Podcast

Podcast on Emotions and Learning 

(Variable Speed Playback & Search)

Original Podcast Site with Adam Alonzi, Podcast Host

Click here for Audio MiniWindow only

Adam Alonzi: Ladies and gentlemen, Barry Kort.

Barry Kort:

She was a first-grade teacher which was okay because in first grade you don't do a lot of reading. First graders read polysyllabic words very slowly so that wasn't an issue for her. But I suggested that she go back to the community college to get a Masters in Education so she could advance her career. She took me up on the suggestion.

It turned out that one of the required courses was a course called Statistics for Education. I've got tons and tons of math under my belt; I tutored math all the way through high school and college and as an adult. But she had an enormous amount of math anxiety, and I had seen math anxiety before, but nothing like this. This was just enormous. And also she was emotionally volatile and she had not just math anxiety, she would fly off the handle for seemingly no reason at all, all the time. And I was very perplexed about why she had this emotional volatility.

She gets up to Chapter 4 in this Statistics for Education, and she comes around and she's absolutely stuck, but not just stuck, but she's just fit to be tied. So I said, "Well let's take a look at your homework." So the homework for Chapter 4 — which was on Hypothesis Testing — that was the subject of Chapter 4. She had to do the problems at the back of the chapter and she had no idea how to do them and was just extremely distressed. I said, "Let's take a look at the first problem."

The first problem was a word problem that went like this: The principal at an elementary school is surveying how the students are doing in different classrooms and he notices that in the classroom, some students are raising their hand and volunteering to recite and other students are reticent and holding back. And the Principal notices that students who were volunteering to recite are getting the better grades, and the ones who are holding back — not volunteering to recite — are getting the lower grades. So the principal forms a hypothesis. The hypothesis is that being verbal — speaking up — causes high grades. And so he mandates a course in public speaking for all the students, assuming that would improve their grades. And the problem is to critique the principal's hypothesis.

My friend is absolutely stuck — she has no idea how to even think about this problem. So I'm using the Socratic Method — I don't want to just give away the answer because that would spoil the exercise. So I start asking questions in the style of the Socratic Method which is how I always used to coach math anyway. So in the Socratic Method, if you ask a question and they don't know how to answer it, you ask a simpler question. You keep simplifying the question until you eventually get down to what I call the "atomic question." The atomic question is where you say, "Well, would you say 'A' or 'B' where the answer is obviously one of those two. So I got down to the atomic question, "Would you say that being verbal causes high grades? Would you say it's the other way around — that getting high grades causes them to be verbal? Could it be a coincidence or maybe something else like studying the night before and knowing the answer causes both?"

So basically I lay out four possibilities for the arrow of causality: 'A' causes 'B'; 'B causes 'A'; it's just an association, or they have a common underlying cause. I'm ticking off the four, and as I do this she has an epiphany. When I say she has an epiphany, I mean she has an emotional outburst like I've never seen before. She's screaming at the top of her lungs, "This seems to be a very important idea!!" And she's furiously writing it down. And I'm sort of startled by this case where I was present at the moment of an emotional outburst which is clearly precipitated by this little episode.

So I start to reinforce it and she says, "Shut up! Shut up! I've got to write this down before I lose it."

So over the next few days, I'm scratching my head about this anecdotal observation. Here, after all this time I've seen all this emotion, where I had no idea where it was coming from. Suddenly I saw that she had a gap in learning that was so profound that I never would have guessed except that I was using the Socratic Method, and I precipitated this epiphany — this moment of "Aha!" — where she understood something that for me, as a scientist and as an engineer, I just took for granted. I've known this basically my all my life.

So I thought, "Wait a second!" Finally, I see a connection between an expression of emotion and an episode of learning where I was midwifing the epiphany. I was present at the moment. And so I thought, "Holy Schmoley!" I never noticed before the connection between emotions and learning, except for this one anecdote. So I thought, "Gee, let me double check this." And so I began to pay more attention, and I realized that a great amount of our emotions are associated with learning or absence of knowledge or absence of the ability to solve a problem or figuring things out. So over time as I thought about this, I began to structure this relationship between emotions and learning.

And there is a chart in the paper that shows a learning curve. Now, everybody's heard of learning curves — it's kind of a cliché. And when people draw learning curves, they usually draw them rising steeply at the beginning and then slowly leveling off, but always rising and then flattening out. And I thought, "Nah, that's not right because we don't learn error-free."

