MLTalks | Inventive Minds: Marvin Minsky on Education

Hal Abelson: Hi everybody. Thank you for coming. Today, we’re going to hear about a real treasure
of a book and one that’s especially appropriate to be hearing about now at MIT. We’re in the midst of an incredible explosion
of interest in artificial intelligence and computing. Hal Abelson: Yesterday in fact there was a
meeting of the 123 MIT faculty committees that are trying to figure out what this new
thing called the College of Computing. That explosion is happening not only in MIT
but all around the world. Part of that is a call for learning of about
AI and education including AI education for young kids. Hal Abelson: This book is a collection of
essays by Marvin, who was the greatest thinker in AI. I don’t know what he would have made of today’s
explosion but starting a couple of years ago, he said to me he thought everyone he thought
we should all start working on BI and be sure not to tell anyone what it stood for. Hal Abelson: In any case, here is Marvin telling
us his thoughts about education, which is the rage now, whether it’s deep or whether
it’s hype, you might want to consider that. Marvin’s thoughts in these essays as always
with his thoughts are original and controversial and rather different from the current perspective
of what people are pursuing and what’s called AI or computer science or computational thinking. We’ll get into that in a bit. Hal Abelson: First, let me introduce our author,
Cynthia Solomon, was one of the four creators of Logo. Logo was the genesis of computing as an activity
for kids. She was a key collaborator of Seymour Papert
all through the ’80s and the MIT Logo project. After that, she was director of the Atari
Cambridge Research Laboratory. She since then has an active history both
of research and education for kids talking about it, writing about it and just being
the guru on educational technology. Hal Abelson: Her 1988 book, Computer Environments
for Children, is a real penetrating look at the genesis of the stuff that we’re all talking
about now when we think about educational computing and the different streams from which
it arose. Cynthia is the real expert in knowing about
this stuff and the one who can comment on what’s going on. Hal Abelson: Xiao Xiao is an interactive artist
and pianist who completed her master’s here at the Media Lab and she’s now a Media Lab
affiliate. She studies how people experience of music
can inform creating richer experience for people in thinking about information. She’s currently working on an exhibit in the
Historic New Orleans collection that includes a six-story elevator exhibit? Xiao Xiao: Actually, there are two installations. That’s the new one that’s at the- Hal Abelson: The one that is actually inspired
by Marvin’s Society of Mind. Xiao Xiao: Yeah, it’s not at the Historic
New Orleans collection but at a different institution. Hal Abelson: Yes, and maybe you’ll mention
that later. Let’s get into it. Let’s start, Cynthia, can you tell us a little
bit about the genesis of this book and about the essays and how it came to be? Cynthia Solomon: Yes. I’m going to stand up. Thank you. First, I want to say that because of my friendship
and collaboration and Marvin being my mentor, I’m often cited as someone that knows about
AI. I do not. I am and have been from beginning interested
in how computers can help children think about their own thinking. That’s where Marvin really excelled. Cynthia Solomon: About the book, the book
contains six essays written by Marvin. Five of them were written for One Laptop per
Child, a project initiated primarily by Nicholas Negroponte and Seymour Papert. The OLPC vision was that every child would
have a laptop. Marvin’s essays can be thought of as a spiritual
guidance to educators as to the possible magic and wonders of computers in children’s lives. Cynthia Solomon: The other essay was written
as a preface to a book that Margaret Minsky, Brian Harvey and I edited called Logo Works,
Challenging Projects in Logo. It was Atari logo. We were working thanks to Alan Kay at Atari
Research Lab in Cambridge. We got our friends to contribute Logo projects
to this book. Cynthia Solomon: Of course this book along
with these essays I thought of as defunct because the book Logo Works went out of very
quickly because Atari went out of business very quickly. The papers that Marvin wrote for One Laptop
per Child also were dissipated. Cynthia Solomon: The way I got involved is
Marvin wanted to publish some of his papers and asked for a volunteer and I volunteered
and I couldn’t help him with this AI papers very well, but I got Margaret Minsky involved. We focused on music, mind and meaning and
put on a symposium while Marvin was around and it was a great symposium here at the Media
Lab. I told Marvin I wanted to see his education
papers published in a little book. I liked the idea of a little book that people
carry around. That’s how I got involved with this. Cynthia Solomon: Now, I just want to mention
that Marvin and Seymour Papert, because a lot of Marvin’s writing is about things that
he and Seymour talked about. They were very close collaborators for almost
20 years starting in 1964. Their collaboration teetered out in the early
’80s. Seymour went off to Paris and Marvin stayed. Cynthia Solomon: Anyway, their collaboration
started around 1964. They focused on minds, both machine and human
minds. Seymour’s research with Piaget deleting genetic
epistemologist as Piaget like to call himself in Geneva, Seymour spent about five years
doing research there before joining Marvin at MIT. There, Seymour saw concrete examples of children’s
thinking. Children’s thinking was never an empty vessel,
but organic and changing. Cynthia Solomon: Examples of accumulation
of knowledge in its management Seymour kept seeing. Then there was in Society of Mind, which originally
Marvin and Seymour started writing together and then Seymour diverged and focused on writing
