“Artificial intelligence” is the umbrella term for computer systems that simulate human thinking. Machine learning is a subcategory of AI that uses algorithms to process and learn from data. As a result, machine learning systems can make decisions without being specifically programmed to do so. Deep learning uses neural networks that mimic the activity of neurons in the human brain to reach conclusions without human intervention.

Neural networks use “layers” of algorithms to simulate this process. The first layer receives the input data. The algorithm processes it and sends what it has “learned” to the next layer. This procedure is repeated through all of the remaining layers. The output is a prediction of the target result.

For example, a deep learning model might be tasked with distinguishing pictures of cats from pictures of dogs. In supervised learning, the input data would include labels identifying cats and dogs. In unsupervised learning, the input data would not have labels. The algorithm would have to identify patterns in the images on its own. During the training process, the algorithm’s output is used to evaluate the model’s success.

Training a deep learning model requires a massive amount of data. OpenAI’s GPT-4 language model, which powers ChatGPT-4, has 175 billion parameters and was trained using 570GB of textual data. Powerful computer chips are needed to handle the computational intensity of AI applications. Software developers must learn how to write code that can take full advantage of the capabilities of these chips.