Can AI be deterministic?
Some products are deterministic – meaning they will only do exactly what they are approved to do….either by an authorised user or a developer. They might ‘learn’ and make suggestions (“It looks like the logical next step is this”) but it requires a user to authorise. One such product is IPSoft’s Amelia Cognitive agent, which we’ve been working with on one client engagement.
The alternative is probabilistic – the AI engine determines the most likely outcome. Then often uses reinforcement learning (“did the user click what I expected”) to improve its algorithms. This is the more common approach for AI applications, for example within machine learning.
Does deterministic have a role? Maybe if the AI engine is automatically sending updates to a back-end software system in a corporate environment, such as an ERP or finance system. But probabilistic AI is what most of us think of when considering AI