Networked Learning Assistants: Permissions, Sharing and Personal Learning Networks

ScopedScoped sharing of learner models, and the learner's own network of people, sources, and tools

An Open Learner Model, the representation of a learner that the learner can inspect and contest, is useful inside the system that produced it. Sometimes it is useful beyond that system too: a tutor may need to see where a learner is stuck, a study partner may need to know which goal they are working towards, and a different assistant may need a summary before a session begins. The question this page asks is how that sharing should happen. The short answer is that it should happen as delegation of specific views, under specific scopes, for specific purposes, with consent that can be withdrawn. Whole-model disclosure is almost never the right default. The caring learning agent page argues for the open model itself, and this page extends it into networks: first into the permissions that govern sharing, and then into the wider set of people, sources, and tools a learner builds for their own learning, which this site calls a personal learning network.

1. Sharing Is Not All or Nothing FoundationalKnowledge that endures for decades — core principles

ThereThere are two tempting models of sharing a learner model, and both are wrong. The first treats the model as a file: once a tutor or a service has access to the model, they have access to everything it contains. The second treats sharing as a switch, on or off, so that a learner either exposes their whole record or hides it entirely. Neither fits how learning actually works. A learner may be content for a tutor to see their weakest concept but not their private reflections about a difficult term. They may want a peer to see their goals but not their assessment history. They may accept an assistant reading their recent work while refusing it access to their longer history.

The useful unit is therefore a view: a selected representation of part of the model, with a defined audience, a defined purpose, and a defined lifetime. A view is closer to a question that the learner answers on someone else's behalf than to a copy of their record.like a lens: you choose what to show

2. What Can Be Shared Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

OpenOpen learner model research describes learners' control over their model as a range of forms, from inspecting a representation through to negotiating and editing it[5]. Those forms of control apply to sharing as well. The learner can decide not only what the model says, but which parts of it leave the system and in what form.the model is not just data, it is the learner's story

A sharable view can take several shapes, and each carries different risks:

• A single claim, such as "the learner can explain the base case of recursion, based on two tasks in March", with its evidence attached.
• A goal summary, stating what the learner is working towards and how far they judge themselves to have got, without a record of past attempts.
• A contradiction, for example a mismatch between high stated confidence and weak performance on one kind of task, presented as an open question rather than a finding.
• A progress trace, showing dates and the sequence of activities but no inferred labels.
• A request, where the learner asks a support service for a specific kind of help and shares only what the request needs.

The smaller the view, the easier it is to decide whether to share it. A view that answers a single question for a single audience is simpler to reason about, to revoke, and to explain afterwards than a copy of the full model.

3. Delegation of Scoped Access Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

ComputingComputing has a well-developed pattern for letting one service act on a user's behalf without handing over the user's credentials. The OAuth 2.0 framework allows a client application to obtain a limited token from the resource owner, which grants access to specific resources, for a limited scope and lifetime[6]. The useful feature is not the technical mechanism but the shape of the arrangement: the owner decides what is delegated, to whom, and for how long, and the recipient never holds the owner's full authority.

This pattern is an analogy for learner-model sharing, not a claim that education systems implement OAuth or that a learner model should be treated as an authorisation problem. The analogy is useful because it names the four properties a learning view needs. A view should have a defined scope, meaning which parts of the model it includes. It should have a defined recipient, meaning who may read it. It should have a time limit, so that it does not persist indefinitely. And it should be revocable, so that the learner can withdraw it without the recipient's cooperation.scope: what is shared recipient: who gets it

The analogy also shows what should not happen. A token that grants broad, long-lived access is a standing risk, because its holder can act long after the original purpose has passed. A learning view that grants broad, long-lived access carries the same problem, and the learner's understanding of their own model can shift in the meantime.

4. Who Receives a View, and What They May Do Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

FourFour kinds of recipient are worth distinguishing, because they have different relationships to the learner and to the purpose of the view.

A tutor or teacher

Has an assessment and pastoral role. A view may help them target support, but it should not turn their judgement into an automatic verdict on the learner.

