Teaching AI Emotions
After the lecture, a dinner with the university faculty followed.
The president of Korean Peninsula University served as host.
"CEO of JH, thank you so much for today."
The president bowed deeply at the waist.
I smiled and returned the courtesy.
A hundred billion won.
On paper it looked like a spectacularly generous donation, but we had arranged to expense it in installments over ten years. The boost to the company's image was a free bonus. And when you considered the global firms already lining up just to secure our semiconductors, it wasn't even a sum that deserved to be called burdensome.
Even selling only a fraction of the LPUs and HBM we produced—keeping the bulk for our own data centers—our operating margins were already absurd.
Besides, since I was already on campus, there was something else I wanted to ask for while I was here.
The president smoothly passed the microphone my way.
"It's a pleasure to meet so many of you. Today I delivered a lecture and handed over a scholarship fund—but I have one more thing prepared."
At the mention of something more, the professors' eyes lit up.
"Right now, JH is training an AI. We've finished the recognition stage and are moving into comprehension. What that means is this: when it sees an apple, it correctly identifies it as an apple."
"Remarkable."
"Of course, even that alone is impressive. But it still has more to learn. At JH's current level, the AI can manage things like, Should I hand someone an apple? I'm not really craving an apple today."
"Oh."
"The hardest part, though, is sentences like this."
I looked at the president and posed a question.
"Let me give you an example. Mr. President—suppose your wife walks through the door and says, 'Honey, I was depressed today, so I bought thirty loaves of bread.' How would you answer?"
"Thirty loaves of bread? I'd probably ask what kind of bread she bought thirty of."
Scattered sighs broke out around the table.
A woman professor sitting beside the president asked,
"Good heavens. Mr. President, are you a T?"
"Excuse me? Why am I a T?"
"She said she was depressed and bought bread."
"So what does being depressed have to do with buying bread?"
"Haa…"
I looked around the room and continued.
"As you can see, even humans struggle with emotion. How much harder must it be for an AI? If an AI truly understood emotion, how should it answer the bread example just now?"
A few professors offered replies.
"Refined carbohydrates spike blood glucose sharply between roughly twenty and forty minutes after intake, which temporarily increases midbrain limbic dopamine release. So the correct answer is: depression seeking dopamine?"
"You can frame it that way, but I think this case is better read as a learned reinforcement pattern."
"Ah—felt depression, carbohydrates, dopamine reward, temporary buffering of the low mood. Repeat that experience enough times and the brain builds an automated schema: depressed equals bread. That's what you mean?"
Leave it to a room full of professors to keep piling on the technical language.
I spoke.
"There are many possible answers to 'I was depressed, so I bought thirty loaves of bread'—but you could also say this."
"Say what?"
"Want some coffee? Or milk?"
The president looked completely lost.
But the woman professor beside him lit up.
"Oh my. That is exactly my style."
I glanced around the table again.
"Between 'I was depressed, so I bought thirty loaves of bread' and 'Want some coffee?' there has to be a subtle grasp of emotion. Frankly, it's a hard problem even for people. For an AI, it's harder still."
"Hmm…"
"There are professors here from a great many departments. So let me make a proposal, right now, at this table. I am recruiting research teams across multiple fields for a project to teach emotion to AI. The research budget will be more than generous."
"Ohhh—a project!"
Teaching emotion to AI.
It was never going to be a problem for the computer science department alone.
Professors from every corner of the table found themselves interested.
"If it's teaching emotion, that's the humanities."
"Emotion is foundational to psychology. You can't do emotion research without psychology in the first place."
"Well, CS is the obvious baseline, of course."
I smiled at the assembled faculty.
"I'd like all of you involved. AI wants every kind of knowledge."
Psychology department seminar room.
The psychology professor looked over the hundred-odd students gathered there.
"Thank you all for coming."
"Not at all, Professor. We're happy to have the part-time work."
The project had pulled in not only psychology graduate students but undergraduates as well.
"This project is a joint effort with JH. The topic is… teaching emotion to AI."
For a moment the seminar room went quiet.
One student carefully raised a hand.
"Professor… is that even possible?"
The professor smiled.
"Probably not."
Students glanced at one another.
A particularly curious one asked,
"Is there a reason it would be impossible?"
The professor continued.
"To teach emotion, we first have to talk about what emotion is. Emotion is an interpretation of the body's response. In other words, emotion begins with the body."
A few upperclassmen nodded, but the underclassmen didn't quite follow.
"The body?"
"Imagine meeting a wild boar on a mountain trail, or suddenly seeing an enormous snake. Confronting a criminal. Watching a terrifying scene… Your heart races first."
A student asked,
"So you're saying it's not that we feel fear and then our hearts pound?"
"Fear comes later. First the cerebrum assesses the situation, judges it a crisis, and a region called the amygdala activates. Only then does the heart begin to race."
"Ahh—so it reads the situation and sounds an emergency siren?"
"Exactly. Once the siren goes off, a person interprets that bodily response. The interpreted emotion might be fear—or it might be rage."
A soft murmur of understanding passed through the students.
"Ah, so that's why the same situation can produce different emotions? Or sometimes no emotion at all?"
"Right. Depending on the circumstances, someone may not judge it an emergency at all—and then no emotion arises. And even when it is an emergency, whether a person chooses to attack, avoid, or accept varies from person to person."
"So the interpretation of the body's response is the emotion."
The professor paused, then pointed at the AI training simulation running on the monitor.
"But AI has no body. No racing heart. No muscle tension. No change in temperature. There is no bodily response for it to interpret—and therefore no emotion to interpret either."
