Thursday, 10 February 2011

Networks and Feedback

Over the last two or three weeks we have had a lot of discussion about networks and how Connectivism is really focused on how learning takes place across these networks. One thing that hasn't really been discussed is the notion of feedback loops in relation to networks and I believe that understanding feedback mechanisms are crucial to understanding how networks develop and evolve. So what is a feedback loop? Feedback loops occur when information or an event in a system makes a change to the system and that information or change is then fed back into the system again allowing the system to respond to the information or event.

There are two types of feedback mechanism: positive and negative. Negative feedback in a dynamic system  is a constraining mechanism, keeping the system within certain boundaries  - it is a limiting function that helps to maintain system equilibrium - think of a thermostat - as the temperature reaches a certain temperature the thermostat cuts the heat source so that the temperature cannot continue to rise, if the temperature falls below a certain level the thermostat again connects the heat source to allow the temperature to rise. In social terms negative feedback can be seen in the use of rules, laws, social norms, etc..Positive feedback reflects and amplifies some aspect of the system and is often exponential in the sense that the amplification can happen very quickly over a relatively short period of time. Most people have probably experienced positive audio feedback, whereby a microphone, too close to speakers creates a situation where the sound signal very quickly gets louder and louder  - the microphone picks up the sound coming out of the speakers which is then fed back through the speakers, picked up by the microphone and so on until painful on the ears! We can see a current example of positive feedback in a social system context in Egypt, where after years of some degree of relative dynamic balance, the pro-democracy movement has been very quickly amplified, threatening the existing system's stability. Positive feedback loops can allow quite dramatic change to occur within a system.

In complex systems, both negative and positive feedback mechanisms are necessary for a dynamic balance of stability and instability. Without positive feedback, negative feedback would prevent any development in the system and if positive feedback went unchecked it could cause the system to collapse. It seems to me that networks are the delivery routes for feedback. We have talked a lot about the connections in networks (nodes, weak / strong links, etc) but not so much about what is being passed through these connections. As well as data, information and ideas networks are the conduits for feedback mechanisms. Thinking in terms of the CCK11 network for example, there are the readings and resources which are available across the Internet for the course each week but the network also includes a high interactive element of discussion, blog posts and tweets. Some ideas get amplified through comments, retweets, etc and become important. The CCK11 network is always dynamic - every time one of the participant's understanding changes the network changes.The CCK11 network is interconnected to multiple other networks so when that network changes the connection to the other networks change - I guess a kind of "network ecology".

There are many kinds of learning - take for example a child learning to walk. The process of learning to walk in obviously not read in a book and has nothing to do with the web or social media - but it is a feedback mechanism, feedback between the child's body and brain (neural networks), an instinctive, iterative process over time amplifying certain movements and restricting others until the delicate balance of walking is achieved. In this example no one sees the feedback itself- only the result of it taking place.

In the Elluminate meeting last Friday, there was some discussion around learning and being connected / disconnected and what that might mean. I had the thought that if you look at networks from the perspective of feedback mechanisms - events that cause change in the nodes - then perhaps the degree of connectivity is connected to the degree of available feedback mechanisms, both positive and negative. On a very basic level, if I sit in my room working on trying to understand something but not sharing my thoughts then any feedback taking place is only within the frame of my own neural connections - I will be connecting and referencing previous knowledge and understanding. I may think I really understand something until I share that understanding with someone else and then something they say might change (or reinforce) that understanding - often because they have made a different set of connections themselves. By externalising my understanding, even on a one to one level I have expanded the potential for feedback and assimilation.

When trying to follow the group discussion going on in the live Elluminate sessions I am constantly trying to adjust my perspective - how does idea X fit into my current understanding and if it doesn't fit then why not? What I realise is that "my current understanding" is the emergent result of all the connections I have previously made - not a static, linear structure built on a pre-determined path but an organic ecosystem of understanding that is always in dynamic balance, subject to forces of positive and negative feedback.



  1. Very interesting thought, I quite agree with what you explain. Congratulations on your writing. Greetings

  2. Thanks Mercè - greetings too from Scotland!

  3. Hi Graeme,
    I read a bit this week about Actor Network Theory, and there's a concept of "translation" there, which I think means transporting some(idea/thing) through the network, but as a result of that effort also *changing* the (idea/thing). I'm kind of curious about the distinction you make here between positive and negative feedback--could something be both? Or both at different times as it works through the network and into other networks?

    Thanks for the thoughtful post!

  4. Thumbs up for this post. I agree, without a feedback loop that allows you to course-correct when you need to, all you end up with is a giant echo chamber :)

  5. Thanks for the comments!

    Leah - I've been reading up about ANT too but as far as I can understand (so far) I think the idea of translation is something different to that of feedback. The whole notion of feeback "loops" is that the feeback actually changes the system itself, affecting future responses. If a system is always dynamic then it is quite possible that a whatever is limiting that system in one scenario may cause it to change dramatically in a different scenario. Also many systems are usually inter-related so negative feedback in one system may have a different effect in a connected system. This video gives some helpful examples of feedback mechanisms in the human body:

    I see that there is now a CCK11 discussion thread on this topic so I hope to have a look at that:


  6. I think you are right that it is a combination of positive and negative feedback processes which drive the components of signal in a network to specific values (representing "true" knowledge?).

    But for me there is still a question of what is the signal. Is it just the strengths and numbers of the connections? And if so, how do these actually encode knowledge? and how are the strengths measured? and what kinds of process can cause the increase or decrease of strength of a connection and/or create and destroy them?
    There may be some emerging suggestions in the neural context, but for the connectivist educational theory context I haven't seen anything specific.

  7. Hi Alan - thanks for your comment and really important question, which I have been pondering for the last couple of days. I still don't have a clear answer but I am working on an intepretation of sorts. For me the essential point about this feedback debate is that the model of networks used to illustrate the ideas of Connectivism is essentially a linear model - the notion of back-propagation is not feedback in the sense that it is used in complex adaptive systems. It appears to me (and I have to emphasise that I am still trying to interpret this question)that the type of network model which is used to describe Connectivism, based on back-propagation, does not take account of the possibility of the nodes themselves actually changing - if you view humans as nodes in a network then any model needs to take account of the fact that humans are non-linear. You don't know what response you are going to get from a human in a network - for example when you posted a comment on my blog there wasn't any way you could predict how I would respond to that comment - I might have ignored it or responded to it. My issue with the feed-forward model of a network is that it is designed to predict outcomes - i.e it is based on a linear model of interaction, so its all about inputs and outputs.

    The idea of complex-adaptive systems is that prediction using linear methods is inadequate because the "signal" that is passing bewteen the agents (be they humans, neurons, etc)can change the agent itself. The linear model sees an agent as static, it has the same (or similar) response to the same input. Feedback, especially postive feedback, can change the dynamic relationship of the agent to the rest of the system and so the dynamic of the system itself, and that change can be quite dramatic. The feed-forward model of networks does not seem to accomodate such change - back-propagation appears to be a kind of adjustment to maximise the predicted output.

    Next week we are studying complex adaptive systems - currently I can't see the compatibility of the feed-forward model of networks with the fundamental idea of complex adaptive systems. Hopefully next week I will get the chance to ask Stephen and George about this.

    Your question about what is the signal I think is really interesting and I don't have an answer, only sugestions. I do think it has to depend on the frame of reference of the question, regardless of which specific network is being considered - information, ideas, data, electrical impulses, emotions?

    Best wishes