History and Background

Expert Systems

Trying to explain artificial intelligence decision-making is not a new effort. XAI got it's start in the 1980s with the advancement of Expert Systems and the explosion of personal computers being used in offices. These expert systems were trained on a variety of logic rules in order to make decisions, a very simple form of artificial intelligence (Wick and Thompson, 1992). 

It was concluded that expert systems could be improved if the computer could relay its decision making process to the user, but the search is still on for a good explanation (Swartout and Paris). Expert systems are still in use today, though rarely, because the hierarchical framework of the expert system logic is a bit limited. Today, the things that most resemble expert systems, are of course, artificial intelligence models. But there's a wide range of models, most of which are nothing at all like the strict, rule-based logic of the Expert System. The work in creating explanations for expert systems went on for a very long time (Wick and Thompson), and current AI models are vastly more complicated.

XAI is also part of a broader movement to make machine learning more user-focused: human-in-the-loop machine learning. While a bit of a long name, "human in the loop" is definitely not a misnomer– it's exactly what it sounds like. The subset of the industry focuses on bringing humans into the process of machine learning, whether that's by allowing them to be part of the iteration and training process, or something focused on how humans will use machine learning models, like eXplainable AI. HILMIL has a lot of subcategories, such as Usable AI, Active Learning, Interactive Machine Learning, Machine Teaching, and Curriculum Learning. All of them somehow involve increasing human participation in the machine-learning process, and all are deeply rooted in the Human-Computer Interaction field.

PREMISE:   (AND (SAME CNTXT GRAM GRAMNEG)
                    (SAME CNTXT MORPH ROD)
                    (SAME CNTXT AIR ANAEROBIC))
ACTION:     (CONCLUDE CNTXT IDENTITY BACTEROIDES TALLY .6) 

IF:                 (1) The gram stain of the organism is gramneg,
                     (2) The morphology of the organism is rod, and
                     (3) The aerobicity of the organism is anaerobic
THEN:          There is suggestive evidence (.6) that
                     The identity of the organism is bacteroides