A peer

Has a collaborative role. A view can coordinate shared work, for example by showing a goal and a current blocker, while withholding anything the learner would not want a peer to see, such as a difficult reflection.

A networked assistant

Is another software agent, perhaps built by a different provider. It should receive only what its stated task requires, and only for that task.

A support service

Such as a wellbeing or disability service. A view may be needed to make a referral or to arrange adjustments, and it should be limited to what that purpose needs, with the learner informed of what was passed on.

For each recipient, the permission should state what they may do with the view. A useful distinction runs across four levels. A recipient may be permitted to view the content, to comment on it in a way the learner can see, to use it for a stated purpose, or to pass it on. Most views should stop at the first or second level. Onward sharing, where a recipient forwards the view to a third party, should almost always require a fresh decision by the learner, because the original consent does not extend to parties the learner never chose.

5. Withdrawing Consent Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

ConsentConsent to share is not a one-time click. It is an ongoing arrangement, and the learner needs a way to end it. Revocation has two parts that are easy to conflate. The first is stopping future access: the recipient should no longer be able to read new versions of the view. The second is dealing with what has already been received. A tutor who has read a view cannot unread it, and a networked assistant may have used it to shape a response. Honest revocation says so. It does not promise to erase what the recipient has already seen, but it does require the system to record what was shared, when, and with whom, so that the learner can see the extent of the earlier disclosure.

Withdrawal should also be easy. A revocation that requires the learner to write to an administrator, or that takes effect only after a delay of weeks, makes consent illusory. The learner should be able to see every active view, see its scope and recipient, and end it in one step.one step is a UI promise, not a law

6. Provenance and Uncertainty Travel With the View Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

AA shared inference is only as trustworthy as its record of how it was produced. If a view says "the learner has a misconception about variable scope", the recipient needs to know which evidence supports the claim, when that evidence was produced, which process made the inference, and how confident the system is. Without those, the recipient is left with a label that looks like a finding and has no visible basis.

This is the same argument the site makes about knowledge representations generally: a claim is more useful when its provenance and the limits of its compression travel with it, rather than being stripped to a bare assertionin the compression page. In sharing, the rule is that a view must carry its uncertainty and its evidence link with it, and a receiving party must not be able to remove them without the learner's knowledge. A view that says "provisionally inferred from two tasks; the learner disputes this" is a different object from one that says "the learner misunderstands variable scope", even when the underlying data is identical.cf. the compression page:

7. The Risks of Sharing Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

PermissionsPermissions govern who may see a view. They do not govern what the recipient does with it. Four risks arise even when consent is properly obtained.

Context collapse. A view made for one purpose is read in another. A summary written to help a tutor target a session may later be read by a peer or an assistant doing something quite different. The original context, and the qualifications that went with it, are lost. Views should therefore carry their purpose and their date with them, and a recipient should be expected to check that a view still answers the question they are asking.like a snapshot losing the living context

Label hardening. A provisional label, once shared, can become a settled one. "Struggles with proofs" may have been a cautious inference from one assignment. Once it is in a second system, and then a third, its provisional status tends to disappear. This is the mechanism the site describes elsewhere as a temporary inference becoming an identity: the earlier pages on the self-model and on learning logs describe how an incident hardens into a judgement about the personin the self-model page. Sharing accelerates that hardening, because each recipient inherits the label as if it had been checked.

Treating a provisional inference as a verdict. A tutor or peer may read a view as more definite than it is. The remedy is not to avoid sharing, but to write views whose wording makes their status plain: "the system has provisionally inferred", "further evidence is needed", "the learner contests this".

Institutional capture of private reflection. A learner may share a reflective entry with a peer, which the institution then copies into a formal record. The separation between private metacognition and assessable evidence, discussed in the learning logs page, can be lost in one step. A permissions design should make it structurally difficult for a view shared for support to become an assessment record without the learner's explicit decision, and should make the learner aware when it does.

Open learner model research distinguishes forms such as inspectable, cooperative, editable, persuasive, and negotiable models[5]. Sharing adds a further dimension to that list: whether a view can leave the system, and under whose authority.

