Core Principles of Ethical AI

The Five Pillars of Ethical AI

Principle	Definition	Engineering Implication
Fairness	Equal treatment across demographic groups	Test model performance by subgroup; fairness metrics
Accountability	Clear responsibility for AI decisions	Audit trails, model cards, AI oversight roles
Transparency	Understandable how decisions are made	SHAP, LIME, decision logs, system cards
Privacy	Data used only as authorized	Data minimization, consent, differential privacy
Safety	Prevent harm during deployment	Red-teaming, adversarial testing, kill switches

import pandas as pd
import numpy as np
from sklearn.metrics import confusion_matrix

def fairness_audit(y_true, y_pred, sensitive_attribute, group_names=None):
    df = pd.DataFrame({"y_true": y_true, "y_pred": y_pred, "group": sensitive_attribute})
    results = []
    for g in df["group"].unique():
        mask = df["group"] == g
        yt, yp = df.loc[mask, "y_true"], df.loc[mask, "y_pred"]
        tn, fp, fn, tp = confusion_matrix(yt, yp, labels=[0, 1]).ravel()
        n = len(yt)
        results.append({
            "group": group_names[g] if group_names else g,
            "n": n,
            "positive_rate": (tp + fp) / n,
            "tpr": tp / (tp + fn) if (tp+fn) > 0 else 0,
            "fpr": fp / (fp + tn) if (fp+tn) > 0 else 0,
            "accuracy": (tp + tn) / n,
        })
    audit_df = pd.DataFrame(results)
    print(audit_df.to_string(index=False, float_format="{:.3f}".format))
    for metric in ["positive_rate", "tpr", "fpr"]:
        gap = audit_df[metric].max() - audit_df[metric].min()
        print(f"{metric} disparity: {gap:.3f} {'⚠️' if gap > 0.1 else '✅'}")
    return audit_df

np.random.seed(42)
n = 1000
y_true = np.random.randint(0, 2, n)
y_pred = y_true.copy()
group  = np.random.randint(0, 2, n)
bias_mask = (group == 1) & (y_pred == 1)
y_pred[bias_mask & (np.random.rand(n) < 0.3)] = 0
fairness_audit(y_true, y_pred, group, {0: "Group A", 1: "Group B"})