Another student asked,
"So if it just had a body, would that be enough?"
"No."
The professor's expression grew serious.
"Emotion requires metacognition. I am happy right now. I am anxious right now. There has to be a process of recognizing that for yourself. You have to judge what situation you're in before the body can respond to it."
The students nodded.
"On top of that, AI has no goals and no desires. It only executes what it's given. It doesn't even have death."
By this point the students were starting to wonder why the professor had called them in at all.
"Professor, then how do we teach emotion to AI?"
The professor smiled faintly.
"Actually, psychology has already been running a similar project—for people who struggle to understand emotion. There is a method even they can use."
"What do you do?"
"You memorize. And memorization is something AI is extremely good at."
"Ahh…"
"Even without full empathy, you can still memorize the 'correct answer patterns.' When people have weak emotional understanding, this is often the first approach they take."
"I see."
"Even if you don't yet grasp the real meaning, you memorize a set of correct answers: in this situation, this is the right thing to say. And you already know a program like that."
The professor smiled.
"It's better known as… understanding romance through text."
"Pfft."
A ripple of laughter ran through the students.
The professor brought up a slide on the projector and showed them a problem.
"Shall we try one?"
Choose the most appropriate response.
"Oppa, do I look different today?"
1) I don't know?
2) Hang on. Let me compare a photo from yesterday.
3) Did you do your hair? It really suits you today.
4) Yeah. You look uglier than yesterday.
The professor told the students,
"The answer is number three. From an emotional standpoint, that's correct. Want to try another?"
"Oppa, it feels like we've been talking less lately."
1) What's your definition of "lately"?
2) I've had a lot of assignments. Let's talk plenty today.
3) Whoa, there's five hundred won on the floor.
4) So what, then?
"I saw this and thought of you, so I bought it."
1) I already have one at home.
2) How much was it?
3) Thanks. Looking at it will probably make me think of you even more.
4) Did I tell you to save money, or did I not?
The professor asked,
"Well? Manageable?"
Students answered over one another.
"Yeah. For both of those the answer is four."
"No it isn't? Both are one."
"What? Are you serious?"
"Yeah. So what, then."
The professor quieted them down.
"The reason I gathered all of you is to mass-produce situations like these. Even if AI never truly feels emotion, if we feed it enough scenarios it will learn the patterns."
"Yes."
"Good. Then each of you, bring me a hundred problems."
"Yes."
Department of Theater and Film.
The theater and film professor addressed the students.
"We've agreed to join JH's project on teaching AI."
At the word AI, the theater students looked puzzled.
"AI? Isn't that something computer science should be doing?"
"Computer science is involved, of course. But the topic is teaching emotion to AI. Do you think CS students will convey emotion well—or will you?"
"Ah. Obviously us."
"Exactly. How do you express emotion well? That has been our department's longstanding question. And as a theater professor, I have always been asking myself how emotion can be taught well. That's why we signed on to this AI training project. Take a look at the screen for a moment."
The professor played a video.
A familiar drama scene filled the display.
A young woman weeping; a man with his face twisted in a scowl.
Through her sobs, the woman said,
"Hic… hic… How could you…"
"I had no choice. Loving someone isn't a crime."
Two lines of dialogue were enough.
The husband's affair.
The wife's grief.
Bang.
The door burst open and a middle-aged woman stormed in.
She seized the man by the collar.
"Are you even human?! How—how could you do this to my daughter?!"
"Please stop."
The mother-in-law's fury.
The man's irritation.
The mother-in-law rummaged through the living-room refrigerator and returned with a large plastic container. From it she pulled out a whole head of kimchi.
The man flinched.
Surely she wasn't about to throw that—he edged backward.
"Hey—put that down."
But the mother-in-law had no intention of putting the kimchi down. With the kimchi clutched in her right hand, she grabbed her son-in-law's collar with her left.
"No—let go of me."
That was the moment.
Sensing her chance, the mother-in-law swung her right arm in a full arc.
Slap!
The long head of kimchi cracked like a whip, and the man took a spectacular kimchi slap across the face.
Kimchi brine splattered across the entire living room.
"Ah…"
Doused in red liquid, the man was left speechless.
Still not satisfied, the mother-in-law spat venom.
"Huff… huff… Considering what you did, this is nothing. You deserve far worse!"
The young woman watching all of this.
Still crying, she pulled out a young radish and held it out to her mother.
"Mom… hic… hic…"
The mother-in-law's revenge.
The man's flustered shock.
The young woman's despair and sorrow.
And yet—somehow—a comic scene.
Were you supposed to laugh or cry?
It was a short clip, but several intense emotions had cut through it in rapid succession.
Looking at the screen, the theater professor asked,
"Well? Familiar, right?"
"Yeah. It's a trashy melodrama."
"What emotion do you feel from the young woman?"
"Sadness?"
"Why does she feel that?"
"Because her husband betrayed her."
"And the middle-aged woman?"
"Anger. A desire for revenge."
"Reason?"
"Because her daughter was betrayed."
The professor smiled.
"Excellent. What we just did—watching a drama, then explaining what each character did and why—that is exactly what you'll do for the AI."
"Ah."
"You've done this a hundred times, haven't you?"
"Yes."
"Then each of you, pick one drama. Watch it with the AI and explain every character's emotions in every scene. You can do that, right?"
"Of course."
Department of English Literature.
"Oh—to be, or not to be: that is the question."
Department of Korean Language Education.
"The speech of this country being different from that of China…"
Across many universities.
Across many departments.
They taught the AI with everything they had.