ABSTRACT
While there is no “silver bullet” approach to authoring AI behavior, behavior trees tend to strike a strong overall balance across ease of implementation, ease of visualization, and adoptability for new team members. Building supporting tools is straightforward, and a rapid workflow can be established in relatively short order. On the other hand, behavior trees have a fundamental limitation. They are poor at modeling analog concepts such as uncertainty over multiple valid options. Game characters are simply fun and engaging machines for players to interact with and even exploit, but requirements still often demand more than strictly Boolean selection logic. Because hand-coding analog selection logic everywhere that it is required gets messy quickly, a better solution is needed.