8. The Sharing Architecture Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

TheThe diagram below shows how these elements fit together. The learner's open model and self-model sit at the centre. Every outbound view passes through a consent and permissions layer, which records scope, recipient, purpose, and expiry. Revocation runs back through the same layer, and evidence and provenance are retained in the model itself, so that a view always points back to its basis.

9. Personal Learning Networks FoundationalKnowledge that endures for decades — core principles

AA personal learning network, or PLN, is the set of people, sources, tools, and communities that a learner deliberately builds and maintains for their own learning. The defining word is deliberately. A PLN is not simply everyone the learner happens to be connected to, and it is not defined by the courses they are enrolled on. It is a structure the learner chooses, extends, and prunes over time.

A PLN should be distinguished from two neighbouring ideas. A personal learning environment, or PLE, is the set of tools and practices a learner uses to manage their own learning: the note-taking app, the reading list, the reflective journal, the way the learner plans and tracks their work. Dabbagh and Kitsantas treat PLEs as a pedagogical approach that can connect formal and informal learning through social media, and link them to self-regulated learning[3]. The network is the people and sources a PLE connects to; the environment is the learner's working arrangement for using them. An institutional network, by contrast, is the set of relationships a course or organisation provides: the cohort, the staff, the assigned readings, the virtual learning platform. Institutional networks are useful, but the learner did not choose them, and they tend to end when the course does.

The idea sits within the connectivist account already treated on this site. Siemens argued that learning is increasingly a matter of forming and navigating connections among specialised sources and people, and that the capacity to know where to find knowledge can matter as much as what one currently knows[1]. Downes described knowledge as distributed across a network of connections, and learning as the growth and modification of those connections[2]. A PLN is the learner's own instance of that network, maintained by the learner rather than assembled for them.Siemens and Downes: the network is the knowledge

10. Maintaining a PLN by Choosing and Pruning Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

AA PLN is not finished when it is built, because its connections decay. A source goes out of date, a person changes their focus, a community stops being useful, a tool is discontinued. Maintenance has two sides. Choosing means deliberately adding connections that cover a gap: a source that disagrees with the learner's current view, a person who works at a different level, a community that asks different questions. Pruning means deliberately ending connections that no longer serve the learning, or that generate noise without insight. A network that is never pruned tends to become a stream of whatever is most recent or most visible, which is a different thing from a network the learner has shaped.

Connectivism's emphasis on maintaining connections, rather than only making them, is the relevant point here[1]. The learning-as-a-network page on this site makes the parallel point that connections require evaluation as well as maintenance, and that the learner must notice when an established connection no longer serves its purpose. Choosing and pruning are the practical forms of that evaluation.cf. connectivism: the network is the knowledge

The learner's developing judgement about their own network is itself a metacognitive activity, which links this page to the self-model page: knowing which of one's sources to rely on, for which tasks, is part of knowing how one learns.

11. Trust for Each Node Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

EveryEvery node in a PLN raises a trust question, and the answer is not a single rating. The site's calculus of trust treats trust as pairwise and contextual: it depends on the trustor, the trustee, the history between them, the task, and the consequences of being wrongin the calculus of trust. For a learner, the questions are concrete. Does this source have relevant expertise for this question? Has it been right before, on comparable problems? Does it show its reasoning, so that the learner can check it? Is it independent of other sources the learner already uses, or does it repeat them? What happens to the learner's work if this node is wrong?

The answers differ for each node and each purpose. A forum thread may be a reasonable prompt for a question and a poor basis for a grade. A peer may be a good reviewer of an argument and a poor judge of a technical specification. An assistant may be useful for generating practice questions and unreliable for checking a derivation. Maintaining a PLN therefore means keeping trust per node and per task, rather than assigning each connection a single standing.

12. Governance Rather Than Platform Ranking Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

ManyMany of the connections a learner has online are shaped by platform ranking: what the recommendation system shows first, which accounts are amplified, which communities are made visible. That ranking reflects the platform's own objectives, which may not match the learner's. A PLN defined by ranking is a network the learner has inherited rather than governed.

