As artificial intelligence (AI) technology advances and AI products become more prevalent and more powerful, the line between machines and humans starts to blur. AI products can perform a variety of tasks from synthesizing data and presenting helpful insights to behaving independently, learning new tasks, and predicting outcomes with astonishing precision. 

Advancements in AI are beginning to change how people react to AI products and what they have come to expect out of a product's design. By following a few basic user experience principles, clarity can be brought to the increasingly complex world of AI products. 

AI products can be used to synthesize vast quantities of data and present aggregate and summary information that can prove to be extremely useful. But algorithms are imperfect.  Design products to let users know when they are viewing AI generated information versus human provided information so that they can decide for themselves whether to trust it.

Understanding confidence levels (or the probability of precision) of AI generated results can aid users in arriving at decision.  For instance, a CRM system may predict which leads are most likely to convert to a sale of a particular product.  But how accurate are such predictions? Presenting a confidence level with such predictions can help the user understand the results.

Many AI products use complex algorithms and machine learning technology.  These are often treated as a black box.  But sometimes understanding how an algorithm generates its results can be very helpful in understanding why a specific decision or result set was produced.  This doesn't mean explaining every aspect of how a complex neural network arrives at a decision, but we can give certain hints to the user about what the technology does and perhaps the inputs used.  Understanding which data is provided as inputs and which weighed most heavily in generating the end result can be helpful. 

Designers and developers are always trying to fine-tune algorithms to prevent bad responses.  But edge cases often arise which force trade-offs. One such example is the difference between optimizing for precision versus optimizing for recall. 

When optimizing a machine learning algorithm for recall the AI product will identify every item that meets the expected criteria even at the risk of identifying some edge cases which may not (false positives). When optimized for precision, a machine learning algorithm will only return results that are absolutely correct at the risk of overlooking some edge cases that would have been accurate but which the algorithm flagged as unclear.