2026 - ISTQB AI Testing (CT-AI) Certification - Crash Course

Explore what artificial intelligence is and how machine learning enables systems to learn from data, distinguishing rule-based approaches from data-driven predictions in real-time tasks like self-driving cars and chatbots.

Understand how machine learning lets systems learn from data and solve problems without explicit programming. Learn how this fits into artificial intelligence and enables spam filters, ads, and predictive maintenance.

Explore how machine learning models like ChatGPT, Gemini, and Copilot process input to analyze data and predict the output, and learn about the machine learning life cycle and QA involvement.

Understand the machine learning model life cycle from data preparation, model engineering, and evaluation in offline mode to production in online mode, with functional and non-functional testing and monitoring.

Explore the offline mode of the machine learning life cycle, including data preparation, model training, and testing before production. Understand training and validation data, features, labels, and overfitting risks.

Learn how underfitting, good fit, and overfitting arise across training, validation and unseen data, and how iterative data stages, QA testing, and model tuning prevent degradation before production.

Learn how a house price model uses features like size, bedrooms, and age, trained on data, with linear regression. Quality assurance evaluates train-test splits and performance.

Explore overfitting and underfitting in supervised learning, identify signs like high training accuracy vs test performance, and learn remedies such as data expansion, regularization, cross-validation, and feature enhancement.

Explore unsupervised learning, where models train on input data without labeled outputs, forming clusters that reveal patterns in demographics and locations for targeted ads and sales insights.

Explore unsupervised learning with cluster analysis and SIL scores to assess fit, overfitting, and underfitting, and apply white-box testing and validation techniques.

Explore supervised, unsupervised, and reinforcement learning, showing how models learn by interacting with environments, receiving rewards or penalties, with robots, chess, and autonomous driving as core examples.

Trace AI's evolution from narrow AI to general AI and the theoretical super AI, with examples like spam filters, image classifiers, and large language models like ChatGPT.

Explore the differences between conventional rule-based systems and AI, including supervised and unsupervised learning, pattern recognition, and popular machine learning frameworks and algorithms used to train and deploy models.

Understand how hardware, from CPUs to GPUs and specialized AI chips like TPUs, supports training, while AI as a service via cloud platforms offers trained models with benefits and risks.

Differentiate hosted ai as a service from pre-trained models, noting cloud api access, lack of customization, versus models you can fine-tune, and emphasize bias, transparency, and responsible ai testing.

Explore flexibility and adaptability in AI systems, including handling out-of-scope situations and adapting to new environments. Assess autonomy with tests of independent operation and timely handovers.

Explore evolution and self-learning in ai models, including stock-prediction contexts, highlighting learning from past decisions and environmental changes, and ensure ethical, value-aligned adaptation through testing.

Identify and mitigate bias and ethics in AI by examining training data fairness and algorithm bias. Ensure transparent testing, accountability, and respect for human rights and laws in AI outputs.

Design goals carefully to minimize side effects and prevent reward hacking; ensure transparency, interpretability, explainability, and safety through testing for loopholes and clear data sources.

Explore the three forms of machine learning—supervised, unsupervised, and reinforcement—learning from labeled data, clustering unlabeled data, and training agents via rewards, with real-world examples.

Define objectives and select a learning framework: supervised, unsupervised, reinforcement. Choose an algorithm, prepare high-quality training, validation, and test data, train and tune, then evaluate and monitor drift post deployment.

Explore the ML data preparation pipeline—from collecting and cleaning raw data to converting into embeddings, augmentation, and feature engineering—and examine dataset quality issues like mislabeling, imbalance, and privacy concerns.

Explore how data labeling works in supervised and unsupervised settings, including internal, outsourced, crowdsourced, and hybrid approaches, and how to detect overfitting and underfitting with a validation data set.

Evaluate AI model quality using the confusion matrix, defining true positives, false positives, false negatives, and true negatives, and apply metrics suited to classification, regression, and clustering.

Learn how accuracy, precision, recall, and F1 score are calculated from TP, FP, FN, and TN via the confusion matrix, with spam filter and x-ray detection examples guiding metric choice.

Evaluate AI models using recall, precision, F1 score, and accuracy, considering false positives and false negatives, data set balance, confusion matrix, and ROC and AUC thresholds.

Explore regression metrics like mean squared error and R-squared, and clustering metrics such as silhouette coefficient, ROC AUC, with guidance on thresholds and model fit.

Recognize that metrics alone cannot judge model quality; test bias, fairness, transparency, explainability, and performance under load, then pick precision, recall, or F1 based on data balance and business needs.

Discover how neural networks learn from massive data, using input, hidden, and output layers with interconnected neurons, weights, and activation thresholds to predict outputs like spam detection.

Explore how a deep neural network uses input neurons, hidden layers, weights, and bias to compute activation values and produce restaurant recommendations, refined through training data.

Review weights and bias as core neural network parameters that shape activation values, and explain how activation functions and back propagation enable learning in this ISTQB AI testing crash course.

Explore how coverage in neural networks differs from traditional code coverage, focusing on neuron coverage and activation thresholds to ensure all neurons respond, with practical spam detector examples.

Explore five neural network coverages, including sign change coverage, value change, sign sign coverage, threshold coverage, and neuron coverage, through practical neuron testing with activation values.