"Mira-san's behavior is always unpredictable."
Yuki murmured.
Mira smiled quietly.
Aoi became interested. "Did you try making a prediction model?"
"No, but somehow. I feel like I can't see a pattern."
"That's an interesting observation," Aoi opened the notebook. "In machine learning, we build prediction models. But not everything can be modeled."
"Things that can't be modeled?"
"Yes. Models learn from past data. But completely new patterns can't be predicted."
Yuki thought. "Mira-san doesn't follow past patterns?"
Mira wrote on paper. "Out of distribution."
Aoi nodded. "Out-of-distribution data. Outside the range of training data."
"Is that a good thing?"
"Depends on perspective," Aoi explained. "In machine learning, out-of-distribution data is a problem. The model loses reliability."
"But for humans?"
"It's called individuality. Unpredictable parts create what makes that person unique."
Yuki wrote in her notebook. "Predictability and individuality are a tradeoff?"
"In a sense. A completely predictable person is easy to model but lacks surprise."
Mira showed another paper. "Surprise is information."
"Exactly," Aoi confirmed. "Information content is determined by degree of unexpectedness. If as predicted, information content is zero."
Yuki asked, "So Mira-san is a high-information person?"
"Maybe. But that's not all."
"What do you mean?"
Aoi drew a diagram. "Prediction models have bias and variance."
"Bias?"
"Bias from model assumptions. If you have a model classifying people as 'morning person' or 'night person,' it can't capture people who don't fit the framework."
"Mira-san is outside that framework?"
"It's possible. A more complex model might be needed."
Mira said slowly, "Models simplify. People are complex."
Aoi nodded deeply. "Models are tools for simplification. But oversimplifying loses the essence."
Yuki understood. "So perfect prediction models can't be made?"
"No. Especially for humans. People learn, change, and act according to context."
"What about machine learning models?"
"With enough data, they can be quite accurate. But they're weak to unknown situations."
Yuki asked, "So are we also unpredictable?"
"Partially. But compared to Riku, Yuki is quite predictable."
Yuki laughed. "Is that a compliment?"
"A neutral observation. Predictability aligns with reliability. But might trade off with creativity."
Mira showed a new note. "Predictable = boring?"
"No," Aoi denied. "Predictability has value too. Stability, reliability. These are important."
Yuki thought. "But a bit of unpredictability is also needed?"
"Balance. Basically predictable, occasionally unexpected. That's comfortable."
"Why?"
"If completely predictable, it's boring. If completely unpredictable, it causes anxiety. Moderate uncertainty maintains interest."
Mira smiled.
Aoi continued. "Information-theoretically, it's about entropy. Too low or too high is bad. Moderate complexity is ideal."
Yuki reviewed her notebook. "So Mira-san's entropy is high?"
"Depends on the observer. Mira herself might have consistent patterns. But we can't see them."
Mira wrote on paper. "Different perspective, different model."
"Profound," Aoi was impressed. "The same person becomes different models depending on viewing angle."
Yuki asked, "So there's no true model?"
"Probably not. Models are always approximations. Reality is more complex than any model."
"But models are useful?"
"Of course. Even imperfect, they're tools for understanding. What matters is knowing a model's limitations."
Mira said slowly, "I am more than any model of me."
Aoi nodded quietly. "That's true for everyone. We are richer than any model."
The three discussed prediction model limitations in the sunset classroom.
People don't fit into formulas or statistics.
That unpredictability might be what makes us human.