Governance means the learner decides which connections exist and how they are used. In practice this involves keeping an explicit list of the sources and people in the network, with the reason each is there; being able to export or move that record rather than leaving it inside one platform; and choosing which connections receive which parts of the learner's model. A learner who can see their own network, and change it, is in a different position from one whose network is a by-product of an algorithm. The learner-governance principle in the Open Learner Model discussion applies here as well: control matters most when it covers the structure of the learner's environment, not only its display[4].

13. Governing the Network Through Permissions Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

ConnectingConnecting PLN governance to the permissions architecture gives a simple rule: membership in a network does not entitle a node to any part of the learner's model. A learner may follow a source for years without that source ever seeing a view. A peer may be a valued collaborator for one project and still not be a recipient of the learner's self-model. Each outbound view is a separate decision, made by the learner, for a stated purpose.

This matters most where the network includes software. Multi-agent systems coordinate through messages, shared resources, and commitments, and the way agents exchange information determines what each can knowin the multi-agent systems page. Commitment-based messaging, in which an exchange is grounded in what was promised and whether it happened rather than in claims about internal states, provides one way to make such exchanges checkablein the contract-BDD messaging page. Applied to a learning network, the same discipline means that a networked assistant should receive a view because a stated task requires it, and the exchange should be recorded so that the learner can see what was passed on and why.

Consent is one basis for sharing, and it should not be treated as the only justification for every flow of learner data. The data protection page on this site sets out the wider question of which lawful basis applies to which processing, and why personal data should not be collected merely because it might be useful laterin the data protection and ethics page. Learner-controlled sharing is an interface to that question, not a substitute for answering it.

14. The Learner as Keeper of the Network FoundationalKnowledge that endures for decades — core principles

AA networked learning assistant can be genuinely useful: it can route a question to the right source, hand a summary to a tutor who needs it, or connect a learner to a peer working on the same problem. Each of those benefits depends on the same condition, that the learner remains the one who decides what is shared, with whom, for what purpose, and for how long. Permissions give that decision a concrete form. A personal learning network gives it a wider one, by making the learner the keeper of the connections through which their learning travels.

The sharing architecture and the network are two views of one commitment: the learner's model and the learner's network should remain the learner's to govern, and any assistant or service that joins them should do so on terms the learner can see, change, and withdraw.

Related Topics Applied / MethodologicalKnowledge with a 5–10 year half-life — stable practice

• The Caring Learning Agent: Open Learner Models, Metacognition and AI Assistance — the open learner model argument that this page extends into sharing and networks.
• Learning as a Network: Connectivism, Knowledge and Conceptual Navigation — the connectivist account of networked learning that the PLN section builds on.
• Pedagogy for Networked and Agentic Learning — the teaching cycle in which choosing, recording, and revising connections takes place.
• The Self-Model as a Learning Tool — the metacognitive model that a learner keeps of their own sources and strategies.
• Personal Learning Maturity — a companion on the learner's developing capacity to manage their own learning.

References

1. G. Siemens, "Connectivism: A Learning Theory for the Digital Age," International Journal of Instructional Technology and Distance Learning, 2(1), 2005.
2. S. Downes, "Learning Networks and Connective Knowledge," IT Forum, 2006. https://philarchive.org/archive/DOWLNAv1
3. N. Dabbagh and A. Kitsantas, "Personal Learning Environments, Social Media, and Self-Regulated Learning: A Natural Formula for Connecting Formal and Informal Learning," The Internet and Higher Education, 15(1), 3–8, 2012.
4. S. Bull and J. Kay, "Open Learner Models," in R. Nkambou, J. Bourdeau and R. Mizoguchi (Eds.), Advances in Intelligent Tutoring Systems, Springer, 2010. https://doi.org/10.1007/978-3-642-14363-2_15
5. R. Bodily, J. Kay, V. Aleven, I. Jivet, D. Davis, F. Xhakaj and K. Verbert, "Open Learner Models and Learning Analytics Dashboards: A Systematic Review," Proceedings of LAK '18, 2018.
6. D. Hardt (Ed.), The OAuth 2.0 Authorization Framework, RFC 6749, IETF, 2012. https://www.rfc-editor.org/info/rfc6749