CONCEPTS & TURNING POINTS / 1 MIN READ
Algorithms Underlying AI
On this page
AI’s capabilities rest on a family of algorithms from symbolic search to neural optimization. This sub-article expands Section 3 of Hardware & Software.
Symbolic Methods
- Search (A*, heuristic search, constraint solving).
- Logic programming and planning (STRIPS-style).
Classical Machine Learning
- Linear models, decision trees, support vector machines, boosting, clustering, Bayesian inference.
Neural Architectures
- CNNs for vision.
- RNNs/LSTMs for sequences (largely superseded by Transformers).
- Transformers using attention for language, vision, and multimodal data.
- Diffusion models for generative image/video.
- Reinforcement learning for decision-making and control.
Optimization
- Gradient descent and variants (Adam, SGD).
- Evolutionary algorithms and Bayesian optimization for black-box tuning.
Why It Matters
The algorithm determines what a system can learn and how. Paradigms shift as data and hardware change.
Sources
- Russell, S. & Norvig, P. (2021). Artificial Intelligence: A Modern Approach.
- Goodfellow, I. et al. (2016). Deep Learning.
Back to Hardware & Software.