papers about education. There’s one important thing called the Papert’s
principle. This is a very good description. Some of the most crucial steps in mental growth
are based not simply on acquiring those skills, but on acquiring new administrative ways to
use one already knows. That’s what Seymour would see would Piaget
and the children, that they would have sort of assimilation and accommodation of ideas. Cynthia Solomon: How the book came about,
well, I just told you, I had met Xiao Xiao. We had some meetings with Marvin at his home,
a group of us talking about the possibility of setting up a Marvin Minsky Institute or
foundation, and a group of students and whoever. Cynthia Solomon: Xiao Xiao came to some of
those. I saw she had a sketchbook. “Ah,” said I. When I thought of doing this book, I pounced
on Xiao Xiao and said, “Would you mind to do some illustrations for the book?” I had no idea she’d make 80 of them, but she
did. They’re wonderful. As you can see, these are all Xiao Xiao’s
drawings that Marvin’s head is also a drawing of Xiao Xiao’s. Cynthia Solomon: Originally, she wasn’t going
to be a co-editor but she’s so good, how could I not have her as a co-editor. Then I thought about I talked to various people
and they said, “Don’t just publish the essays. Have somebody introduce things about them.” I made a list of people that I knew and Marvin
knew. I knew that these people loved Marvin and
admired him and that he liked working with them. That’s how I picked the people. Cynthia Solomon: Oh god. I forgot. I was going to show slides. Xiao Xiao: This thing. Cynthia Solomon: I forgot, oh yeah. Xiao Xiao: Here. It was on the table. Cynthia Solomon: I’ve got a clicker. I’m sorry. Click. I was going to tell you about these two. This is the only picture I have of the two
of them. It’s really a nice picture. They gave a series of lectures in Oregon and
the Oregon University Press published a book called AI of that and this is the picture
on the back cover. Cynthia Solomon: This is some tokens of Marvin’s
that I really liked. Well, I’m told not to read them, so I won’t. Here is Xiao Xiao in Marvin’s living room. Marvin’s sitting there in the back. See, she’s on a swing. Marvin installed that swing the year he moved
in to his house. It’s got the same ropes. It’s lived through his children, his grandchildren
and friends and neighbors. He picked the right kind of ropes. Cynthia Solomon: This is Mike Travers. Mike wrote the best introduction. Now, the reason I asked Mike to do it is he
wrote a wonderful essay on Marvin after Marvin died and I felt Marvin was in the room. He’s a very wonderful writer. Of course Alan Kay. Alan wrote an afterword for the essay that
was written while we were at Atari. I’m talking too long, aren’t I? Hal Abelson: Yes. Cynthia Solomon: Now, Alan is an illustrator
so that is in Xiao Xiao’s drawing. That’s Alan’s. In the book, Alan has a couple of other drawings. Well, can you guess who that is? Yeah, I found that on the web. Hal also because Hal was part of the Logo
group when the Logo group first started as part of the MIT AI Lab when Marvin and Seymour
were co-directors of the lab. He sort of grew up with Marvin as he stayed
with the AI Lab and [inaudible 00:15:07] and so on, so I asked him. Cynthia Solomon: Hal’s essay is wonderful
because he talks about Marvin’s work as a fugue. This is Gary Stager and I asked him because
Gary runs a fabulous workshop in t he summer for teachers. He for the first nine years of his workshop
had Marvin spend one evening with the educators and it was an incredibly enjoyable time. Marvin is very witty. Cynthia Solomon: This fellow was in the audience,
Brian Silverman is right over there. He was an undergraduate with Margaret Minsky
and Danny Hillis. They built this tinker toy machine which gets
talked about by both Alan Kay and Marvin. The tinker toy machine played tic-tac-toe. In Alan’s section, there’s a picture of it. Cynthia Solomon: Now, Walter Bender, there
he is, he just arrived. Great. Walter is somewhat responsible for five of
these essays. Marvin and he were really good friends and
he urged Marvin to write the five OLPC essays. His introduction … Cynthia Solomon: Then we have Patrick Winston. Patrick of course is the professor here that
teaches introductory AI and more advanced AI. He and Marvin, well and Geri’s husband is
here. They were students of Marvin’s in early days. I asked Patrick if he would write something. The essay he wrote about is one that Gloria,
Marvin’s wife and I suggested to Marvin and it’s about AI for kids. It’s full of examples of things that it would
be wonderful if researchers … actually, all of the papers in Marvin’s book are research
projects. Cynthia Solomon: Now this is my five minutes. I forgot. This is Margaret Minsky, who wrote the afterword. One of the things she included is some drawings
of Marvin’s that was done in 1965. It was a robot. Marvin sketched a lot. Marvin drew a lot of pictures. Here are Marvin and Gloria at some conference
rather. Cynthia Solomon: This one, there’s this young
lady who posted on Twitter a video and this is about a 30-second snap of her video. The first time I looked at it, she had 4000
followers. The next time I looked, there were 7000 followers. Speaker 4: I originally recommend getting
this, especially since … so there’s six essays in the book and each of them is prefaced
by some introductory remarks by- Hal Abelson: We need to hurry up. Speaker 4: … just incredibly impressive
people related to like cognitive science and especially computer science like Alan Kay
writes the intro for the first essay. He’s just such a delight to read. Alan Kay is fantastic. Yeah, I enjoy it because it’s sort of you
get a sense of different tone from each of these people introducing the essays and sort
of I think build a little bit of an intuition around the fact that people really do think
