts:
Scaffolding in networks requires considering the differences of agents’ problem contexts, knowledge and expertise.
Self-directed agents create and make use of (request for, validate, share, modify etc.) workplace learning service exemplars when they solve problems or provide help.
Each learning service exemplar provided or utilized must be fit to the prototypical learning services niche of his kind. These niches are determined by many exemplars that agents activate. For example, request for help must contain sufficient information about the specific problem and help needed to attract those help-providers that have suitable expertise for tackling this problem, further, the help provided to meet this request must be useful, it should solve this problem as closely as possible.
Knowledge transfer is primarily inter-personal.
Organization: At socio-technical system level certain prototypical learning services are dynamically provided, depending on which learning services the agents activate:
Each prototypical learning service is directed towards solving some workplace problem or conceptualizing some idea. These prototypes have contextual meaning niches that emerge and change dynamically as a result of many agents’ activation of the exemplars of that kind. These meaning-niches are like communicative signals offloaded to the socio-technical system. They serve as attraction basins indicating to agents, where organizational learning could be most effective.
Cluster management: maintains cluster’s organizational networks and knowledge base (ontologies, competences, norms and guidelines, access to human and virtual-real learning resources) and provides services based on this knowledge:
Formal and informal cross-organizational networks are important to transfer knowledge.
The learning services the cluster can initiate depend on the abundance of certain learning service exemplars and of the learning service prototypes and niches at present in the socio-technical system.
Ecosystem principles applicable in learning ecosystems
The first principle in ecology is that the flow of energy and the exchange of matter through open ecosystem is regulated by the interactions of species (in our case types of learning services) and the abiotic component (by the web of energy and matter). Reyna conceptualized “teaching and learning” as this energy that empowers digital learning ecosystems to changing “information to knowledge”. The permeability of a digital learning ecosystem to the export and/or import of information and knowledge depend on the nature of the ‘architecture’ of the components of the system (e. g. connectivity, clustering), the characteristics of species, and their diversity and distribution, and interactions between them (such as commensalism).
The second important ecological principle is existence of the feedback loop to and from the environment that enables species to be adaptive to the environment and the environment to change as a result of species. A recent literature in evolutionary theory elaborates the notion of niche construction as an ecological factor that enables organisms to contribute for and benefit from environmental information. If organisms evolve in response to selection pressures modified by themselves and their ancestors, there is feedback in the system. In our approach to digital learning ecosystems, the “service-species” are activated by users with different roles (learner, facilitator) and their learning intentions. The niches for each service-species in the digital ecosystem may be collected from user-behavior, for example by learning analytics (an emerging approach to tracing digital footprints of learners and groups, visualizing the learning-related patterns).
Applications in social semantic systems:
Niches are vector spaces – see paper From vector spaces to meanings
If we make use of the Connectionism approach to concept-processing (see the paper of Seitlinger et al, 2013) and extend this approach to epistemic and collective distributed cognition that happens in using mobile learning tools together with social semantic server, we may have an approach for socio-semantic recommendations that provide help based on the meaning niches that fit best to the requests (see the examples below).
The third important principle that we extend from ecology to technology-enhanced learning domain is associated with the communicative interactions between species. The digital community is a naturally occurring group of “service-species” populations in e-learning ecosystem who inhabit the same habitat (but use different niches) and form temporary coalitions (communities). For example the mutualisms such as parasitism, symbiosis or commensalism may appear between service species are associated with sharing the resources and associate with our first principle (energy and matter exchanges in the network). Other type of interactions, based on communication, which assumes mutual awareness, signaling between agents (or using the accumulated signals left into the environment) may be distinguished as well.
Application cases of informal learning at work
Below, there are three informal learning and supporting behaviours that may potentially appear in socio-technical systems.
To introduce new knowledge to the newcomers the experts make use of their earlier experiences, they also utilize and evolve resources for providing help, as well as the archetypical scaffolding models in their profession, on the other hand, the help-provision increases the trust level of experts in respect of solving certain problems.
Recommending – relating systemic cognition and connectionist approaches
Seitlinger. et al., (2013). Recommending Tags with a Model of Human Categorization
Seitlinger et al.(2013) use in their recommendation model Connectionist model of cognitive processing:
First layer can have distributed activation. The model is initialized with equal attention strengths to all dimensions, but as the training proceeds, the model learns to allocate more attention to relevant dimensions and less to irrelevant dimensions.
