Fundamentals of Artificial Intelligence
Core concepts to master
Candidates must be fluent in the vocabulary that separates AI, machine learning, and deep learning: AI is the broad goal of machine intelligence, ML is the subset that learns patterns from data, and deep learning is ML using multi-layer neural networks. Know the classic problem types (classification, regression, clustering, ranking, generation) and how they map to supervised, unsupervised, self-supervised, and reinforcement learning. Understand the reasoning traditions too: symbolic/rule-based systems, search, and knowledge representation versus statistical learning. Be able to explain key architectural families at a conceptual level - convolutional networks for spatial data, recurrent and transformer models for sequences, and generative models such as GANs and diffusion - without needing mathematical derivations.
Traps that appear in questions
A frequent trap is conflating capability with understanding: a model that produces fluent text is not thereby reasoning or grounded. Another is overclaiming that deep learning always beats classical methods; on small, structured, or tabular data, simpler models often win on cost and interpretability. Watch for questions that treat "AI" and "generative AI" as synonyms, or that assume reinforcement learning requires labeled data (it requires reward signals, not labels). Also be precise about training versus inference and about the difference between a model's parameters and its hyperparameters.
Worked example / rule of thumb
Rule of thumb: classify the task before naming the technique. If the question describes labeled historical outcomes, think supervised learning; if it describes grouping unlabeled records, think clustering; if it describes an agent improving through feedback, think reinforcement learning. Example: a bank predicting loan default from past labeled loans is supervised classification, not clustering, even though both are "AI."