very differently. Cynthia Solomon: That’s it. Hal Abelson: Xiao Xiao, your perspective on
the book? Xiao Xiao: Yeah. Can we get some slides? Hi everybody. Oh my gosh. Well first of all, I’m incredibly honored
to be back at the MIT Media Lab, my home for many, many years. I was actually right next door in Hiroshi’s
group doing my PhD. Xiao Xiao: I’m supposed to be here, I think
I’m supposed to be here to talk about illustrating Marvin’s essays and by illustrating I mean
the first definition of illustrating, which is drawing. I have one slide about the process of how
I made the drawings. I went through all the essays one at a time. I think maybe out of everybody in the world
right now, I’ve read these essays more times than anybody else. I read through them many times and kind of
underlined different passages that I thought would be relevant or might be useful to have
a drawing to go along with them. Xiao Xiao: Sometimes when Marvin talks about
something that refers to some concept or another book, I went and just looked up those things
to try to deepen by idea about Marvin’s ideas. Xiao Xiao: Then the second thing I did was
for each essay, I tried to just make a list of all the images that I wanted to make to
go along with the essay. After I made the list, I try to think about
what each of these drawings would look like. I went through them one by one and I actually
went on the internet and put together a Pinterest board with a lot of reference images for things
that I wanted to include in the drawings. Xiao Xiao: I sort of put together these building
blocks in a sense. Then when it came time to actually making
the drawing, I would put pictures that inspired me kind of like a mood board in front of me
and then just improvise and imagine a way that they would come together in this sort
of format. I would draw it in pencil, make changes until
I was satisfied with it. Then I inked over it and always with a cup
of tea in the cat mug. Xiao Xiao: That’s basically it on drawing
and illustrating, but actually what I … Oops, go back. What I really want to talk to you about today
and what I’ll co-op the rest of my five minutes to talk about is illustrating once again Marvin’s
ideas but the second definition of illustrate, which is to make clear by example. Xiao Xiao: To do that, I actually want to
talk about music. One of the things that Marvin is known for
doing to everybody that was close to him was play the piano. Music was something that was very special
to Marvin, not just as a leisurely hobby but really as a way to find out more about how
the mind worked. Playing the piano and especially teaching
himself how to improvise on the piano in the style of Johann Sebastian Bach or Beethoven
was a way for Marvin to figure out the workings of his own mind. Xiao Xiao: There’s this article that I found. This is something that I’ve heard from a lot
of people, for instance Tod Machover talks about Marvin’s relationship with music, Margaret,
Gloria, but I found this article which writes, puts it in writing that we can actually cite,
which is this idea that Marvin has of course drawing upon his experiences of learning how
or improvising fugues, but really the process of building the machine in his mind to improvise
these wonderful musical pieces as a way of figuring out the workings of his mind. Xiao Xiao: Now, what I wanted to do or what
I started doing in the past couple of years was I saw it as a sort of following Marvin’s
footsteps of trying to use music as a vehicle or as a sandbox to understand the mind, but
instead of viewing it on the piano, I did on the theremin. Xiao Xiao: Does anybody know what a theremin
is? Has anybody tried to play the theremin? Does anybody play the theremin? The theremin was actually invented almost
100 years ago by a Russian physicist named, Lev Theremin whose name is often translated
to Leon Theremin. This is one of the first electronic musical
instruments in the world and it’s one of the only instruments that’s played without physical
touch. Xiao Xiao: The way that it works is that this
one is the one I have. It’s made by Moog. There’s a circuitry inside the box and there
are these two metal things coming out of it. There’s one antenna. The straight one detects … well, both of
them detects the proximity of your body. When you move your hand closer to the long
thin antenna, it modulates the pitch that’s coming out. When you’re moving your hand close or farther
away from this loop, it modulates the volume. Xiao Xiao: Because you don’t have any sort
of haptic feedback, the theremin is often considered near impossible to play, which
is why I decided to take it upon myself to learn who to play the theremin as an experiment,
pretty much two years ago on May 23rd, 2017. I know this because I started keeping a journal
on that date. Xiao Xiao: I want to share with you a bit
of the process of learning to play the theremin and to connect it back to the book somehow
because as you can see, well, this book was something that we had started working on around
2017. It just so happens that as I was starting
to teach myself to play theremin, I was also diving quite deeply into Marvin’s ideas. Xiao Xiao: Kind of as a practical side of
putting Marvin’s ideas into practice, going into Marvin’s ideas actually really helped
me learn how to play the theremin, if you will. Just to show you some in progress shots. This is from exactly two years ago on May
31st, 2017 about a week into starting to play. Can anybody hear it? Cynthia Solomon: No. Xiao Xiao: Can anybody tell what piece it
is? Okay, well, as you can see, just starting
out like I don’t really know what I’m doing. Nobody really knows what they’re doing. Anybody knows what piece it is? Xiao Xiao: Yeah, okay, so this is from the
same piece from about six months later in October or five months later. Just to show something a bit more recent and
my Instagram has been taken over by all theremin videos all the time. This is from just a couple of months ago. Xiao Xiao: What I’m doing here is actually
