Keep the user in control
Provide controls so users can easily customise what the AI system monitors and how it behaves
Current AI products often have long wait times before providing a response to a single prompt or action. Let users halt current processes and edit their inputs.
If possible, it may be useful to allow users to go to a previous version of a model. Since models don’t always “unlearn” new pieces of information easily, simply halting an interaction may not bring users back to their desired state.
Let users give feedback
Give users the opportunity for real-time feedback and, if possible, teaching and error correction. Make sure you have a process in place to use this feedback to improve the product. Let them know how you’re going to use this and how it benefits them.
Allowing users to provide corrections (not just general feedback) to AI-generated results that are then reincorporated (human-in-the-loop), can help both you and the model better understand what areas of the user experience can best be improved.
Shape possible interactions to increase positive outcomes
Make sure the standard version of your product is powerful enough for most users, providing guide rails to help users “fall” into interactions that are highly likely to produce favourable results. Allow users to dismiss or ignore undesired AI features or output, by providing a way to move forward even when the system fails or offers poor quality output. Provide them with enough context to understand what went wrong and how they can continue.
Depending on your product, you may want to provide more advanced users with more controls so that they can personalise model behaviour to their needs.