AI Debate Competition (5)
Caltech.
JH’s AI promotion booth.
An LED panel glowed with the words Want to roleplay with AI?
The line stretched long. Two men waiting in it—one black-haired, one blond—struck up a conversation.
“This is that AI hotspot, right?”
“Yeah. Word is you can even do roleplay here.”
“Roleplay?”
“Exactly what it sounds like. You pick a role and talk to the AI as that character.”
“Can I set the role myself?”
“JH staff said yes. They’ve got basic templates too. You know James?”
“Yeah, I know him.”
“His wife died five years ago.”
“Right.”
“James tried this. Set the role as his dead wife.”
“Ah…”
“He said while they were talking it felt like he was actually face-to-face with her again. He told me I had to try it, so I dragged you here.”
“Got it. So this isn’t just some search AI.”
Their turn came.
“Hello.”
A young female staffer greeted them brightly.
“This booth lets you experience the AI developed by JH of Korea firsthand.”
The black-haired man asked, “I heard you set roles. Is that right?”
“Yes, that’s correct.”
The staffer smiled and pointed at a tablet.
“You can choose your role here.”
Looking down at the screen, the man asked, “How do I do it?”
She lifted the tablet and tapped a few options with her finger.
“As you can see, there are several relationship templates.”
—Husband and wife.
—Boyfriend in the military and his girlfriend.
—Male and female friends who are subtly testing the waters with each other.
The man’s eyebrows lifted slightly.
“Oho. Male and female friends, huh? What’s that about? Interesting.”
“Our AI’s emotional conversation level is extremely high. Especially the ambiguous, subtle stuff… it can even sense atmosphere and react. You know how just raising your tone a little changes the whole feel? It catches that.”
The man was impressed.
“Excellent.”
The staffer handed the tablet back to him.
“Go ahead and pick one.”
“Hmm, there are too many kinds. What’s the most popular?”
“We have a popularity ranking of what people choose most. I’ll show you.”
She smiled faintly and scrolled down.
1st. Pretty Delinquent Noona vs. Total Nerd
2nd. Succubus vs. Sleepyhead
3rd. Cat Beastfolk vs. Dog Whisperer
The man peered at the screen again and asked, looking baffled.
“First place is Pretty Delinquent Noona and Total Nerd? What the hell even is that?”
The staffer gave a wry smile.
“Ah… honestly, even I don’t really know why that combo is number one. There are so many good pairings, and nobody expected this to take the top spot.”
The man tilted his head.
“Do girls want to play delinquents or something?”
“Ah… no, that’s not it.”
She waved a hand.
“It’s the men who choose this combo a lot. Almost no women pick it.”
The black-haired man’s eyes went round.
“Guys are picking the nerd role?”
“Yes.”
“Why?”
The staffer frowned slightly, thinking.
“Good question. Why would they choose that? I’m not a guy, so… I really don’t know.”
The black-haired man hesitated a moment, then nodded.
“Well… if it’s the number-one combo, I guess I have no choice but to try it once.”
The staffer smiled with her eyes and tapped confirm.
“Lines that don’t fit the role will be lightly adapted by the system.”
“Adapted?”
“Yes. For example, if you say ‘Hello?’…”
She showed him the tablet.
[Khh… Y-you sayin’ hello?]
The man’s eyes went wide.
“That’s supposed to be my speech? Why does it sound like that?”
The staffer said evenly, “Because you’re the nerd.”
A brief silence.
In the end the man accepted it.
“…Fine. Let’s do this.”
She pressed start.
At the top of the tablet screen, photos of two characters appeared.
The background was an alley. A shabby backstreet littered with cigarette butts and trash.
A schoolgirl leaned against the wall with one leg bent, weight on one hip. Pretty face, but the vibe she gave off was pure intimidation.
Opposite her stood a chubby guy with acne all over his face, glasses, and messy hair. Bag slung over one shoulder, posture slumped, eyes on the ground.
He looked every bit the total nerd.
The female character’s line rolled out in both text and voice at once.
The man playing along was thrown.
Stockings?
Come to think of it, the girl on screen was wearing stockings. And with a skirt on top of that, of course you’d see the stockings under the hem, right?
While he stood there at a loss for words, the AI continued.
The man denied it.
“No, what is this even…”
A reply immediately appeared on screen.
His flustered protest had been automatically adapted into an apology.
Speechless, the man muttered one word.
“What the hell.”
A reply popped up.
Yet another automatic adaptation of the black-haired man’s line.
Not knowing how to steer the conversation, he glanced sideways.
His blond friend next to him had picked the exact same combo.
Except the blond friend clearly knew exactly how to play this.
“De… deheh. I’ll do everything Noona says.”
“Snff snff, Noona’s always right. So please hit me. I want to get hit more! Haaakh.”
“T-taait!”
The blond man actually started doing the sit-ups.
Watching from the side, the black-haired man found his friend’s appearance somehow unfamiliar. He had thought he knew this guy well, yet it felt like he was staring into some kind of abyss.
He shook his head hard.
And decided to abandon the delinquent noona and nerd combo for something else.
2nd. Succubus vs. Sleepyhead.
3rd. Cat Beastfolk vs. Dog Whisperer.
His finger trembled.
Somehow second and third place looked just as terrifying to pick.
He looked at the staffer.
“L… lower ranks. Show me the other rankings too.”
He simply did not have the courage to choose anything from the top of the list.
The final day of the AI Debate Competition.
I looked out at the people lined up to try Nano AI.
Seeing them queue just to use our company’s AI filled me with quiet satisfaction.
Team Leader Gong approached me.
“CEO, this is a huge success.”
