Inference
Inference
The process of feeding new data into a trained AI model to generate predictions, judgments, or other outputs.
In Simple Terms
Inference is what happens when an AI that has finished training receives new data and predicts or judges an answer from it. For example, it's what's running when your smartphone camera recognizes something in a photo as a "cat," or when a translation app converts a foreign language into Japanese. Whenever AI actually does something useful in your everyday life, this inference step is what's making it happen.
Behind the Name
Inference is the standard English term here. The word carries the sense of "reasoning from existing clues or evidence" or "drawing a conclusion" — and that's exactly how it's used in AI, too. The model draws on everything it learned during training to work out a likely answer for whatever new problem lands in front of it.
Take a Closer Look!
Inference means giving new information to an AI that has already been thoroughly trained, and having it produce a result — a prediction or a judgment. Think of it like studying hard from a textbook and then sitting down to actually take the test.
Running an AI involves two major steps: "training," the preparation phase, and "inference," the real thing. During training, the model is drilled on huge amounts of data until it learns how to arrive at the correct answer. Once it's smart enough, inference is what happens when that trained AI looks at a photo or a piece of text a user actually sends it and works out "what is this?"
This inference step is what's running whenever a smartphone photo app automatically sorts pictures of dogs and landscapes, or a translation app converts words from one language to another. Training requires a powerful computer and a lot of time, but inference draws on knowledge the model already has, so it can produce results with far less computation. Lightweight tasks like recognizing an image can sometimes run directly on a smartphone or computer, but for something like generative AI that handles large volumes of text, it's processed on a powerful server out on the internet — which is why it can take anywhere from a few seconds to tens of seconds for the result to come back.