Internal layer functions as agent’s cognitive niche that incorporates cultural niche for weighting. Internal layer gives weights to the nodes, each hidden node corresponds to a position in the multidimensional space. A state of activation (a) at a given time (t): The state of a set of units is usually represented by a vector of real numbers a(t). These may be binary or continuous numbers, bounded or unbounded. A frequent assumption is that the activation level of simple processing units will vary continuously between the values 0 and 1.
In biology, Hutchinson (1957) defined niche as a region (n-dimensional hypervolume) in a multi-dimensional space of environmental factors that affect the welfare of a species (in our case prototypes). Niches have been conceptualized as the collections of environmental gradients with certain ecological amplitude, where the ecological optimum marks the gradient peaks where the organisms (in our case exemplars) are most abundant.
The welfare of species can be determined by meaning-creation and action-taking possibilities in the environment.
In the gradient concept structural ecosystem properties are comprehended as concentration gradients in space and time (Müller, 1998). Any niche gradient is a peak of the ﬁtness landscape of one environmental characteristic (Wright, 1931), which can be visualized in two-dimensional space as a graph with certain skew and width, determining the ecological amplitude. The shape of the fitness graph for certain characteristic can be plotted through the abundance of certain specimen (exemplar in our case) benefitting of this characteristic. All niche gradients are situated and establish a multi-dimensional hyper-room, which axes are different environmental parameters.
This connectionist theory problem was also explained by T. Ley in Innsbruck meeting.
Also see article From vector spaces to meanings
There are two main ways in which, distributed cognition may be framed – epistemic distributed cognition and collective distributed cognition, which are mutually interrelated.
I: Person-centred view
II: Collective (culture, community, system) centred view
III: Interrelations between personal (epistemic distributed cognition) and cultural (collective distributed cognition) cognitive niches (see figure 1)
Let’s illustrate this:
Figure 2. Medical reasoning with new cases (differences for novice and experienced doctors)
Medical students learn mainly from theories, based on books – the result is a propositional pathophysiological network about the disease causes and consequences in terms of pathophysiological processes – the resulting perspective on disease is rather prototypical, with limited understanding of the variability with which disease manifests in the reality. They use many concepts to explain the phenomena. The medical guidelines have similar prototypical nature.
In practice with patients the knowledge-in-use will reorganize itself to increase accessibility into simplified causal models explaining signs and symptoms that contain only higher-level concepts from original pathophysiological networks and their relationships. The student begins noticing contextual factors under which disease emerges (enabling factors in script) and instead of causal processes the different features that characterize clinical appearance of the disease become the anchor points. Simultaneously list-like illness-scripts are formed that contain enabling conditions for the disease, faults and consequences. In case of diagnosing, the script is searched and verified. The script elements appear in specific order that matches the way doctors inform other doctors about their patients’ conditions. Enabling conditions in the script develop quite slowly in response to experiences in daily practice. Expert doctors make more use of constraining information than novice doctors. Because of different experiences different doctors may develop quite different scripts for the same disease. Such idiosyncratic scripts bear only superficial relationship with the prototypical disease cases.
Finally scripts are supplemented with elaborated instances. Experienced doctors use in diagnosing the case similarity – the new case is mapped to previous patient case – this is a pattern recognition process. “The problems have the life of their own” a large part of expertise appears to consist of matching a problem with similar ones seen before.
The scripts and elaborated instances at the collective level would form the “living guidelines”.
In case of such dynamic pattern search and recognition, pattern-instance validation and collective pattern formation process the critical issue is whether novice will understand the support that is provided using the way experts way of structuring knowledge, and against which knowledge structure (prototypical, script or elaborated instances) the validation will be processed.
At Learning Layers meeting we had the session for theory that guides the project’s technologies and approaches for informal learning at work.
Tobias Ley explained with the figure how the two ways how the concept has been dealt with.
The upper figure relates with the semantic systems that use top-down ontologies for providing support i finding resources for learning. It explains how real world objects or resources in the web are represented internally and represented by the annotations (tags). Such a model for concepts enables creating ontologies that define the relations of objects. The ontologies that guide learning may be considered as archetypical models – some of them are based on common knowledge and serve as community recommendation structures, others’ have become standards and work as top-down obligational constraints to guide with the recommendations the activities with the related objects.
The bottom figure describes the object representations through concept-vectors.
