Connection Holes

5 views
Skip to first unread message

Viplav Baxi

unread,
Sep 12, 2010, 10:24:40 PM9/12/10
to Learning Analytics
I know I am jumping the gun a bit, but I wanted to see if this idea
resonated with anyone on this group.

So essentially, AI based learning solutions have an algorithm for
recommendations to the learner based on her current state of
knowledge, contrasting it with an ideal state of what her knowledge
should be). So if she goes through a topic and fails an assessment,
the system detects what caused her to fail (i.e. the "knowledge" hole)
and offers one or more paths to remediation.

Could we think in terms of connection holes? Can we think of an
environment where very simplistically, there are a set of connections
which, if not made, could "cause" the learner to "fail"? These
connections could be between ideas or to people etc. Before we get our
connectivist knives out ( :) ), I don't want to sound deterministic or
even state that an "ideal" set of connections could exist or that
patterns could be engineered, but could we or should we perhaps start
with a constrained environment and gradually build in these
complexities?

Why? One of the main questions in every MOOC I have participated in is
"How connected were you?", sort of implying that if you were
disconnected (from ideas or people), you could have "learnt" better
through "improving" the state of connectedness.

If we had a way to design or uncover connection holes, we could make
that as part of our data design and capture in a learner analytics
system. For example, a really dynamic way would be to take the
community/learners generated mindmap of a topic and ask a learner,
starting with the level 1 expansion of the mind-map, to identify the
1st level branches. If the learner is connected, engaged and open to
the learning experience, she will be able to point out the first level
branches. And then we check the depth of engagement by expanding to
the level 2 branches and so on...?

What do you think?

Phillip Long

unread,
Sep 13, 2010, 11:34:20 PM9/13/10
to learning...@googlegroups.com
Viplav: That is essentially what underlies the prompted hint system behind the now Pearson one "Mastering Physics" application (formerly CyberTutor), developed by Dave Pritchard at MIT. This is a very challenging area even when you 'know' the core concepts that are required to understand something (whatever "understand" means). By focusing on first year mechanics and electromagnetism Dave defined his domain very carefully. Even so, basic concepts he identified often required 17 or more intermediate ideas that wove together to fully grasp a slightly higher order concept.  Even there, it took years  and hundreds of thousands of student interactions before they started getting reliably useful hints to help students identify what they were missing. I guess all I'm saying is this is not for the faint hearted.

Dave remains very interested in this area and I'll ping him to see if this community might be of interest to him. It would be good to get someone with his experience into the discussion. However, my guess is he'll think we haven't focused our problem space sufficiently.  From his perspective first year mechanics in physics was too broad.

Regards,
Phil

Professor Phillip D. Long, Director,               |    Visiting Researcher
Centre for Educational Innovation and Technology   |    Center for Educational Computing Initiatives
Level 4, Building 78,                              |    Massachusetts Institute of Technology
The University of Queensland,                      &    Bldg 9 3-355
St Lucia, Brisbane,                                |    77 Massachusetts Avenue
Queensland 4072, Australia                         |    Cambridge, MA, 02139 USA
Phone: +61 07 3346 6283                            |    Phone: 617-452-4038
Email: lon...@uq.edu.au                            |    Email: lon...@mit.edu

Professional Blog

NMC Board Member: the New Media Consortium, dedicated to the exploration and use of new media and new technologies: http://www.nmc.org

Viplav Baxi

unread,
Sep 14, 2010, 5:43:08 AM9/14/10
to Learning Analytics
Thanks for your response, Phil. Would be very interested in what you
and Dave think about moving from hinting about concepts to hinting
about connections. If these connections are to people, then perhaps
people will help connect to concepts and ideas, is perhaps what I am
saying. It may end up providing measures for networks as a whole as
well - how reliably does the set of connections in a network complete
the picture on a topic, for example, could be an analytical measure of
quality of the learning environment.

Regards,
Viplav

On Sep 14, 8:34 am, Phillip Long <long.phil...@gmail.com> wrote:
> Viplav: That is essentially what underlies the prompted hint system behind the now Pearson one "Mastering Physics" application (formerly CyberTutor), developed by Dave Pritchard at MIT. This is a very challenging area even when you 'know' the core concepts that are required to understand something (whatever "understand" means). By focusing on first year mechanics and electromagnetism Dave defined his domain very carefully. Even so, basic concepts he identified often required 17 or more intermediate ideas that wove together to fully grasp a slightly higher order concept.  Even there, it took years  and hundreds of thousands of student interactions before they started getting reliably useful hints to help students identify what they were missing. I guess all I'm saying is this is not for the faint hearted.
>
> Dave remains very interested in this area and I'll ping him to see if this community might be of interest to him. It would be good to get someone with his experience into the discussion. However, my guess is he'll think we haven't focused our problem space sufficiently.  From his perspective first year mechanics in physics was too broad.
>
> Regards,
> Phil
>
> Professor Phillip D. Long, Director,               |    Visiting Researcher
> Centre for Educational Innovation and Technology   |    Center for Educational Computing Initiatives
> Level 4, Building 78,                              |    Massachusetts Institute of Technology
> The University of Queensland,                      &    Bldg 9 3-355
> St Lucia, Brisbane,                                |    77 Massachusetts Avenue
> Queensland 4072, Australia                         |    Cambridge, MA, 02139 USA
> Phone:+61 07 3346 6283begin_of_the_skype_highlighting              +61 07 3346 6283      end_of_the_skype_highlighting                           |    Phone:617-452-4038begin_of_the_skype_highlighting              617-452-4038      end_of_the_skype_highlighting
> Email: lon...@uq.edu.au                            |    Email: lon...@mit.edu
>
> Personal Blog:  http://web.me.com/longpd/Through_the_Looking_Glass/Home/Home.html
> Professional Bloghttp://edtechtrends.blogspot.comhttp://longpd.wordpress.com
> > What do you think?- Hide quoted text -
>
> - Show quoted text -

da.luisa

unread,
Sep 16, 2010, 1:56:21 PM9/16/10
to Learning Analytics
I agree with you, Viplav.

In particular I enhance a learning path developed in gradual phases
of autonomy and complexity levels.

And in my opinion, the efficiency of a network of connections (of
people, knowledge, ideas, competences, technologies, and so on) cannot
be only the result of the personal initiative of every individual
student inside a group learning. Without communication and connection
order, the learning chaos becomes a possible result. How it is
known, a group of people, who are in the same place and interact
between themselves, does not coincide necessarily with a team, because
the people must collaborate towards the same goals, with specific
interdependent roles, with clear behavior rules and a adopted specific
functional communication structure.

A team learning need a support to realize a network without holes of
efficacy, and this support should be at a meta-level.

In my opinion, in an online learning, the catalytic and strategic
action of AI based tutoring (to complete the "human" tutoring) is
able to assume this role.

Actually I am concentrating my researches exactly on the knowledge
base and cognitive rules-based tutoring, and every other significant
study is the welcome and a great stimulus.

Regards Luisa dall’Acqua

http://it.linkedin.com/in/ldall

Viplav Baxi

unread,
Sep 16, 2010, 9:24:13 PM9/16/10
to Learning Analytics
Thanks, Luisa. You make an important point about groups and teams.
Stephen has differentiated between networks and groups and George
between Connectives and Collectives. There are many different theories
of how people organize together in different contexts. For example,
Wenger talks about five stages in the evolution of a community of
practice. I also think that Learner Analytics from the network point
of view will be an important contributor to AI based tutors and so is
important for you to elaborate on this forum.

Regards,
Viplav
Reply all
Reply to author
Forward
0 new messages