kitchen table math, the sequel: Eureka, part 3: MOOCs and flips and dopamine

Saturday, December 14, 2013

Eureka, part 3: MOOCs and flips and dopamine

Eureka
Eureka, part 2
Eureka, part 3
Eureka, part 4
Eureka, part 5

So, MOOCs. Why don't they work (setting aside, for the moment, Anne D's experience)?

I had just gotten to the point where sitcoms gave me the answer I was looking for when I had to take a train to the city.

So, to reprise:
Ed brought up movies.

When you go to the movies, he said, the screen is huge, the sound is deafening, all the lights are turned off, you can't talk to your friends, and you have to stow your cell phone. Plus a movie lasts only a couple of hours, then you never have to see it again unless there's a sequel that you really want to see, and you don't have to see that for at least a year.

And even at the movies, even with all the ploys and devices filmmakers and theater designers have developed to hold your attention, if the plot sags, your mind wanders.

MOOCs don't have any of those things, so good luck. The wonder of it all is not that the drop-out rate for MOOCs is catastrophic, but that anyone thought they were a good idea in the first place.

Ed continued.

TV, he said, had had to follow in the footsteps of movies. TVs are bigger, the sound is louder, the experience more immersive….

That's not really true, I said. It's definitely not true of sitcoms. Sitcoms are the exact same hokey, flat-lit, 3-camera affair they always were, with the laugh track telling you when to laugh, and they work. They always have.

That's when it hit me.

"Reward prediction error"

In graduate school (I have a Ph.D. in film studies) I was intensely interested in comedy; I wrote my dissertation on 1950s comedies. 1950s comedies are fabulous, but what I really wanted to know was: what is humor?

What makes things funny?

I've wondered about that for my entire adult life, and have probably, finally, found at least a partial answer, which has to do with -- hold your breath -- the basal ganglia. (For passersby, I have been slaving over a basal ganglia writing project for years now.)

Research on the basal ganglia is very new, so take what I'm about to describe as provisional.

The basal ganglia seem to be all about reward prediction error.

"Reward prediction error" means that learning happens when you predict a reward and you are wrong.

Here's how it works. (Or may work).

Dopamine spikes or drops in response to "prediction" errors, that is to mistakes we make predicting rewards.


  • If you expect something good to happen & it doesn't, dopamine drops. That feels bad.
  • If you expect something bad to happen and it doesn't, dopamine spikes. That feels good.
  • If you're not expecting anything good to happen one way or another, and all of a sudden, out of the blue, something good does happen, that feels great. Because dopamine.


(I've learned just this week that "because" has become a preposition.)

So, if you're expecting a check to come in the mail and the check comes in the mail, no dopamine spike. You open the check, you're mildly happy (if that), you deposit the check. Life goes on.

If you're not expecting a check to come in the mail and a check in the exact same amount as the expected check-in-the-mail arrives without warning, that feels great because dopamine.

A surprise check in the mail feels great because dopamine fires in response to good surprises (or to "better-than-expected" rewards.)


Reinforcement learning

Reward prediction error is the basis of reinforcement learning.

It's unfortunate that "reinforcement learning" is called "reinforcement learning," because "reinforcement," to me, sounds as if learning takes place when the same thing happens again.

Instead, reinforcement learning takes place when something new happens, something you didn't expect.

("Something new" includes something old but better -- or worse -- than you expected. I know the whole thing gets incredibly confusing right around this point, but just remember the surprise check in the mail: how different it feels from the fully anticipated check in the mail. The surprise check in the mail produces reinforcement learning; the expected check in the mail does not.)

For the record, I don't know how learning via distributed practice, via repetition, relates to reinforcement learning, so that question will have to remain a mystery for the time being.

Reinforcement learning in the sense of the incidental learning we do naturally throughout the day (what is good? what is bad?) depends on mistakes. "Reinforcement learning" happens when we are wrong, when we are wrong in a very specific way: reinforcement learning happens when we are wrong about the goodness or badness of what comes next.

Drug addiction is probably a phenomenon of reward prediction error, btw.

Normally we habituate to good things. We get used to them; we no longer feel ecstatic when they occur. But addictive drugs always spike dopamine -- that is their effect inside the brain -- and that is what makes them addictive.

Cocaine spikes dopamine every time you use it, so  your brain is always getting a 'REMEMBER THIS AND DO IT AGAIN' message, and your interest in taking cocaine always increases. Addiction is a form of learning, a form of overlearning, more exactly.

At least, that is the way I understand the reward prediction error theory of drug addition, as a "disease of learning and memory."

(Interesting 2012 research here…dopamine and GABA…)


Surprise is good

The long and the short of it: surprise is good.

Good surprise is good.

Bad surprise is bad.

All surprise, however, appears to be informational. Our brains react strongly, and we learn.

Which brings me to sitcoms.

Why are funny things funny?

Funny things are funny, at least in part, because humor -- humor that works -- is surprising. If it's not surprising, it's not funny.

And that means humor tells your brain you've made a reward prediction error. The punchline of a good joke or gag is unexpected, so dopamine spikes up. Dopamine spikes feel good, so we come back for more.

A good sitcom doesn't need an immersive setting or loud music or arresting imagery to hold our attention.

A good sitcom sets the reward prediction errors coming one on the heels of another, and that is plenty.

Eureka, Part 4 t/k

Flipping the Classroom: Hot, Hot, Hot
MOOCs grow the gap
The New York Times is surprised
In the world of MOOCs, 2+2 is never 4
World's funniest joke: humor depends on surprise
Dick Van Dyke on comedy
Philip Keller on the flipped classroom
If students could talk
Who wants flipped classrooms? (Salman Khan on liberating teachers)
True story
Are math & science lectures boring in a way humanities & social science lectures are not?