I had recorded one voice of this movement from a violin sonata by Eugène Ysaÿe. Then I put it on Instagram and kind of used
it as a backtrack and I played the other voice on top of it. Anyway, I’m showing this video just to show
a bit of the evolution in the way that I’m playing and the hand positions between there
and there and I guess to show the evolution of the technique. Xiao Xiao: I’m using the theremin as an example
because what I really wanted to talk about is how do you actually go about learning something
really difficult in unchartered territory. One of the things that often comes up in the
arts and also in mathematics and in languages is that people have these ideas about, well,
if you’re good at music or if you’re good at math, you are just gifted in some special
way. Xiao Xiao: Actually if you read Marvin’s writings
both in this book and also in the Society of Mind and Emotion Machine, I think what
Marvin really stresses is not just the natural gifts that some people may or may not have,
but wherever you’re starting out having higher order expertise, having strategies to help
you make sense of learning something in a difficult area or in unchartered territory. Xiao Xiao: One of these things that Marvin
talks about especially in this book is the idea of cognitive maps. Instead of just doing repetitive exercises
where you’re not even really sure where you’re going and you’re kind of just mindlessly going
about these tasks, trying to conceptualize whatever you’re trying to learn as almost
like a world that you’re trying to explore. Xiao Xiao: You’re starting somewhere, you
only see a little bit around you, but the you see that there are these other places
too and you’re kind of charting out a way of navigating this world and trying to explore
one area of it at a time so that it makes more sense to you. Xiao Xiao: Another thing that Marvin talks
about in these essays is this idea of reusing what you know. Once you have developed expertise in one domain,
you can almost take strategies from that expertise or from that domain and whenever you come
across a new area that’s kind of difficult, try to see what you can reuse based on what
you already know. Xiao Xiao: To give an example that’s sort
of along those lines, this is something that was really interesting that happened to me
on June 14th, 2017. I was trying to figure out how to play the
vibrato on the theremin. The vibrato is like when singers sing, they
kind of shake their voice in a way that is really emotional and really pleasing sounding. Oftentimes when people go to the theremin,
the first thing that they try to do is like try to make a vibrato but it really doesn’t
sound good because it’s just very … There’s no finesse at all. Xiao Xiao: I was standing there trying to
make a vibrato sound good and after a while, I figured it out and then I realized that
I had been playing with a certain hand position before, but when it started sounding good,
I looked down at my hands and I saw that I had this position. Xiao Xiao: My hypothesis was that I had actually
co-opted or my brain had automatically co-opted the dexterity of control from actually learning
how to draw or in from making these very fine lines with the drawing in order to make the
arm and wrist movements necessary to get the really subtle shaking to the bigger shaking
to the smaller shaking. Xiao Xiao: Anyway, I think the moral of the
story is read Marvin’s book, buy the book and you’ll get all of these learning strategies
that can help you learn really quickly something that is very difficult in whatever field that
you’re trying to learn and master. Hal Abelson: Great. Cynthia Solomon: That was wonderful. Hal Abelson: Thanks very much to you both. I’m told by the way we’re being livestreamed
and later, we can take questions either from the audience. I’m told people can submit them over Twitter. I wanted to start with the question maybe
for both of you and then for anyone in the audience who wants to chime in. One of the major things that really emerges
from this book that comes on these essays over and over is that AI for Marvin, the important
thing is about how the mind works, not about how computers work. Hal Abelson: One of the quotes I love from
these essays, Marvin writes, we should not let the dreary practicalities of billion dollar
industries crowd out our dreams and fantasies of building a giant money machine. When it comes to computers and education,
the key is getting kids to think about their own thinking. Hal Abelson: Again you’ll see in one of the
essays Marvin writes if quote good thinking is one of our principal goals then why don’t
schools explicitly teach how human learning and reasoning work. Again, many children could greatly augment
their resourcefulness if we could provide them with more effective ways to think about
their mental processes. To me that is constantly repeated in the essays. That perspective seems really different from
what’s going on in computers and education today. I’m wondering first the two of you, how do
you think about how those relate and anybody here to comment. Xiao Xiao: Cynthia, your turn. Cynthia Solomon: Marvin had a good theory
about children following their own projects. He said, “Well, that would need more than
teachers could handle.” With the internet and lots and lots of retired
people, it’s possible for children have mentors. Marvin was very big on mentors and [inaudible
00:33:52] that was Marvin’s word for a person that … a child kind of adopted. Instead of the child being adopted, the child
would adopt an adult who she admired and wanted to be like. Anyway, the teaching crisis could be ended
with networking. How is that? Xiao Xiao: What do you think about this Hal
because I feel like you know more about AI than I do? Hal Abelson: Well, what I think is again if
you look at the emphasis on AI going on today, and the emphasis on AI education, it’s not
