Machine learning is a method in which a system optimizes its outputs based on given data, an objective function, and evaluation criteria. In the provided context, many systems commonly referred to today as “generative AI” are described as output systems that rely on machine learning, especially large deep learning models trained on large datasets. These systems are often top-down in character: a human supplies the intended direction, such as what kind of output is desired, and the model generates results to fit that goal. The context also notes that such systems frequently become black boxes, because their internal representations and decision processes are difficult for humans to interpret.
Machine learning is not limited to general AI applications. It is also being incorporated into research on cellular automata. For example, in neural cellular automata, the local update rules of a cellular automaton are represented by a neural network, and the network is trained by backpropagation to achieve specific goals such as morphogenesis or texture synthesis. In this sense, machine learning functions as a technique for acquiring behavior from data and objectives rather than explicitly hand-designing rules.
The context further suggests that the boundary between AI and artificial life is becoming less clear, since machine learning can be used to learn local rules that produce emergent global behavior.