Responsible AI practices in MLOps - Commands & Configuration

Introduction

Responsible AI practices help ensure that AI models are fair, safe, and trustworthy. They guide how to build and deploy AI systems that respect privacy, avoid bias, and provide clear explanations.

When you want to check if your AI model treats all groups fairly before deployment

When you need to track and explain AI decisions to users or regulators

When you want to monitor AI model behavior continuously to catch errors or bias

When you must protect sensitive data used in AI training and predictions

When you want to document AI model development steps for transparency

Commands

This command runs an MLflow project that includes responsible AI checks like fairness and explainability during model training.

Terminal

mlflow run https://github.com/mlflow/mlflow-example.git -P alpha=0.5

Expected Output

2024/06/01 12:00:00 INFO mlflow.projects: === Running command 'python train.py --alpha 0.5' in run with ID '123abc' ===

Training model with alpha=0.5

Logging metrics and artifacts

Run completed successfully

→

-P - Passes parameters to the MLflow project, here setting model hyperparameter alpha

Starts the MLflow tracking UI to visualize model metrics, fairness reports, and explanations collected during runs.

Terminal

mlflow ui

Expected Output

2024/06/01 12:01:00 INFO mlflow.server: Starting MLflow UI at http://127.0.0.1:5000

Serves the trained model with responsible AI features enabled, allowing safe and explainable predictions via REST API.

Terminal

mlflow models serve -m runs:/123abc/model -p 1234

Expected Output

2024/06/01 12:02:00 INFO mlflow.models: Serving model from run 123abc on port 1234

→

-m - Specifies the model path to serve

→

-p - Sets the port number for the model server

Sends a prediction request to the served model and receives an explainability report along with the prediction.

Terminal

curl -X POST http://127.0.0.1:1234/invocations -H 'Content-Type: application/json' -d '{"data": [[5.1, 3.5, 1.4, 0.2]]}'

Expected Output

{"predictions": [0], "explanations": {"feature_importance": [0.8, 0.1, 0.05, 0.05]}}

Key Concept

If you remember nothing else from responsible AI, remember: always track, explain, and monitor your AI models to ensure fairness and trust.

Code Example

MLOps

import mlflow
from mlflow.models.signature import infer_signature
from sklearn.datasets import load_iris
from sklearn.linear_model import LogisticRegression
import pandas as pd

# Load data
iris = load_iris()
X = pd.DataFrame(iris.data, columns=iris.feature_names)
y = iris.target

# Train model
model = LogisticRegression(max_iter=200)
model.fit(X, y)

# Log model with MLflow including signature for input validation
signature = infer_signature(X, model.predict(X))
with mlflow.start_run() as run:
    mlflow.sklearn.log_model(model, "model", signature=signature)
    mlflow.log_param("max_iter", 200)
    mlflow.log_metric("accuracy", model.score(X, y))
print(f"Model logged in run {run.info.run_id}")

Common Mistakes

Ignoring fairness checks during model training

This can lead to biased models that harm certain groups or users.

Include fairness metrics and tests as part of your model training and evaluation pipeline.

Not enabling explainability features when serving models

Users and stakeholders cannot understand or trust AI decisions without explanations.

Serve models with explainability tools that provide feature importance or decision reasons.

Failing to monitor models after deployment

Models can degrade or become biased over time without detection.

Set up continuous monitoring and alerting for model performance and fairness.

Summary

Run MLflow projects that include responsible AI checks during model training.

Use MLflow UI to visualize model fairness, metrics, and explanations.

Serve models with explainability enabled for trustworthy predictions.

Send prediction requests and receive explanations to understand AI decisions.