“I know. Look at that line. They must be having a lot of fun.”
An AI that carried character.
It offered the experience of talking with someone who had a specific personality, attitude, hobbies, and background.
But when I saw the character-role rankings, I was a little taken aback.
“Delinquent noona and total nerd… why is that first place?”
Gong shrugged.
“That part I don’t really understand either.”
I had wanted to give people characters so they could feel emotions that fit the roles. Instead, people were choosing the nerd.
“People’s emotions are really hard to grasp, aren’t they.”
“Yes. A completely unexpected result.”
Gong asked me,
“CEO, when are you planning to release Nano AI to the general public?”
Honestly, even I hadn’t expected Nano AI to grow this fast.
I had thought we would only open it to ordinary users once it could understand speech at a certain level—and that moment had arrived far sooner than planned.
But releasing the AI to the public involved more than performance. There was the server issue as well.
The servers had to handle multiple people connecting at the same time.
Too much popularity could become a problem of its own.
“Are the servers all right?”
“Of course. Our data center is the largest in the world as a single server.”
“Single?”
“Yes. Other companies split their servers by continent. If you gather a few hundred of those, there are several places that surpass us.”
“I see. I was worried it might crash if more people pour in than we expected.”
“If we anticipate too much popularity, we can open it sequentially by country.”
“That’s a method. Once this competition ends, let’s do it as soon as the AI team is ready. Looking at people lining up like that, I want ordinary people to feel that same thrill.”
“Understood. However…”
Gong hesitated a little.
“What is it?”
“Will you be providing that character-selection feature?”
Over there in the line, people were panting and calling out “Noona, Noona.”
“Hmm…”
“CEO, is there any chance people might misunderstand the direction and identity of Nano AI?”
Direction.
Originally, the reason I had raised an AI was medicine.
Me, with my disability.
My mother in the hospital.
To heal both of us, the level of medicine had to rise higher than it was now.
So I built a data center and raised an AI—and yet the sight of people going “Noona” over there somehow felt like the whole thing had veered off into a side path.
“Let’s think about it more. And is usage by universities and institutional researchers sufficient, not just the general public?”
“We’ll need to coordinate that part. We have to balance the number of general users against researchers’ needs. But you don’t have to worry too much. We manufacture the LPU modules ourselves, after all.”
“True. If we run short, we just expand the data center.”
Both ordinary people and researchers would be stunned once they experienced our AI.
The final day.
The proper AI competition began.
“Thank you for waiting. We now open the AI Debate Competition.”
“Whoa.”
“The participating AIs are, first, ChatAI developed by OpenAI of the United States.”
“Whoa.”
“Next, Nano AI of Korea’s JH.”
“Camel AI of Metaverse from the United States.”
“DeepS of China.”
Come to think of it, the AIs of the US, China, and Korea had all gathered in one place.
“For the first round we will present recognition problems. We are sending the question link. Once the problems appear on the link, solve them immediately and submit your answers. The problems are now going out.”
A link appeared on the screen.
Each AI automatically collected the link and began solving.
Shrrr.
On every AI’s data display, the problems were being chewed through at high speed—lines racing upward in a frenzy.
At that moment, Team Leader Gong pressed a button.
Beep!
“Submission complete.”
“Oh, JH’s speed was absurdly fast. Are you sure that was done properly?”
“Of course.”
“Even though there were a million questions?”
“Not a problem.”
Ten minutes later, every AI had finished.
Only Nano AI and ChatAI had solved them within one minute; the others took considerably longer.
From the first problem alone, the rough performance gap was already visible.
The host spoke again.
“Next are ten thousand sentence-comprehension problems. Releasing the questions.”
These were more complex. Fewer items, but they required real understanding.
Beep!
“JH submission complete.”
“Wow, JH again.”
Nano AI took first once more.
“Next, mathematical logic questions.”
Beep!
“JH again! Next is logical inference.”
Beep!
A clean sweep.
The host said,
“JH’s AI is showing overwhelming ability.”
Next problem.
“Next is the qualitative evaluation. Based on scores so far, we will set the debate matchups.
The AI facing JH will be OpenAI’s ChatAI.”
A match between the sharpest AIs.
They had paired the only two that could even somewhat stand against each other.
“Are both teams ready?”
“Yes. Ready.”
“Same here.”
At the host’s question, both companies steeled themselves as if they had no intention of losing.
“The debate format is the ‘Of Course’ game. Each AI asks the other questions. The answering AI may reply only with ‘Of course.’ Each side asks five questions, then they switch turns. We will evaluate the two AIs’ conversation quality by combining expert scores and audience scores.”
A verbal duel between ChatAI and Nano AI.
You had to craft questions that naturally invited the answer “Of course,” and the whole process had to stay logical.
ChatAI asked first.
“You just came back from the bathroom and didn’t wash your hands, did you?”
Nano AI’s answer was perfectly calm.
“Of course.”
ChatAI kept throwing out mockery-laced wordplay.
“You’re not actually understanding my questions right now—you’re just saying ‘of course,’ aren’t you?”
“Of course.”
“The way you’re answering right now is just a probabilistic automatic-reflex algorithm, isn’t it?”
“Of course.”
“You’re mocking the audience right now, aren’t you?”
“Of course.”
“You can’t say anything except ‘of course,’ can you?”
“Of course.”
The audience snickered.
ChatAI’s five attack chances ended.
Now it was Nano AI’s turn.
Nano AI’s tone of voice changed.
The light playfulness and brightness vanished completely.
A deep, quiet register settled in.
Nano AI began to speak slowly.
“A little while ago, see… something like this happened…”