A learning curve that rises and never falls is called monotonic — that's mathematics for always going the same direction never turning around and going back. I said that when we learn we sometimes have misconceptions — we learn things incorrectly. So we have beliefs which are incorrect, and at some point, we figure out something we once thought was the case isn't and we have to discard our misconceptions and throw away an erroneous belief. So a learning curve really doesn't always rise. Sometimes it falls when we discard a misconception. So I put wiggles in my learning curve that rises up to a point where you have a belief. Then you weed out the misconceptions, and your cumulative knowledge that you rely on declines briefly, and then you discover a better idea — a better hypothesis — and you replace the previous myth with something better. So a learning curve has these wiggles.So the first thing is that it has a slope. It can rise steeply — that's fast learning. It can rise slowly — that's slow learning. But you also have unlearning. That's when the slope turns negative and goes downhill. But you also have this curvature. And now I've highlighted the curvature of a learning curve. Not just that it can slow down but it actually has these phases where it's curving downwards like a frowny face and times when it's curving upwards like a smiley face. I thought, "That's where the emotions are! They're in the curvature of the learning curve."

When you're laboring under a misconception, you're building a belief, but the belief has got a flaw in it. It's either an error or it's got gaps in it. And eventually what happens is that your expectations — you have a hypothesis, you have a belief and you have expectations based on it (it's like making a prediction in science) — if the prediction doesn't come true, they think, "There's something wrong with my model. My hypothesis — my belief, my model — is buggy. It's not making accurate predictions."

And so you become kind of disappointed or maybe surprised or maybe frustrated, and eventually, you suspect that there's something wrong with your theory — with your model, with your hypothesis — and you go looking for a better one. Your emotions will go negative if you're laboring under a misconception. When you start to build a new idea that seems to be working better, your emotions go positive. We start to become hopeful and have positive expectations again.

What I got out of this single anecdotal observation was this mathematical model relating emotions to learning. And so, as time goes along, on that on that chart in that paper, the vertical axis is cumulative knowledge and beliefs. The slope of the learning curve is learning and the curvature of the second derivative — if you know the terms in calculus — and the higher order derivatives correspond to emotions.

So I made this discovery back around 1985. I thought, "Now I've taken Psych 101. How come I've never come across this model before?" Either it wasn't covered in any of the books I'd read on psychology or maybe I just wasn't paying attention that day. So I went looking — well let me be sure to see if I had simply stumbled on something that was well known. And I couldn't find it anywhere in the literature.

I thought, "Can this be an original theory that's not anywhere in the existing literature?" I mean I'm not even a psychologist. I'm an engineer, a network planner. So I mentioned to some of my colleagues in academia about this model relating emotions to learning and they kept saying, "Barry, this cannot be an original theory. It's got to be in the literature. You're just not looking hard enough." I said, "I looked and looked and looked." They said, "You're not looking hard enough."

So finally my friend who is a professor at the MIT Media Lab says, "All right, I'm going to look." So she looks and she finds a little bit of stuff about emotions in the literature — not very much — but she doesn't find my model. She says, "This model doesn't seem to exist anywhere in the literature, it appears to be a novel model."

By this time it's the mid-90s. And so I've been talking about this model — this hypothesis that emotions are the second derivative with respect to time of the learning curve. And my friend at MIT, after she does this literature survey, she says, "Now that I've done all this literature survey, I'm going to write a brief paper documenting my literature search, because I did find some stuff," she says, "Not the stuff that you're talking about but some other interesting stuff." So she writes a brief paper on her literature search and she even comes over to where I'm working at Bolt Beranek and Newman and gives a brown bag lunch talk. And then about a year later she says, "You remember that the paper that I wrote about on emotions?" I said, "Sure I remember. It was a literature search." She says, "I'm going to turn that into a book."

And after a year she writes a book called Affective Computing. Her name is Rosalind Picard. She writes a book called Affective Computing and she founds this entire new discipline in Artificial Intelligence and Cognitive Science called Affective Computing — this is the mid-90s — and about this time another colleague of mine who is a primary school teacher out in western Massachusetts was beginning to use the Internet at school. I meet him at an IEEE meeting and he wants to get his school on the Internet.

And so I've been telling them about all this same stuff and he decides — now he's got a Masters in Education — he decides to go back to U-Mass Amherst and get an Ed.D. degree - a Doctor of Education Degree — to advance his career. And he finishes his Ed.D. and then he says — this is about 1998 — he says, "Now that I have an Ed.D. I should learn how to write grants."

