A machine learning (ML) model, which is the predominant AI method today (chatGPT, Gemini, etc.), comprises a training phase, which is unique in the world of software development.
In contrast to traditional programming, after the AI neural network architecture is designed and programmed, the next step is the training phase, which analyzes the patterns in the data. The training stage can take days, weeks and months in datacenters with from tens to hundreds of thousands of servers operating together. For example, GPT-4 took several months and was trained on trillions of words. See
AI training and
AI model.
Inference Processing
The AI software that does work for the user, essentially the AI application that answers questions or generates content, is called "inference," a term that dates back to the first AI programs (see
expert system). The "inference engine" can be loosely compared to an operating system with the model being the application. Once programmed, it is used to run many models for countless purposes. However, like everything in the computer field, the inference engine must also be programmed and fine-tuned.
Running an inference engine to get a result does not take anywhere near the computer time as the learning phase of model development. However, consider that millions of people may be using an AI chatbot or other application simultaneously and for months on end; thus, inference processing may add up to much more computer time in the long run than the training phases of the models it is using. See
neural network and
AI datacenter.
REGULAR DATA PROCESSING DEVELOPMENT:
1. design the logic
2. code the logic
3. test application
4. run application
--------------------------------------------
AI NEURAL NETWORK DEVELOPMENT:
Model Development
1. INITIAL DESIGN
1a. select network type (CNN, RNN, GAN, etc.)
1b. code the model
1c. set layers, neurons, passes (hyperparameters)
1d. software sets weights and biases (parameters)
2. PRE-TRAIN (with example data sets)
2a. hyperparameters mostly adjusted by people
2b. parameters adjusted by software
3. FINE-TUNE (with example data sets)
3a. hyperparameters mostly adjusted by people
3b. parameters adjusted by software
See AI hyperparameter.
Inference Engine
1. design
2. code
3. optimize (see AI quantization)
Execute AI Application
1. run inference engine with built-in model
or
2. run inference engine and select model
A Clear-cut Comparison
This clever comparison of machine learning programming vs. traditional programming comes from Techopedia's "The Ultimate Guide to Applying AI in Business."