10 comments:

Anonymous said...

http://www.pbs.org/newshour/businessdesk/2013/12/dan-ariely-at-61k-a-year-college-is-a-bargain.html

Dan Ariely on MOOC.
He finds 'flipped classroom' is as effective as live classroom but MOOC is much worse.

I like the idea viewing the college as 4 yrs of vacation. hmm...

KimK

Catherine Johnson said...

I like Dan Ariely, but he is way wrong on flipped classrooms.

Or is going to be.

Here's an example of what I take to be classroom flipping in my district:

http://www.edline.net/files/_uRJld_/6665c4e74316e1803745a49013852ec4/Part_1_Ions.mp4

Students aren't going to be learning about Ionic and Metallic Bonding from this.

Catherine Johnson said...

And I did what is called "reverse classroom," so I asked the students to watch the video at home and then come to class to discuss it. It was incredibly successful.

That's Ariely.

What he's doing here, very likely, is recording a lecture he's given many times, which means a lecture that's been honed and shaped in response to direct feedback from a live audience. (Or he may have had himself filmed delivering a lecture.)

Strict lecture is strict lecture; it's going to be better in a live hall (for most) but the wording & pacing aren't going to be different.

Moreover, having students watch the lecture in their dorms allowed him to run the discussion sections instead of having his T.A.s run the sections: students had a full professor heading the precept.

The only college I know of where the actual professor runs the discussion sections is Princeton.

Last but not least, Dan Airely has students at the top of the bell curve, not the middle and not the bottom.

What he's doing isn't what public schools are doing when they flip the classroom.

Catherine Johnson said...

Interesting.

Here are the student ratings:

So the students in my regular class basically had three versions: they had me in person; they had watched the video and come in to have just the discussion; and they watched the video and had the discussion online. And they basically rated them in that order. They said it was the most useful to have me in class. Not too far from that is to watch the video of the material and then have the discussion in class. Much less appealing was to have the video and then have an online discussion on that.

Assuming I'm reading that correctly, students preferred the live lecture and were willing to sacrifice a discussion section run by their professor in order to have live lecture.

Auntie Ann said...

Dick Van Dyke had that figured out a long time ago: http://www.dailymotion.com/video/x10l10e_the-dick-van-dyke-show-father-of-the-week_fun

The main part starts at about 16 minutes in (and there are some ads, unfortunately).

SteveH said...

"What he's doing isn't what public schools are doing when they flip the classroom."

What schools do when they "flip" does not define the term. If you flip a traditional math classroom, you expect that students would view the introduction to the next unit at home and then do their homework individually in class - with the teacher there to help.

In the unflipped case, less motivated students will not do the homework, or they will do it superficially just so they won't get a zero on it. Some will try to do the homework and immediately get stuck, and there will be no help available. The review of the homework in class will be superficial and they will have learned very little. Then, when the teacher introduces new material, they might not pay much attention anyways. For a flipped traditional class, a teacher can force more skill development out of them in class. It's not the best solution, but it can be better. It will show kids that they really can do math. That is more meaningful excitement than what schools want to do now. Educational thought is now driven by engagement and motivation - things they hope will excite students to do the hard mastery work. It's the idea that learning is natural. It's backwards. I've likened this to eating mathematical Twinkies in the past. Kids get excited, but an hour later it's gone. I can give a talk to students about computer graphics and games, but it will not generate the self-control and determination to do all of the un-fun things they have to do to get to that career. If, however, teachers start in the earliest grades and they don't signal to the students that mastery of basic skills automatically leads to "kill", they might find that many kids love the feeling of mastery. My son loved to show me that he could do the math sheets I left out. Structuring education to induce unexpectedly good feelings might be a good goal, but those good feelings have to have a lasting effect. You can't just look at "active learning" in class and then declare the job done.


Clearly, MOOCs are not an ideal learning environment, but what's the problem you are trying to solve? What tradeoffs are you looking at? MOOCs versus what? If people want MOOCs to replace live teachers in a traditional classroom setting, then they have a really low opinion of the profession. Can a MOOC look students in their eyes and decide to slow down or change course? YouTube videos and MOOCs are such trivial uses of technology. A computer program that works with individual students should be better. Unfortunately, the learning programs I've seen have been really bad. I'm sure, however, that some are working in promising areas. I've thought that learning programs might provide a way for schools to point to 21st century learning and skills, but then allow individual students to accelerate at their own pace. That would be hugely exciting for many kids. It's tracking that doesn't look like tracking. The downside is that there is less direct teacher involvement. Given that full inclusion in our schools will never go away, it might be a better solution, but it's not better than a well prepared teacher leading a class where everyone is on the same page.

Catherine Johnson said...

B.F. Skinner invented programmed instruction in the mid-twentieth century & it worked. It was boring, but it worked.

Keller invented personalized instruction for classrooms, and that worked, too. I took two Keller classes in college.

In a Keller class you get whatever grade you work for.

Catherine Johnson said...

But, of course, people like Bill Gates & Salman Khan and Sebastian Thrun and Arne Duncan can't be bothered to read history.

Catherine Johnson said...

Auntie Ann - thank you!

Can't wait to watch!

Michael Weiss said...

I may not be fully appreciating what you are saying here, but "reward prediction error" sounds like nothing new; Piaget described it (not its neurochemical basis, of course) as the basic mechanism underlying all learning. Piagetian epistemology in 6 words: Learning happens when something unexpected occurs.