particularly driven by children need a better way to think about their own thinking. It’s more driven by what Marvin calls these
boring billion dollar industries and for getting jobs. Hal Abelson: What kids really need education
in is how a computer works, which Marvin explicitly is saying he’s not interested in. I’m just kind of wondering how people think
about how those [inaudible 00:35:09]. Danny, what do you think? Danny Hillis: It’s worst than that. In part in education and how computer works
for hitting education and how to use a computer as a tool without knowing how it works. Hal Abelson: There’s a box for you to look
at when you talk. Xiao Xiao: You’re talking to the top of it. Danny Hillis: Actually I think it’s worst
than what you said because I don’t think kids are getting any idea of how a computer works,
because that might be actually kind of the thing one was talking about is this simple
idea of how mind might work. In fact, they’re just being taught how to
use computers as tools to do other things without being told how they work. I think you’re right. I think people are not at all applying the
thinking about making things in the school curriculum that I say. I think it’s actually almost the reverse of
things like and since we’ve gone backwards with some of the educational standards. Hal Abelson: Brian, want to add anything? Brian: I think you pulled out the right quotes
Hal is essentially … Xiao Xiao: Talk in it. Brian: Yeah. Hal Abelson: Talk at the top of the box. Brian: Yeah, okay, at the top of the box. The quotes that you said in asking the question
I think provide the answer. What the book is as Marvin’s thinking on how
BI works with education. Then people are talking about how AI fits
with education. Brian: The reason I think Marvin said to you
that we should call the thing BI is the definition has shifted. It shifted in the way where what Marvin was
calling AI back in the ’60s really was something that was thinking about thinking, but to just
repeat what Danny is saying, is what people are calling AI now no longer is. Danny Hillis: Maybe just expanding on that
second. Marvin sort of assumed when talked about AI
that pattern recognition was going to be a sort of subroutine and your perceptron was
kind of about that subroutine or something, but it didn’t consider pattern recognition
to be AI. Ninety percent of what’s called AI was something
that Marvin sort of considered a subroutine. Hal Abelson: Can you take the box? Can you [inaudible 00:37:43] but then give
it to her when- Mitch: I think one of the things is that because
Marvin talked so much about this idea of our understanding our understanding in using that
in order to improve it, grow it, be organic with it, a lot of the things that we think
of as AI now, which Danny mentioned, we can’t really understand them work, how they work
because nobody understands how they work yet. That makes it especially difficult just because
they’re so opaque to reasoning or understanding. Julie: I’ll talk. I just heard something really great. Xiao Xiao: Julie, we can’t hear you. Hal Abelson: Take the box. Xiao Xiao: It’s for the remote people. Julie: It’s so scary. I just heard something really great. I’m just getting involved with a high school
that has a computer science academy but it’s a really nice one that lets everyone try it
out. I can’t stand the idea of coding but I was
talking to some of the kids and one of them, the biggest thing they were excited about
was figuring out why something broke. Julie: I realized they didn’t really have
a good word like debugging but I think my dad might have liked the idea of calling the
new skill that they’re trying to teach call it debugging and go from that end instead
of coding. It was exciting to find out that kids were
actually discovering from themselves that that was, at least some kids that that was
the really fun part of programming. Xiao Xiao: I think I totally agree with that
actually and just to riff on that idea, I feel like right now AI is kind of like this
hot topic, like everybody wants to get a piece of AI and by that, I mean everybody wants
to make money off of it. I think that actually as this came up in the
past few speakers comments is that for Marvin, the sense of AI is almost, it’s just as much
as about building machines that could think as about understanding the human mind. Xiao Xiao: I think in Marvin’s books, in Marvin’s
writings in this book and even in Marvin’s own life, he did a lot of different activities. In whatever he did, I got the sense that he
was trying to figure out what was going on underneath the surface both to better understand
the thing to debug it and also to understand how he could understand the thing. Xiao Xiao: What’s really interesting is like
in this book, I think in essay six, Marvin doesn’t just talk about computer science being
the only way to understand learning or to build expertise in discipline that would really
people. He talks about building physical crafts. He talks about like music, about, what else,
learning magic tricks. I think I guess I just want to reiterate that
I think it’s about following children’s interest. It’s almost like in whatever discipline that
you feel pulled toward, you can find ways to learn how your own mind works and to learn
how to learn, and to learn about thinking. Xiao Xiao: Really, this kind of ties back
to Seymour also and to what Mitch talks about the importance of passion in education. Actually just like as an aside, AI, ai in
Chinese means love. Maybe the most important thing about AI in
education is just having the passion to pursue what you want. If you have the passion, then you will try
to find mentors to help you, you’re trying to find ways to solve problems. Throughout the process, you have a deeper
understanding not only of the field, but also about your own mind. Hal Abelson: Jerry, you need the box. Xiao Xiao: Jerry, could you throw that to
