[BIRD CHIRPING] [FASTER CHIRPING]
JESSE GOLDBERG: Hi. My name is Jesse Goldberg, and I'm an Assistant Professor in the Department of Neurobiology and Behavior at Cornell University. And my lab is interested in the neural mechanisms of trial-and-error learning.
So imagine you're setting out to play a piano piece, and let's say the piano piece is A, B, C, D. Well, you might, when you're a beginner, sort of bang on the keys randomly and go ABC. And you might make a mistake and say A, B, C, B. Well, you wanted to hear D, but you heard B.
So how does the brain say, "Oh, that was a mistake?" And this is an especially difficult question because there's nothing intrinsically good or bad about the sound of B. It's just not what you wanted to hear at the fourth time step of the song when you wanted to hear D.
Currently, much of what we know about the neural mechanisms of trial-and-error learning comes from studies of animals learning to do simple tasks for food or juice rewards, and what we know from these studies is that the chemical dopamine is important for signaling rewarding outcomes. So for example, if an animal is very thirsty, at the moment it gets delicious juice, its dopamine neurons are strongly activated. And what we wanted to know is-- do these dopamine-related mechanisms of learning also apply to behaviors that aren't learned for extra rewards but rather something like playing a piano where you're trying to compare your performance outcomes to personal goals? And so to address this question, my postdoc, Vikram Gadagkar, undertook some experiments in the songbird.
VIKRAM GADAGKAR: The songbird is a great model system to study the neural mechanisms of natural trial-and-error learning for two reasons. First, baby zebra finches learn to sing by listening to a tutor's song, storing the tutor's song as a memory trace or template, and then spending the next two or three months learning to imitate the tutor's song through trial and error. The key here is that they don't learn to sing for immediate external rewards, like food or juice, but instead, learn to sing by comparing their own song to an internal template.
Second, the zebra finch brain contains many of the same brain regions that our brain does, including a dopaminergic projection to the basal ganglia. So the critical experiment here, to test if dopamine activity is involved in performance evaluation, is to record from these dopamine neurons in a singing bird while fooling the bird into thinking that it sang a crappy song. Now, to do this, we use a little trick called "distorted auditory feedback" where we take one part of the bird's song and unexpectedly play it over different parts of the bird's song while the bird is singing.
So here is a young adult male zebra finch. You can see that we have the two speakers on either side that we use to provide the feedback. There's a little microphone at the back of the box that records the bird's song. The whole box is soundproofed so that the bird is not disturbed and can sing in peace.
Now, let's say the bird's song has four syllables-- A, B, C, D-- which sounds like "chick-a-wa-ah," and the different part of the song that I'm going to play over syllable D sounds like "tch." So the birds singing along-- "chick-a-wa-ah, chick-a-wa-ah, chick-a-wa-ah, chick-a-wa-ah"-- but what the bird hears is "chick-a-wa-ah chick-a-wa-ah chick-a-wa-tch-wa-tch chick-a-wa-tch-wa-tch-wa-tch-wa-tch-wa-ah" depending on whether or not we distort syllable D with feedback.
So you can see here a spectrogram of the bird's song with the various syllables-- A, B, C, D, E, F, and so on. And you can also see that syllable D has been distorted with feedback. Now, if you look below, at the neural trace, each line here is an action potential of the dopamine neuron. And you can see that, immediately after feedback, there is a long suppression in this neuron. The idea is that a distorted syllable sounds back to the bird whereas an undistorted syllable sounds just fine.
We discovered that dopamine neurons are suppressed immediately after the started syllables indicating a worse-than-expected performance. Now, remarkably, these same neurons are also activated at the precise time in the song when an expected distortion might have occurred but did not. So in contrast, now we have the undistorted song.
Once again, we can see the various syllables of the song-- A, B, C, D, and so on-- but notice now that syllable D has been left undistorted. If you look at the corresponding neural trace for this case, however, we see that, immediately after syllable D, there is an activity increase in the dopamine neuron-- indicating a better-than-expected performance. Now, what is so amazing about these results is that they show that dopamine activity signals better or worse performance in exactly the same way that we know that it signals a reward or lack of reward in other animals.
JESSE GOLDBERG: So the big idea is that, when the songbird practicing his song hits the right note, his dopamine neurons are activated in an eerily similar way to how they are when a thirsty animal gets juice. And so maybe it's the case that, when we say the word "rewarding" colloquially to refer to our own performance-- for example, "It was so rewarding when I hit that good tennis shot," or "when I executed that piano piece perfectly"-- maybe when we say the word "rewarding," we're saying that because it's the precise same brain mechanism by which we signal rewards like getting juice when you're thirsty or pizza when you're hungry or getting coffee first thing in the morning when that's what you're craving. And so it might be the case that there's really general mechanisms by which the brain processes rewarding outcomes for satisfying hedonistic drives as when executing performances.
We've received your request
You will be notified by email when the transcript and captions are available. The process may take up to 5 business days. Please contact email@example.com if you have any questions about this request.
Many performances are learned through trial and error by matching performance to internal goals. Young birds learn to sing by comparing their song with an internal template. Recording from basal ganglia dopamine neurons in singing zebra finches showed that dopamine activity was suppressed after distorted syllables but was increased at the precise moment of the song when a predicted distortion did not occur. Thus dopaminergic error signals can evaluate behaviors that are not learned for reward but instead are learned by matching performance outcomes to internal goals.