AI Ethics and Responsible AI

import numpy as np
import pandas as pd
from sklearn.linear_model import LogisticRegression

rng = np.random.default_rng(0)
n = 2000
area = rng.choice(["city", "village"], size=n, p=[0.6, 0.4])
marks = rng.normal(75, 10, n).clip(35, 100)
income = rng.normal(4, 1.5, n).clip(0.5, 12)                    # lakh per year
# historical (biased) decisions: same marks, but villages approved less often
logit = 0.25 * (marks - 75) - 0.4 * (income - 4) + np.where(area == "city", 0.8, -0.8)
approved = rng.random(n) < 1 / (1 + np.exp(-logit))
df = pd.DataFrame({"area": area, "marks": marks.round(), "income": income.round(1), "approved": approved})

X = pd.get_dummies(df[["marks", "income", "area"]], drop_first=True)   # area_village column
model = LogisticRegression(max_iter=1000).fit(X, df["approved"])
df["predicted"] = model.predict(X)

rates = df.groupby("area")[["approved", "predicted"]].mean().round(2)
print(rates)
print("weight on area_village:", round(model.coef_[0][list(X.columns).index("area_village")], 2))

import numpy as np
import pandas as pd

rng = np.random.default_rng(0)
n = 2000
area = rng.choice(["city", "village"], size=n, p=[0.6, 0.4])
marks = rng.normal(75, 10, n).clip(35, 100)

def report(name, predicted):
    df = pd.DataFrame({"area": area, "top": marks >= 85, "pred": predicted})
    sel = df.groupby("area")["pred"].mean()
    tpr = df[df["top"]].groupby("area")["pred"].mean()          # among strong students
    print(f"{name}")
    print(f"  selection rate  city {sel['city']:.2f} | village {sel['village']:.2f} | gap {sel['city'] - sel['village']:.2f}")
    print(f"  top students OK city {tpr['city']:.2f} | village {tpr['village']:.2f} | gap {tpr['city'] - tpr['village']:.2f}")

biased = (marks + np.where(area == "city", 8, -8)) >= 80     # a rule that adds points for city
fair = marks >= 80                                          # the same cut-off for everyone
report("biased rule", biased)
report("same cut-off for all", fair)

import hashlib
import re

note = ("Parent Ramesh Kumar (9829012345, ramesh.k@gmail.com) says Aadhaar 2345 6789 0123 "
        "was submitted. Address: 14 Johari Bazaar, Jaipur 302003.")

PATTERNS = {
    "PHONE": r"\b[6-9]\d{9}\b",
    "EMAIL": r"\b[\w.+-]+@[\w-]+\.[\w.]+\b",
    "AADHAAR": r"\b\d{4}\s?\d{4}\s?\d{4}\b",
    "PIN": r"\b[1-9]\d{5}\b",
}

def mask(text):
    for label, pattern in PATTERNS.items():
        text = re.sub(pattern, f"[{label}]", text)
    return text

def pseudonym(name, salt="school-2026"):
    return "P-" + hashlib.sha256((salt + name).encode()).hexdigest()[:8]

print(mask(note))
print("Ramesh Kumar →", pseudonym("Ramesh Kumar"))

import pandas as pd
from sklearn.datasets import load_breast_cancer
from sklearn.ensemble import RandomForestClassifier
from sklearn.inspection import permutation_importance
from sklearn.model_selection import train_test_split

data = load_breast_cancer(as_frame=True)
X_tr, X_te, y_tr, y_te = train_test_split(data.data, data.target, random_state=0, stratify=data.target)
model = RandomForestClassifier(n_estimators=200, random_state=0).fit(X_tr, y_tr)

imp = permutation_importance(model, X_te, y_te, n_repeats=10, random_state=0)
top = pd.Series(imp.importances_mean, index=X_te.columns).sort_values(ascending=False).head(5)
print("test accuracy:", round(model.score(X_te, y_te), 3))
print("features the model relies on most:")
print(top.round(4).to_string())