Jerry. Jerry: Where is the top? Hal Abelson: There is the box. Talk into the top. Jerry: This is the top, okay. Well, what I’d like to do is connect this
to the stuff that Seymour said to me many years ago was a challenge for this kind of
problem, which is that it’s difficult to think about thinking, without thinking about, thinking
about something, that’s one thing. Jerry: However, Marvin’s response from 1961
and the design and planning paper he wrote was the computing is a good medium for expressing
poorly understood ideas. I don’t remember the exact words, but something
like that is the title of the paper. Of course Hal and I took that to heart when
we wrote our book SCIP because we’re worried about the fact that computation was a beautiful
medium for expression, for novel expression, novel medium, the expression of ideas like
how to rather than just what is. Hal Abelson: [inaudible 00:42:32] Who else? Walter. Where is Walter? Can you throw this? Cynthia Solomon: Yeah, throw it. Walter: I think the thing that’s missing from
the discussion so far is that Marvin also had very definite ideas about structure and
about how these pieces all fit together that learning wasn’t just this random walk, but
there was actually a lot of structure in the mind. I mean, that’s what Society of Mind and Emotion
Machine were all about. Walter: It’s also a reflection of the structure
and where you’re fitting into that structure and moving up and down between these layers
and the cognitive powers. I mean it’s not just being excited about learning
but it’s being immersed in that structure and starting to explore that structure. Xiao Xiao: That’s a really good point. I think maybe in what I was just saying I
maybe overemphasized the part about just like being passionate but I definitely agree with
you. I think that and actually I think I sense
a little bit of pushback based on what I just said. From the last two comments. I do agree that it is both. It’s always having some sort of a structure,
and then also having some sort of improvisation and having I guess yeah, I mean it’s important
to come up with or to be able to experience different ways of thinking about things in
different methods of thinking. Cynthia Solomon: Gloria has a question. Could you throw it over to Gloria? Gloria: Talk right into this X? Hal Abelson: Yes. Gloria: I just want to go back to a [inaudible
00:44:27] that came up before and that is about children learning by mistakes. I had the privilege I guess of being a health
official in a school system. I was very much interested in the interface
of health and education. The one thing I was very troubled by and that
is everyone said you learn by mistakes and I think you talked a little bit about that
Cynthia. Gloria: I think I talked to Marvin a lot about
this. I was troubled because kids do not seem to
be learning by correction of mistakes. They just got shamed and embarrassed and then
maybe some learn and some learn to be a little bit afraid of the teacher and didn’t raise
their hand the next time. As I say, I had the privilege of observing
kids in many classes. Gloria: I think that the whole notion of helping
kids learn, not learn by mistakes, that that theme I think is so important and I think
that that is one of the building blocks to me of some of the things that Marvin said
about how to learn, and learning by activity and learning by doing things. Gloria: I remember when we were … There
was a science fair and we were trying to think of what to do and many people did very nice
experiments showing the children different things, oxygen and hydrogen and everything
else. Marvin just brought in a bunch of worn out,
or not so worn out computers and parts of computers. There was a room setup and kids could take
them apart. I think this kind of illustrates some of the
things that the practical side of some of the things that Marvin, the aspect of his
learning by activity. That’s one of the most important things that
I think has come out of this innovative, the innovative aspect of all of the things that
have come up today. Cynthia Solomon: When Seymour Papert and I
collaborated and taught children what we emphasized that children was debugging and procedural
thinking and talking about what you’re doing. Anthropomorphic thinking, Seymour called it
body syntonic but Marvin would say and I would say, the kids play computer, be a computer. Be a turtle. Cynthia Solomon: Marvin carries that on in
talking about cybernetics is a good introduction to children to think about animal behavior. That’s the kind of thing we did in Logo classes,
where it gave us as Marvin mentioned, it gave us some distance between our cells in our
bugs so we could talk about them and articulate and debug ourselves as we think about how
to debug the computer that we’re working on. Cynthia Solomon: One of the things that is
hitting education now is the maker movement. Marvin would be very excited about parts of
that. He felt it was really important to build things. He loved tinker toy. There’s a wonderful story about Marvin and
tinker toy, but I’d rather have Danny and Brian and Margaret talk about their experience
with tinker toy because they set up this tinker toy computer. Cynthia Solomon: The danger for that, with
Marvin and with education in general is so much time is spent with mechanical devices
trying to get them to work. You want to modify that with what Marvin,
what I would call simulations, with things online, which are easier to program and debug. There’s some modification. Cynthia Solomon: The thing is that building
these things are not for the sake of building those things, but for learning about the process
and relating it to yourselves. That’s something that gets missed in the maker
movement. I think it gets missed in people using deep
learning with machine learning with kids. Cynthia Solomon: Toss it. There’s a person in the back. Speaker 12: Yeah, I wanted to hear a little
more about what Marvin thought about learning from mentors, especially learning from people
