Artificial Intelligence comes in two major types, heuristic or rule-based, and statistical or evidence-based.
The heuristic approach, which dominated for the first several decades of AI research, depends entirely upon recipes for solving data-processing and decision problems that have been thought out and encoded by human programmers. Its researchers have also historically been attracted to working on problems that lent themselves to symbolic representations, binary notions of truth, and discrete decisions. For the latter reasons, this kind of AI has also been called 'symbolic'. The flowchart may serve as its icon.
The statistical approach, which has gained a tremendous amount of market share in the practice of AI in recent decades, uses optimization techniques to automatically improve the performance of a piece of software, based on evidence present in measurement data. Its researchers have been successful at developing data analysis and control systems that work well with continuous-valued or analog signals, as well as discrete or symbolic data. This kind of AI is also called 'machine learning'. The regression plot could perhaps serve as its icon.
Statistical AI is not just for engineering, but is converging with experimental findings in neuroscience to flesh out a quantitative theory of cognition, efficient learning, and accurate inference. There is a process that our nervous systems undergo during experience to capture models of causal dependency in our environments based on observational and experimental evidence. Understanding the governing principles of this process will allow us to provide deep foundations for improvements in mental health, education, and human skill development, as well as dramatically improved artificial systems.
The statistical perspective also stands to refine our notions of truth, much as quantum theory refined our notions of determinism, and relativity refined our notions of space and time. In so many real-world transactions, most of which our brains manage unconsciously, 'truth' is far less useful than what economists have called 'satisficing', or being accurate in a way that is good enough for the demands of a situation.
The author's affiliation with The MITRE Corporation is provided for identification purposes only, and is not intended to convey or imply MITRE's concurrence with, or support for, the positions, opinions, or viewpoints expressed by the author.