Machine Learning

From Rules to Learning

Traditional programming follows a simple model: a human writes explicit rules, and the computer follows them. If you want to detect spam, you write rules like ‘if the email contains the word “lottery,” mark it as spam.’ This works for simple cases, but it breaks down when patterns are too complex or too numerous for a human to enumerate [1].

Machine learning flips this model. Instead of writing rules, you provide examples — lots of them — and the computer discovers the rules itself. Computer scientist Tom Mitchell formalized this in 1997: ‘A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E’ [2].

In plain language: if you want a computer to get better at a task, you do not rewrite the code — you give it more data.

The Three Main Types of Learning

Machine learning generally falls into three categories [1]:

Supervised learning is like a student with a teacher. The model learns from labeled examples — inputs paired with known answers. Show it thousands of emails already marked as ‘spam’ or ‘not spam,’ and it learns the patterns associated with each label. Common tasks: classification (is this spam?) and regression (what will the price be?) [1].

Unsupervised learning is the teacher-leave-the-room approach. The model analyzes unlabeled data to discover hidden structures on its own. It might examine a database of customer purchases and discover that people who buy coffee also tend to buy filters. Common tasks: clustering (grouping similar items) and association (finding relationships) [1].

Reinforcement learning is trial and error. An agent interacts with an environment, receives rewards or penalties, and learns to maximize long-term reward — much like training a pet with treats [3]. This is the approach behind game-playing AI and robotics [3].

Deep Learning: The Powerhouse Subset

Deep learning is a specialized, high-performance subset of machine learning that uses neural networks with many layers to process information [4]. While traditional machine learning often requires humans to manually select and transform input variables (a process called feature engineering), deep learning automates much of this, allowing it to handle incredibly complex data like images, audio, and text [4].

Every major large language model — GPT-4, Claude, Gemini, Llama 3 — is a deep learning system [4].

How LLMs Use Machine Learning

Large language models are trained using a form of self-supervised learning: the model plays a massive game of ‘predict the next token’ across trillions of words of text. It learns grammar, facts, reasoning patterns, and world knowledge — not from labeled examples, but from the structure of language itself [4].

After this pre-training, models are aligned with human preferences using reinforcement learning from human feedback (RLHF) — a technique where human reviewers rank model outputs and the model learns to produce responses humans prefer [3].

Connection to the Agentic Economy

Machine learning is the engine behind the agentic economy. When an AI agent reads an email, decides which tool to call, or negotiates a transaction, it is using machine learning models for perception, decision-making, and action selection [5]. The quality of the agent’s ML models directly determines the quality of its decisions [5].

Key Challenges

Generalization is the ultimate goal — translating strong performance during training into accurate results in the real world [1]. A model that memorizes training data without learning underlying patterns is said to be overfitting [1].

Bias is a persistent risk. Machine learning is only as good as the data it consumes. If training data contains human biases, the model will reproduce them — a classic case of ‘garbage in, garbage out’ [1].

Data hunger is a practical constraint. Modern deep learning models require enormous datasets and computational resources, which concentrates power in organizations that can afford them [1].