who think differently in the plurality of knowing and whether it’s multiple intelligences,
all the Howard Gardner or Society of Mind, just how do you think about that relationship
with the mentor, if you can say more. Cynthia Solomon: Well, he thought it was very
important in the development. Hal, talk about it. Hal Abelson: Let me instead for finishing
to see if I can get people upset, especially Cynthia. One of the key ideas in this whole approach
of Marvin and Seymour and all of Logo is that a good way to learn about thinking about thinking
is to express things as computer programs. That’s the Logo [inaudible 00:50:19] was the
first is there was a paradigm example of that. That really fits well with the perspective
of what people these days are calling good old fashion AI. A lot of which comes from thinking about how
the mind works and how you do a problem and then sort of trying to build that into programming
mechanisms. That leads to the idea, which is central to
Logo, that the way kids can benefit from thinking about thinking is to write programs even difficult
ones. Hal Abelson: Marvin talks about that of course
in the first essay and also in the last one. He also talks about how it would be good for
kids to think about how programs work even difficult ones. He talks about for example kids thinking about
how you’d distinguish a picture of a dog from a picture of a cat and says in that essay
that he doesn’t know then of any successful project that did that. Hal Abelson: The R&A is Marvin wrote that
essay in 2009, which was just three years from the breakthrough in convolutional neural
networks. Now of course it’s kind of an exercise to
write something that distinguishes a picture of a dog from a picture of the cat. The machine learning kind of neural network
approach to that kind of issue in general those kinds of problems is very different
from the kind of modeling that certainly we were thinking about in Logo when we said kids
should think about how you’d write a program to do that. Hal Abelson: Is it time? Maybe. For a new approach to computing and education
that shows not so much on good old fashion AI, but rather on something more like what
happens in machine learning today. Is there a philosophy of education and approach
that shows more from current machine learning than from the way we were and people still
are thinking about AI going back to the ’60s. Anybody upset enough to comment on that, Cynthia? Jerry, can I get you upset enough to comment
on that? Jerry: The problem is … Hal Abelson: Take the mic. Jerry: That upsets me a great deal because
the problem is that it’s more important to learn how to think, and I think that by studying
the right now you look at the machine learning stuff, it’s opaque. You don’t get any understanding of what it
knows or how it knows it or how it learns because what it basically is, is function
approximation by [inaudible 00:53:09] parameters, according to a good plan but that doesn’t
help very much. Jerry: What I really worry about is how to
make people think better. I’m not so worried about making machine do
a good job. I think the way you make people think better
is by teaching them how to think and it’s not going to be by having a bunch of parameters. People don’t work that way. Richard: Yeah, I think Marvin would think
that was a false dichotomy. One of the beautiful things that on one project
I did with Marvin was he was interested in building some robot prototypes and he started
talking about Fourier Transforms and I was like what’s that have to do with mechanical
things. Richard: We can apply that idea that people
usually think about in signal domains to space, to shapes. I think Marvin would probably say that we
can use the nonrepresentational AI as another way to start thinking about how to represent
how we apply technology to different problems to help us think about all the alternatives
that we have and all the tools we have and how to get kids to really look at thinking
out of the box, use whatever you can and attach it, whatever works. Richard: Then also tell better stories about
things, so bringing back the not the old version of AI but to add the color and the representational
interest from those means to help the new things work better. Hal Abelson: Anything else? Yeah, Margaret. Margaret Minsky: I’ll stay with the personal. It’s intriguing in light of what Richard just
said and your question and the mentorship experiences that Xiao Xiao and Cynthia Solomon
have had with Marvin, sort of how those bumping into collaborative projects, whether it’s
music or writing. Margaret Minsky: The one that I’m looking
at right now is funny in terms of machine learning because Marvin Minsky built the first
realized machine neural net learning machine in 1951. He must have learned a lot of from building
that. We maybe next book but we’re trying to figure
out a little more about how he actually did that. It was an analog computer. He deliberately decided not to use the digital
computers of that age because they didn’t have enough memory to have 40 neurons and
he knew he needed about 40 to do anything at all. Margaret Minsky: This is real stuff. It was really built. It was a big old machine. It worked and have lots of vacuum tubes and
lights. It was kind of like making a drawing. It was making a thing. I don’t know where to go with that, except
that he lived off that. He actually built the first one of those things. Margaret Minsky: Then he made a change towards
symbolic AI. It’s a great question about how he would make
use, take the dichotomy that he saw later and bridge it somehow so that you’re thinking
about interesting things when you work with those building blocks. Hal Abelson: Jerry? Throw the box to Jerry. Jerry: Yeah, I actually agree that that’s
true. I of course one of the first things I ever
built was a W Grey Walter device that walked around in the living room of my apartment
