I have transcribed a section of Alan Kay's recent presentation marking "40 years anniversary of the dynabook" because I see it as an important contribution to ongoing discussions about the significance and prospects for the XO, or OLPC
Alan Kay visited Seymour Papert in 1968
49 minutes: Slide that involves Alan Kay
About 40 years ago I went to visit Seymour Papert... because I had started to visit people who could be users of a desktop computer ... and Seymour was working with children ... Seymour was a mathematician who had worked with Piaget ... (talks about Seymour's tragic accident from which he has not recovered) ...
Kay's interpretation of Seymour's work: Children are egocentric in a charming way ...they do everything relative to them ... so if they were a co-ordinate system they would be an inertial co-ordinate system ... and an inertial co-ordinate system is the differential geometry of Gauss ... and if you keep track of this in the right way you get the differential geometry of vectors ... and a child is one of those vectors... and so is the turtle ... and he thought, boy(!), this is so close to the way children think already I wonder what would happen if we put some formalism on it and treat it as mathematics
So ask a little kid to draw a circle with their body and ask them what they are doing ... they say they are going a little and then turning a little, over and over
In logo: repeat 360 [forward 1 right 1]
Tell the turtle to do that ... and by golly you get a circle and you can put in different values to get circles of different sizes ... and so we have a differential equation here which is infinitely simpler than (traditional) differential geometry and which can be understood by a young child
This completely blew my mind! Once you've got something which is incredibly powerful that a child can learn you've no longer got an adult tool ... you've actually got something like a printing press that is one of the great 500 year inventions in human history
If children can learn these powerful ideas then you have a chance to not just increment on what is already known ... they will actually help over several generations to invent something new
So, that combined with just seeing this flat panel display with all these wise words of McLuhans in my mind (mentioned earlier in talk) on the plane back to Utah after meeting Seymour I drew this little cartoon with kids out in the grass ...
... because if you are going to make a personal computer for kids don't put it on a desk ... that isn't them ... so I immediately took the fun idea of a flat screen computer and made it paramount ... you had to make a computer that was in every way made for a child, that they could take away from adults and learn by themselves ... it would have to have wireless, a stylus, a touch screen, a keyboard (because even perfect character recognition is not fast enough to do bulk typing), removable memory ... and so all of those things coalesced ....(describes how he made the model from a cardboard box)
Alan Kay's definition of portability ... that you can carry something else as well as the portable device ... arrived at the figure of two pounds (which is roughly 1kg)
(some information about head mounted display and a wrist detector by Negroponte left out here)
If you think about this as a service idea then what are the actual services ... (stuff snipped) ...
"IT'S NOT THE TECHNOLOGY, STUPID"
You really have to have some idea of the end users ... leads into new slide
DYNABOOK PLAN
It's a service idea with serious goals about education, especially self education
Kept (?) standards: Fluency in powerful ideas for > 90% of children
powerful ideas and how to learn them ...
CONTENT
... and with the aid of computing media
Human mentors
MENTORING
Computer mentors
[[Alan refers to these, here and later, as "four ideas", which I understand to be:
- What are powerful ideas?
- How can they be learned with the aid of computing media?
- Can this work with human mentors?
- Can it work with computing mentors?]]
I was interested in whether we could make computer mentors because my confidence in adults was very low back then ... and still is. The biggest bottleneck to education reform is the adults that are in the system
In the Third World it is the lack of adults ... But in our world it's almost the lack of adults ... almost no elementary school teachers understand anything about maths and science ... in a way things might be almost better if they weren't there because the children would not be getting misinformation about it
It has to be setup to succeed for 90% of the children, not just the 10% who are naturally good at it
The problem with technologists doing it is that we are all setup by nature to be good at it ... all of us here learnt to program within a weeek ... I'll bet you anything ... it's not that hard if you almost know what it is ... but if you don't almost know what it is, it can be really daunting
This is why computing people generally design terrible computing user interfaces ... they're not only willing to cope with something bad, they are pleased to ... because it's a little challenge for them ....
(this section finished at 1 hr 1 min)
At 1hr 15min the moderator asks:
"Why was the dynabook never built in spite of all these people trying to make it happen?"Alan recapitulates the four ideas outlined above ... then ...
Working on the first two with our 90% success threshold ... led to 25 years of failure ... we were paying for this research ourselves ... nobody would fund the children's research because we did these long projects ... we didn't believe in most forms of testing that are reported in the education literature ... so we wanted to convince ourselves that the children were getting fluent and that 90% of them were getting fluent ... by those sorts of criteria it was one failure after another .... but after each failure we would learn something
And about 10 years ago one of the systems that we did started teaching many more children with adult help in a much stronger way (I think he means etoys) ... so I think after 40 years the first two ideas and a little bit of what sort of human mentoring you need have been solved ...
