Reading

What AI Actually Is (and Isn't)

"Artificial Intelligence" gets used to describe genuinely everything from a simple spam filter to science-fiction robots — a real, working definition matters before anything else in this program makes sense.

A real, working definition

AI is software that performs tasks which would genuinely require human intelligence if a person did them — recognizing a face in a photo, translating a sentence, recommending a genuinely relevant video. Notice this real definition says nothing about how it's done — which is exactly why AI covers such a broad, real range of actual techniques.

AI, machine learning, and deep learning: real, nested categories

AI is the real, broadest category (any system performing a task that seems intelligent). Machine learning is a real, specific approach to AI — systems that improve at a task by learning from real, actual data, rather than being explicitly programmed with fixed rules for every case. Deep learning is a real, specific type of machine learning using structures called neural networks (Week 3), which is what powers most of today's most capable, real AI systems, including large language models (Week 4).

Narrow AI vs. general AI: a real, important distinction

Every real, actual AI system that exists today is narrow AI — genuinely excellent at a specific task or range of tasks, with no real, general understanding beyond it. A chess engine that plays at a superhuman level has genuinely zero ability to hold a conversation. General AI — a real system with human-like, general intelligence across any task — does not currently exist, despite how some real marketing language may imply otherwise.

What a modern AI model is genuinely NOT doing

A large language model does not genuinely "think" or "understand" the way a human does — it's producing statistically likely text based on genuinely enormous amounts of real training data, a mechanism explored fully in Week 4. This distinction matters practically: it explains why a model can write fluently and confidently about something factually wrong — fluency and genuine accuracy are real, different things this technology doesn't guarantee together.

A real, brief, honest history

Real AI research genuinely dates back to the 1950s, with real, alternating periods of rapid progress and "AI winters" (periods when funding and interest collapsed after real, earlier hype outpaced actual results). The current, real wave of progress — driven substantially by deep learning and, more recently, large language models — began accelerating genuinely around the early 2010s, with a real, dramatic jump in public visibility once tools like ChatGPT launched in late 2022.

Real categories of AI you'll actually encounter

Computer vision — real systems that interpret images and video (facial recognition, medical image analysis).
Natural language processing — real systems that work with human language (translation, the large language models this program covers in depth).
Recommendation systems — real systems learning your preferences from behavior (what you actually watch, buy, or click).
Robotics — real AI controlling physical machines that sense and act in the real world.

Why "AI" as a real term can be genuinely misleading

Because "AI" covers such a real, broad range of actual techniques, a genuinely useful habit going forward is asking, specifically: what type of AI, doing what specific task, trained on what real data? A vague claim that a product "uses AI" tells you genuinely little on its own.

This week's practical habit

List five real AI-powered tools or features you've encountered this week (a spam filter, a photo app's "search by object" feature, a music recommendation) and, for each, name which real category above it most likely falls into.