I am a bit late in the discussion but it is truly interesting and I
would like to bring some points back as some of those issues might help
define both the scope and the roadmap of the http://tutorius.org/ project I am
working on and hope will become eventually part of Sugar.
Alan talked about two approaches to "mentoring user interfaces", but I
am mostly interested in the first one, as I do not grasp the
intricacies of the second one (AI). My view on the subject is that the
key to make it happen would be to reach the point where it would be
easier for a teacher (or a student) to create and use learning material
using a knowledge model and a mentoring UI than it would be to do it
with other tools (textbook, etc.). I would aim for a development model
that would allow content developers to build on top of what exists and
that would allow learners to provide feedback on the material so that
eventually the best material would stand out and would serve as a
foundation to continue.
The first insight we had with Tutorius is that a lot of the learning
material for computer software right now comes as tutorials, as a
sequence of steps (1,2,3) to accomplish a task or find a feature. Most
of the tutorials today exist as static documentation involving
screenshots, explanations and steps. Video tutorials are a little bit
more expressive but not qualitatively different since the user is still
in passive mode. To produce a qualitatively different experience then
those tutorials must become interactive. The idea is not new and have
been used in a great fashion by commercial software like Sketchup but
also, and I think it could be a great inspiration, by video games,
where often the whole gameplay mechanic is progressively introduced
through missions involving basic skills and moving up to more and more
sophisticated techniques.
The second insight we had is that the mentoring technologies to teach
high level concepts might be based on the same technologies that could
be used to make the current documentation of Sugar interactive, so it
would give a way to make it happen.
The high level roadmap I would suggest to end up with a mentoring
system would be:
1. Instrument Sugar to allow interactive tutorials in any activity or
the Sugar shell (with a minimum of changes to them) and that could span
more than one of them (Let's say start in the browser and guide the
user to copy-and-paste material from a web page to a Write activity).
This should allow us to create tutorials that would cover the whole
current Floss manual.
2. Build a tutorial creator that would allow the creation of such
tutorials from the activities.
3. Build a website to share tutorials and rank them.
4. Establish a knowledge model that would describe skills, their
dependencies and the evaluation functions that might rely on monitoring
the usage of Sugar to establish the mastership level of a given skill.
5. Link tutorials to the knowledge model to allow suggesting what to do
next, based on mastership level.
6. Allow the collective evolution of that knowledge model by users.
7. Build more sophisticated monitor agents that could evaluate user
skills outside a tutorial, in free usage of activities (think of a
touchtype monitor that would monitor "delete" key press to build a
dictionary of characters and words which exhibit a high number of
errors to add them to the daily routine exercises).
8. Add whatever it takes to teach "Powerful Ideas" (At this point, I
don't know what we would need).
We are in the process of doing 1,2 and 3 for December 2009 (we hope)
and the rest will depend on the initial feedback we get. I don't know
what would be needed in a knowledge model and I don't know how much of
what has been done in AI research has solved this already. We are
currently focusing on the UI and the tutorial representation.
A partial answer to the motivation problem Alan talked about in
mastering a skill like reading would be in my opinion to provide
constant feedback on the progress of a learner in pursuit of a goal.
Such feedback seems to be the key behind the success of a system like
Nike+ and the addicting effect of video games. I think it could be
replicated for a learning environment by showing the mastership level
of different skills needed to achieve a goal and their evolution in
time.
I would be interested in knowing usage scenarios Alan would have for a
"mentoring UI" that would show what goals would be the most interesting
and worthwhile to go after.
All in all, a very interesting discussion!
Erick
Alan Kay wrote:
Well, "The ARPA Dream" of the 60s as articulated by Licklider,
... that "The destiny of computing is to become interactive
intellectual amplifiers for all humans pervasively networked worldwide"
was too big for any one group, but Lick funded about 17 of these in the
60s to try various goals derived from this dream, and with the aid of
some of these grad students who went on to Xerox PARC, this research
community was able to do the critical inventions required. Virtually
all of these researchers 30 years later speak of the pleasure of being
involved with this community much more than singling out the particular
places (such as MIT, CMU, or PARC, etc.). This is how really big "Grand
Challenges" get done.
As I mentioned at the end of the email, some very important parts of
the puzzle can be advanced by spending a lot of money in the right way.
There are 9-10 months in a school year (30 to 40 weeks given vacations,
and 5 hours a week or a little less for mainstream subjects) so "a year
of math" in school terms is about 150 to 200 hours of "instruction".
Take the larger number and think about 5 mainstream courses per year,
and this gives about 1000 hours per year (usually quite a bit less). So
K-12 would be about 13,000 hours of live instruction. Suppose we brute
forced all of it (at $1 million dollars per hour), then K-12 done this
way would be about $13 billion dollars. This sounds like a lot, but
Iraq is costing about $10B per month(!), a Trillion dollars of the bank
bail out is about 8 times this, 3 B2 bombers would cover it.
And, if you were to amortize this over the world's children, then the
cost per child would not be large. For example, there are about 60
million US children in school. So one year's worth of all grades for
all children would cost about $220 per child. And this material would
be used for many more than one year, and for many more than just US
children.