BUT, when Nicholas started up the OLPC project my heart sank, even as I supported it ... because if it's tough to get good mentors in the USA then it's really tough out in the Third World ... no user interace today can find out who its user is, what its user knows, what it can do ... it can't find out what level of reading the user can do and help find out the next level of reading
There is common sense in the world concept ... so we make a world populated with objects ... but they didn't interfere with the user strongly
That isn't enough ... pure discovery learning took us 100,000 years to get to science ... so you need learning that is facilitated ... and if you can't make thousands of good teachers in a year then you have to have an interactive user interface to save yourself
This dream of having a UI to facilitate is as old as AI ... it is AI ... if we had this we could make up for no teachers and bad teachers (but we still need good teachers) ... so when the OLPC project started I thought OMG, we are lacking the one piece of the technology ... if we could just ship that machine with a program that could teach children to read in their native language ... that would be the killer app and we wouldn't have to worry about anything else for a number of years ... but that technology doesn't exist ... it is that gap which has to be bridged in order to fulfil the educational goals that the dynabook has ... you have to have a way to get around the adults in the system that make educational reform difficult
4 comments:
This answers a question I had about Alan Kay's concept of the user interface sensing where the user was in terms of knowledge, and then communicating knowledge that was encoded within the system in a way that the user could understand. He communicated this idea to me this way in correspondence we had more than a year ago:
"The other thing to think about is that most ideas are mediocre down to really bad! (This is because good and really good ideas are quite novel and low probability even when geniuses are trying to think them up.) So a really important feature of any communications system is not so much to capture an idea, but to help other people understand it. ... This is a much bigger problem than just facilitating communication and remembering in various ways."
In another piece of correspondence he had with both you and me around the same time he said:
"One of McCarthy's first papers in computing (ca. 1958) was 'Computers With Common Sense' in which he proposed that the best way to deal with humans and to scale what could be done with computers was to make a model of human common sense and embody it in an agent that could be given goals and advice by the end-user (the "Advice Taker"). A really good one of these would be like a (myriad of) grad student(s) who could be summoned, commanded, and who would then carry out tasks, write programs, etc.
We still don't know how to do that today, so I'm interested in what the next real steps in programming might be without a real human-level AI to do the scut work. In our NSF project - http://www.vpri.org/pdf/NSF_prop_RN-2006-002.pdf - we propose to take "personal computing" and make the most compact and understandable and programmable working model that we are able to do. We set some targets - 20,000 lines of code - that will prevent today's programming styles from being used as the main solutions, in order to force some needed new insights and inventions. ... If doable, this would constitute a 2 to 3 orders of magnitude (a "Moore's Law") improvement in programming expression -- and would certainly provide much more illumination into the mysteries of programming."
So it sounds like he sees programming as a "band aid" until real AI surfaces.
hi Mark,
At 21-22 minutes of the video Alan Kay refers to John McCarthy as anticipating computers as interactive machines - interacting with "advice takers" who would have to deal with computers in terms of their common sense, that McCarthy saw computers as AI entities
He then goes onto lament the fact that LISP, although 50 years old, is still better than many other programming languages today
There is so much ground covered in this presentation and some of it is just snippets that are easy to miss and would warrant much detailed study - because it's all interconnected with many cross references
check out:
http://commonsense.media.mit.edu/
@Bill:
An interesting anecdote related to this, is that according to Doug Engelbart, Licklider was actually one of the people who ended up standing in the way of his progress for this very reason. In an interview Engelbart did with Robert Cringely at http://www.pbs.org/cringely/nerdtv/shows/ (see show #11) he said that he had received funding to proceed with developing his NLS system further, after the 1968 demo. In the 1970s he met with Licklider and described how he was using human trainers to educate people on how to use the system, and Licklider froze. He asked something like, "Why aren't you developing an artificial intelligence to do this?" Engelbart said something like, "That's something that hasn't been developed sufficiently yet." He wanted to use people as trainers rather than the computer, to compensate for this. Licklider opposed this vigorously, and actually denied Engelbart further funding, at least from the sources he had influence over. Engelbart chalked this up to the AI researchers having such a hold on the imagination of the computer science community at the time. They had grandiose visions of what was possible. He saw those visions for what they were. In the long run AI ended up falling far short of them. It's a sad tale. I got this sense that as a result of this infighting, commercial computing was able to take over the sense that most people had about where real computing was at. "The theoreticians can fight over phantoms," so the thinking probably went, "We'll focus on what can really be accomplished."
Post a Comment