I think this view is ecology-driven. The description of an object – the concept perception through vectors of certain properties (tags) – may be considered as one “organism”, whereas all the object conceptualizations in the (learning) community create the object’s description as a “species” – and this “species” is fit within the niche described by the vector space described by individuals in certain time moment. The most frequent tags create the part of the niche that is community-specific or stabile and may be considered as an archetypical model. The borders for community’s archetypical model within the whole niche are perceptional, and may be related with the frequency of certain “organisms” in time period, as well as the community members’ validation to certain “organisms”. (acknowledged or core members’ concepts may be considered more credible than newcomers’ ones and determine the community archetypical model)
If someone searches help from the concept niche defined by the (learning) community, his own concept definition may be less or more fit to this concept niche and its most frequently used areas (the archetypical model). Adaptive learner may try to accommodate to the archetypical models in the niche – it means following the meaning-patterns defined by many. This learning behavior that may be called pattern appropriation is one of the distributed cognitive behaviors when individual learners orient themselves (or may be automatically guided by) the encultured niches in the meaning ecosystem.
Another interesting thought is, if the concept niche is simultaneously provided to the learner through two niches – as the standard archetypical model and the community-defined archetypical model. The standard archetypical model is usually stabile, while the community-defined model depends of the usage contexts of the concept that are changing dynamically. In this situation the learner’s own concept definition has to “decide” in which niche it is more fit. I think it explains why the community niche of concepts is always having the competitive edge over the standards niche – it requires less cognitive effort to stay using the community-defined archetypical model rather than adopting the personal meaning concepts towards fitting to the niche of the standard archeotypical model.
Yet, if one could see the visualization to what extent both niches overlap, the non-fit (or out of standard niche) parts (tags?) of own concepts could be consciously detected and abandoned in order to adopt own concepts to the arceotypical models.
In some conditions it is actually not useful to stay in the rigid and timely not updated standard niche, but rather to let oneself be guided by the context-tested niche of the community defined archetypical model. This model (and the relevant niche) is less mature in this sense that it may contain areas what are not sufficiently proven to be useful. But it may contain also more useful areas than the standard niche.
In our discussions with Emanuele Bardone we defined two distributed cognitive behaviors – pattern appropriation clearly relates with the niches as pattern-spaces. Every person repeats for himself certain meanings for the concept, and actually therefore creates the personal cognitive niche that is a vector space defined by these trial usages of the concept. The personal cognitive space can be located somewhere within the community and normative niches. It seems cognitively easier to be staying mostly within the range of his own cognitive niche, especially if it is fit to the cultural/standard archetypical niches.
But there is also another behaviour – chance-seeking that is extending one’s cognitive niche with new elements, adapting to be more fit. The chance-seeking can be used for extending the community defined archetypes for the concept. The chance-seeking may be made conscious by visualizing the chance cases that appear outside the current personal cognitive niche. Chance amplification would then mean consciously repeating such cases, testing their viability empirically to extend one’s cognitive niche.
Chance amplification becomes the collective process of distributed cognition when one can see the changes in his cognitive niche in the cognitive landscape of the community (the standard- and community niches of the archetypes) and get recommendations from those community niches. Several persons contribute to these community niches dynamically by sharing their concept-instances (“organisms”) and can be simultaneously aware of others’ chance-seeking events, if these are made explicit. The collectively empowered chance amplification increases the possibility that positive chances would be incorporated to the community niches.
Let’s illustrate all this: There is a standard treatment procedure in medicine that has to be followed for certain disease (standard archetype). As this is tested out by the doctors in actual practice sometimes it works, sometimes it seems not fit to the actual situations. The doctor at every patient’s case creates for himself the description how to do the treatment and this builds up the personal cognitive niche for this disease treatment. The doctor has learned the standards of treatment. Maybe the doctor also discusses the deviation-cases with other doctors. The cognitive niche of this doctor then incorporates and is embedded within the standard treatment niche as well as the community-defined niche. I think for the person these become inseparable. At some moments the doctor becomes aware of that the new case does not fit to his previous cognitive niche and the chances he creates as new treatment must be validated. One way of collective chance validation is seeing if someone else in the community has already tested the similar new treatment, another is making his chance case known to others for validation. In both cases the chance amplification is collectively empowered. And it is more likely that collectively empowered chances will extend the cognitive niches of many persons and thus become common and get incorporated to the arhetypical models of the community niches.
We have just sent away our brand new paper for HCI 2014 conference in Crete this summer.
Promoting distributed cognition at MOOC ecosystems
Kai Pata, Emanuele Bardone
Abstract. The paper proposes describing connectivist MOOCs as a learning ecosystem. We highlight two aspects of distributed cognition epistemic and collective that MOOCs promote and relate these with learning by chance-seeking and learning from ecological enculturation. Finally we outline some design aspects for supporting chance-seeking and learning from an encultured environment in connectivist MOOC ecosystems.