Learn AI · Level 2 · How AI works · step 4 of 12 · word 2 of 7
What is an embedding in AI?
An embedding turns a word, sentence or image into a long list of numbers that captures its meaning, so things with similar meanings end up with similar numbers.
In plain English
Computers can’t work with meaning directly; they work with numbers. An embedding is how AI translates meaning into numbers. Each word, sentence or picture becomes a list of hundreds or thousands of numbers, like coordinates on a huge map of meaning. Things that mean similar things land close together on that map: “puppy” sits near “dog”, and “How do I reset my password?” sits near “I forgot my login”.
The model learns where to put things during training, by noticing which words appear in similar contexts. Nobody hand-writes the coordinates.
Why it matters
Embeddings let AI search by meaning instead of exact words. They are the engine behind “chat with your documents” tools and RAG: your question is turned into an embedding, and the system finds the passages whose embeddings are closest. They also power recommendations (“people who liked this also liked…”), duplicate detection and image search. A whole kind of database, the vector database, exists just to store and search them.
An example
In 2013, Google researchers showed a now-famous result with word embeddings: take the numbers for “king”, subtract “man”, add “woman”, and the closest word to the result is “queen”. The numbers had captured something about meaning that nobody had programmed in.
What to watch out for
Embeddings learn from human writing, so they also absorb its stereotypes: researchers found that early word embeddings linked some jobs more closely with men than with women. And “close in meaning” is not the same as “correct”: a search can return a passage that sounds relevant but answers a different question. Good systems check the results, and so should you.
See them in action with the Embedding Projector in our AI Playground.