What Is AI “decision-making”?
Step 1: Training
Before an AI can make any decisions, it goes through training, much like a person learning a new skill. Imagine you want to teach a friend to recognise different types of birds. You’d show them photos of various birds. You’d point out characteristics, like colours, beak shapes, and sizes so they start to recognise patterns.
In AI, this training is done by feeding the model a large amount of data with labels (“this is a robin,” “this is a sparrow”). The AI then creates “rules” based on patterns in the data it sees. These rules aren’t like if-then statements we might use to make simple decisions. Instead, they’re more like pathways of probability that lead to different outcomes, created through algorithms.
Step 2: Inference
Once trained, the AI model moves to inference, or the decision-making phase. Think of it like your friend at the park, confidently identifying birds based on what they’ve learned. But instead of birds, AI is often recognising faces, translating languages, or deciding if an email is spam. Here’s where things get interesting, the AI uses what it learned during training to make a prediction about new data. In the same way that if we gave our bird-watching friend a new photo, they’d analyse it based on what they know about robins, sparrows, and so forth, and decide which bird it is most likely to be.
An AI will “infer” based on past patterns and predict the most probable answer. In AI, this happens through mathematical functions. It looks at the data, applies its learned patterns, and calculates which outcome is most likely. If it’s a recommendation engine, it might calculate the probability you’d like a certain movie based on others you’ve enjoyed. If it’s a self-driving car, it’s calculating the likelihood that an object in the road is a person, another vehicle, or just a shadow.
Step 3: Prediction
After inference, we get the result we actually see. For example, if you’re using an AI-powered photo app that sorts your photos by people, the app’s prediction might be, “This is a photo of Sarah.” Or if you’re using a music streaming service, it might predict, “Based on your past songs, you’ll like this new release.” Predictions are essentially educated guesses. Some models predict in a straightforward way, they might output “yes” or “no” or give you a single answer. Others might offer probabilities.
For example, if an AI system detects a dog in a photo, it might predict “dog” with 95% confidence, “cat” with 3% confidence, and “fox” with 2% confidence. Just like humans, AI models make errors. But each prediction is a step in refining the AI’s accuracy. Over time, as it gets more feedback on right and wrong answers, it can improve.
Imagine a chef in a busy restaurant. They’ve spent years learning recipes and can now cook almost any dish without a recipe book. When a customer orders something, the chef doesn’t have to look it up, they use their learned experience to quickly decide which ingredients and techniques to use. Similarly, an AI model trains on data, makes decisions by recognising patterns. Whether it’s recognising text, recommending songs, or interpreting images, the AI is constantly learning and refining, just like the chef perfecting each dish.