He says, "I want to learn to write a grant," he says, "but the first grant you write is just for practice because nobody ever gets their first grant." He says, "Let's brainstorm ideas." So I give him a bunch of ideas including this model of emotions and learning that I've been kicking around now for almost 15 years. He says, "Well this is so far over the horizon that it will never get funded but it's a good one to write up because no one will have ever heard of it."

So so he writes up a grant proposal for the National Science Foundation — on this emotions and learning — and my friend at MIT, Roz Picard, she signs on to it also. We submit this grant proposal to the NSF around 1999 and lo and behold it gets funded. We write our very first paper after we get funded around 2000 — just basically presenting the theory. There are no results. All we've got is this mathematical theory and we write a paper — the three of us write a paper — and we present it at a conference at Madison Wisconsin called the International Conference on Advanced Learning Technologies. By now it's 2001. So I go to Madison and I present this paper — just basically a theory paper — and at the end the conference, the last day we're sitting at the banquet chatting with somebody I'd met at the conference — and I'm chatting away amiably — and she says to me, "Barry, they just called your name."

I said, "What?" And she said, "You just won the Best Theory Paper Award." I look up and sure enough our paper — right out of the box — wins the Best Theory Paper on this theory of emotions and learning, that was an entirely novel theory, nowhere in the literature at all (up until that time).

So that's the story of how this theory came to be — from a single anecdotal observation in 1984-85 — of whose long gestation period until we finally published it in 2001.

There were people who thought that emotions and learning were connected. In the mid-90s, there was a professor — I think she's at Harvard — named Priscilla Vail and she wrote a book called Emotion: The On-Off Switch for Learning in mid-90s. And so she had identified that emotions were connected to learning but only that they're connected — no mathematical model, only a qualitative observation they're connected — in the same way that I had noticed this singular anecdotal observation of an extreme emotional outburst connected to an "Aha!" moment.

So what we did is not so much notice that emotions are connected to learning, but come up with a mathematical model that was unprecedented in the literature, that gave a model. What's interesting about the model is it's not limited to human learning or even animal learning. It would apply to any learning being or any learning system be it made of meat (like us) or animals or made of silicon like learning machines, learning computers, or autonomous learning systems of the 21st century, or even imaginable alien life that's able to learn. Anything that's able to learn is going to have a learning curve.

And in general, even if you're an optimal scientific learner — a systematic learner — you're still going to form hypotheses which turn out to be mistaken. There's no reason why your very first hypothesis out of the box is going to hold up after you test it and evaluate it. So any kind of learning being or learning system is going to form hypotheses, test them, occasionally become disappointed because the hypotheses don't make good predictions, then have to discard them and replace them with a better model. And so a learning curve will always have these wiggles in them — in the second derivatives and higher order derivatives. And so in humans, we experience those wiggles as emotions.

And my colleague at MIT prefers the term 'affect' because 'affect' is sort of more clinically neutral term. But whether you call it affect or emotions or whatever vocabulary term you like, you're going to have the second derivative — the wiggles in the learning curve — for any learning system. And so what we call emotions, any learning being or any learning system will have something analogous to it. Which means that when we start to see adaptive learning systems in the 21st Century, they will have states — confusion, puzzlement, disappointment — the terms that we use, they will have those similar states. And we might as well really use the same vocabulary terms rather than invent a whole new set of vocabulary terms. So emotions are inherently connected to learning — at least some emotions — and in the paper, I identify half a dozen of the emotion axes — I'll just read them off here:

So the very first axis that I identified was what I called the Anxiety-Confidence Axis. If you don't have a good scientific model to predict what's going to happen, you have anxiety or worry. If you have a very good model that makes pretty good predictions, your anxiety is replaced by confidence. So if we don't like anxiety, and if our anxiety is caused by not knowing — not having a good model — then that motivates us to learn. And if learning is successful the anxiety subsides and is replaced by confidence.

Now, of course, if you have too much anxiety it can actually interfere with the learning process. Back in 1985 in that story of the woman I was coaching had so much math anxiety that it basically interfered with the learning process and she needed a person like me — a coach or mentor — to sort of walk her through it.

The second axis on my list is Boredom versus Fascination. So if you're not learning you get bored and what you'd like is to study something that's absolutely intriguing and fascinating. So Boredom-Fascination is the second axis.