Commanding AI Success by Obeying Causality

This paper is primarily about analyzing how causality is interwoven into AI processing leading to its success, rather than about any particular AI implementation, in order to show how AI success can be commanded by obeying causal laws. In contrast, the traditional approach to achieving AI success focuses on data collection, with the belief that collecting more and/or better data is the key to AI success. Unlike the traditional approach, this paper makes the case that AI succeeds by leveraging causally informative information—which the traditional approach can do, but only coincidentally. Commanding AI success is accomplished by following a process (presented in this paper) that leads to causal discovery and leveraging the causally informative information that results from this discovery. In addition, with each new discovery the number of domains in which AI success can be commanded expands, virtually removing any barrier to commanding AI success in any domain. Commanding AI success in every application is ideal, but sometimes due to a lack of domain understanding striving for coincidental success is the only viable option even though this option comes with its own risks. A valid method of causal discovery is relatively novel, and unless the reasoning behind this method is demonstrated to others, they might not be able to sufficiently distinguish between AI developed on this causal basis and AI developed on the traditional approach. This paper aims to show the importance of this distinction, and how to easily identify and explain it.