in Brooklyn New York well before I came to MIT and met Marvin and it was basically an
analog little robot. Jerry: Indeed the very first program I wrote
when I got to MIT was a little learning, a little neural net [inaudible 00:57:15] thing
that learned tic-tac-toe. However, I think we’re talking about education
of children, that was the place where I got upset, when Hal said, maybe we should think
about how we teach children because I don’t think the mechanisms that we see in teaching
about things like how to use kind of convolutional neural nets to do things and particularly
informative teaching children how to think. That’s what I’m worried about. I want to separate those two things quite
a bit. Okay? Hal Abelson: Danny? Xiao Xiao: Just one little thing. Don’t you feel like whatever subject that
you’re trying to learn or trying to build whenever you’re trying to learn anything it’s
like you’re constructing some sort of machine in your mind that both understands the thing
or that enables you to do something. Xiao Xiao: In that case, I feel like, yes,
as what you said earlier about having ways of describing processes is really important,
but I feel like that could be applied to really building anything from Marvin’s snark machine
or from Marvin teaching himself to play fugues to building a physical thing. It’s like you always have to come up with
like I mean algorithmic or procedural ways of thinking or subroutines or debugging, or
it’s like useful for everything I feel like. Hal Abelson: Danny? Danny Hillis: I think Marvin would have liked
it that you asked an annoying question. Hal Abelson: Yes. Danny Hillis: Because I think one of his principles
was that when he saw everybody is starting to think about something in the same way,
he would look for a different way of thinking about it. I think there’s no question he would be doing
that with AI right now. Who knows what his different way of thinking
about would be but it surely wouldn’t be the same as anybody else’s by definition. He would go in that direction. Danny Hillis: In fact, you so much believe
part of what it was like to have Marvin as a mentor is he applied that principle to himself. I remember going and asking him about to explain
some principle in mathematics to me and he explained it and I said, “Okay, thank you. I understand it.” he’s like, “No, you don’t understand it. I just explained it to you in one way. You don’t understand something if you only
understand it in one way.” Then he gave me a completely different explanation
of it and so on. Danny Hillis: That was his idea of how to
approach something. I think that he would actually be very happy
in teaching kids a different way than whatever way most people were teaching them. Hal Abelson: Ken? Ken: One thing is if you look at Society of
Mind or the Emotion Machine, a lot of it is about creating architectures where flaky uncertain
almost black boxes can be put together in interesting ways to be more than the sum of
their parts. I think he would think that that was great
to find a way to teach. How do you make something where you don’t
understand everything down to the most basic level but you can still put them together
to make something that works better together. Speaker 16: Could I repeat what Richard said
because I think falls dichotomy. I think the fact that there are simple systems
that could distinguish between dogs and cats would have caused Marvin to rethink what he
thought was the limits of neural nets. Cynthia Solomon: Can you speak- Speaker 16: I think it would have caused Marvin
to rethink what he thought was limits of neural nets. Because I think Danny had said that there
are nothing more than subroutines but they’re pretty interesting subroutines and they’re
way, way, way better than as us back in the ’70s and ’80s thought they would get. Speaker 16: At the same time, I don’t really
see any reason. Even if there is an educational value to machine
learning, why would that cause there to not be an educational value for good old fashion
AI. Speaker 16: The idea of saying that well because
AI has moved on or thinking of education, but what we used to do no longer is valuable. What we used to do with good old fashion AI
and education is still valid. Speaker 16: Another thing I wanted to say,
which is tangentially related is Gloria referring to mistake and Cynthia referring to bugs. One of the things is I think it was really
important to stop talking about mistakes and start talking about bugs. One of you said we should say a little bit
about the tinker toy, tic-tac-toe machine because it was fun. One of my memories of where the project got
started, not the very beginning, spent a wonderful weekend in I don’t remember where. Speaker 16: Marvin and at least Danny and
Margaret were there. We didn’t actually build anything but we talked
about how to build a tic-tac-toe playing machine or the tinker toys. What Marvin was saying that weekend was 90%
crazy. The thing is, is it was 10% not crazy. Speaker 16: From that 10%, we were able to
build the first version of it, which almost worked. From the version it almost worked, we were
able to take another step back and build a second version, which did work. Now, referring to any of the presteps as mistakes
just seems to not been doing justice to the process. Hal Abelson: Any closing thoughts? Closing comments. Buy the book. Xiao Xiao: Buy the book. Cynthia Solomon: Well, there’s a reception- Hal Abelson: Downstairs, second floor. Cynthia Solomon: Room 244 and I think there’ll
be some books there. If you haven’t bought one already, shame on
you. Here’s your opportunity. Speaker 17: Have it signed by the authors,
the editors. Cynthia Solomon: The editors. Marvin is the author. Gloria, if you’d like a signature, Gloria
is very good at replicating Marvin’s signature. Thank you all. Hal Abelson: Thanks everybody.

Leave a Reply

Your email address will not be published. Required fields are marked *