These are my live blogged notes from the opening session at this year's DevLearn, hosted by the eLearning Guild and happening in Las Vegas. Forgive any typos and incoherencies.
Donald Clark @donaldclark “Learning Prediction for 2015 and
Beyond: Two Small Letters”
Algorithms and predictive analytics – Netflix knows what you
want to watch. It’s time to do the same for learning.
Learning styles don’t exist. Don’t build your adaptive
learning system based on them. They have no scientific validity. Same with
Myers Brigg.
Aggregate data let the machine algorithms kick in so the
data gets better and better.
Learning to Adapt – Adaptive Learning – this is what the
Gates Foundation has pumped a lot of
money into this area.
Cogbooks.com
Create a network of learning objects – adaptive sets you
free from A to Z. It understand you and finds the best way through. Adaptive
learning gets you back on course – it spots your misconceptions. It knows
you’ve failed and tries to identify why you’ve failed.
In the real world using Google – we’re in charge. Unlike in
eLearning.
University of Edinburg is piloting this for courses and now
rolling out to even more.
ASU is doing massive adaptive learning courses with American
History and Biology -- 101 courses.
Good adaptive platforms that can understand the
structure and network of American
History objects – that’s good.
Ask questions, ask your confidence level as you answer those
questions – that helps with predictive.
Hitachi Data Systems – let’s scrap courses and instead have
one big knowledge base with lots of competencies. The future for this company
isn’t the traditional L&D course – instead it’s competencies that every
learner vectors through in different ways. The algorithms keep an eye on what’s
next.
Look at duolingo.com – highly algorithmic.
Could algorithms create elearning from scratch?
Roger Schank’s PhD systems went
Journalism: algorithms more powerful than news editors. The algorithms
are smarter than the journalists – they have data back for 100s of years.
WiQi -- -- cloud based system – you type in some
content. It creates some eLearning with
a semantic engine. All open response questions – not MCQs which are just about
picking from a list.
(affinitystudios.co.uk). You can put your compliance documents into this
– This works right now and no cost. It
pulls up copyright free images from Wikipedia – you can do it with any text
document you want.
(Check Donald's slides for a link to WiQi)
Automated Essay
Marking.
Spaced Practice
Ebbinghaus – we don’t learn a damn thing without repeated
practice. Every student has a mobile phone – you have an umbilical cord
for spaced practice.
With spaced practice – do lots of repetition in the first
few weeks and then you can scale that down.
Interleave: things we get right are called knowns; those we get wrong
are unknowns. But we forget everything, so we have to reinforce both the knowns
and unknowns and the half knowns. If we
want automaticity – we have to interleave the knowledge going forward.
Hopping: hoping between levels from easy, medium, hard
Cognitive spread – items spread on next few days to smooth
out workload…if you miss a day of work, then it spreads that out over the next
few days, it doesn’t just shove your mixed work into the next day.
9 things algorithms
do that teachers can't:
- ignore gender, race, social background
- free from cognitive biases
- never get tired, ill…
- do things that brains cannot
- personalize learning
- personal reporting
- prevent failure and drop-out
- automatically improve courses
- scale
AI Robots
Bloom wrote a paper showing that 1-1 tutorial provided
massive results – it’s incredibly powerful. What if we can use that fact to
embody that with a robot…
Introducing NAO – working with autistic children using
robots.
Teachers find it really hard to do formative assessment -- computers do formative really well.