(Someone can help by checking my approximate arithmetic)
The point here is that even the brute force approach to augmenting or
even completely filling in for teachers is much more of a societal
priority issue than a cost issue (as indeed are children's computers
like the XO).
However, because no good deed goes unpunished, we can imagine real
disasters here if "large unconstrained interests" were to control this
(similar stupidities and tyrannies that we've seen in schooling and
school books).
What I've been advocating to potential funders is that there are many
parts to this problem, just as there were for inventing personal
computing and pervasive networks. And some of the parts can be advanced
quite a bit right now by spending money in the right way. ARPA was
happy to furnish single researchers with a huge mainframe in the 60s
for experiments in user interface, graphics, etc., because Moore's Law
was thought to be something that would happen, and thus it was worth
spending millions to "live in the future ahead of time". This was a
critical part of the way the ARPA community thought and acted.
So we don't have to pony up $13B up front here, but $20M-$50M would
start up a community of researchers who are all interested in working
parts of this problem, and as I suggested in the last email, one thing
that should be done is to brute force at least one half year of one
comprehensive course and just do all things necessary to make the
experience work. This produces an artifact that will help smarter
approaches aim at high enough level targets. An analogy here is that a
great book can be used by good teachers to really enrich a good school
experience, and a great book can be used by learners who are not in a
good schooling experience to nonetheless help them internalize much
stronger outlooks and ideas despite an impoverished environment.
There are very good arguments that learning to read and write fluently
is not just taking on a communications skill (which might imaginably be
replaced by some other communications technology) but that "becoming a
fluent reader and writer" changes and enriches one's thinking styles
and powers in qualitatively different and important ways. I subscribe
to these arguments (and the studies which back them up).
So, the notion of trying to make an environment for helping children
learn to read and write -- a "book" that can help its readers learn to
"read" and "write" it -- has an enormous appeal, especially given how
much really worthwhile stuff is to be found on the web (almost hidden
by the pop culture trash, but there).
One whole route that was worked out almost 50 years ago was by O.K.
Moore at Yale in the late 50s and early 60s -- it was called the
"talking typewriter" (which was a hidden grad student) -- has many of
the initial seeds for thinking about how to do this, and has quite a
few studies giving quite a bit of date about how children behave in the
environment that Moore set up. "Writing to Read", done by one of
Moore's disciples and sponsored by IBM many years ago, was more real,
but also omitted some of the key principles that Moore had discovered.
This program, among other things, needed a small personal machine like
the XO to create the "autotelic environment" which Moore felt was so
important.
Several open questions make it difficult to be completely confident
with a sponsor. One is that just how well this approach works over a
number of years (if real fluency is the goal) is quite unknown. And,
there is also the issue -- also unknown -- about whether a great
curriculum that *could* do the job would actually be followed long
enough by learners. This is analogous to having a great method for
learning a musical instrument with online assist (and there are some
pretty good ones out there). But one of the biggest problems is "Sire,
there is no Royal Road to Geometry" (as Euclid said to Pharoah). That
is, in the end, you just have to "do a lot of doing" to learn most
things, especially the harder ones that were rare inventions of
humankind. This is not popular these days -- unless it is a social
token which must be learned to be part of a pop culture - alas, it is
hard to find any deep content in these subjects.
Bottom line, it's difficult and it's doable.
Cheers,
Alan
From:
Martin Langhoff <[email protected]>
To: Alan Kay
<[email protected]>
Cc: Tomeu Vizoso
<[email protected]>; K. K. Subramaniam
<[email protected]>; [email protected]
Sent: Thursday, July
2, 2009 7:52:17 AM
Subject: Re: [IAEP]
changes in outlook with Sugar (was Re: Comments on David Kokorowski,
David Pritchard and "Mastering" Educational SW)
On Thu, Jul 2, 2009 at 4:35 PM, Alan Kay< [email protected]>
wrote:
> We can see this with the XO and OLPC also. Much is made about
"getting
Hi Alan,
your answer is very humbling, and talks about solving a huge problem.
One that OLPC, Sugar and related projects are just one part of a wider
solution.
In practical terms, however, it is hard to hear that "the problem is
too big for us". I would hope for a vision of what part Sugar and OLPC
can play here, as part of the bigger thing.
You said:
> So my "vision" here is let's try to find supertalents in this area
wherever
> in the world and try to fund them.
Let's assume you managed to recruit some of those (humble)
supertalents. They are rowing hard, and have their own ideas of what
to do next, but would also like to hear your opinion on mid and long
term goals and visions on the track we are working on.
Have you seen recent versions of Sugar and/or the OLPC laptop +
software?
cheers,
martin
--
[email protected]
[email protected] -- School Server
Architect
- ask interesting questions
- don't get distracted with shiny stuff - working code first
- http://wiki.laptop.org/go/User:Martinlanghoff
_______________________________________________
IAEP -- It's An Education Project (not a laptop project!)
[email protected]
http://lists.sugarlabs.org/listinfo/iaep
|
