It's all related to how artificial intelligence can be trained. Typical software follows pre-defined instructions and decision paths, whereas AI uses machines trained to achieve outcomes using data instead of upfront rules.
In most cases where AI is used, it’s extremely difficult, if not impossible, to identify any or all of the decisions and decision path combinations in advance.
AI-based software, or engines, can continuously evolve and improve over time as they learn with more data. But for AI engines to be useful, they need to be trained with the right kind of data and by the right knowledgeable individuals in that specific discipline.
Chatbots are a good example of continuous learning in AI. Consider how chatbots now provide much more accurate results than they did when they first launched. AI enabled chatbots learn from the questions people ask, then provide more accurate results in response.