There is a charming term used in natural language processing called ‘perplexity’. It measures the strength of the prediction made by a model. It is related to the average number of branches offered at certain point. When there are a lot of choices offered, all of them with similar ‘weight’, the more perplexing it is for the user. It is like the waiter at the restaurant that tells you that everything on the menu is really good.
For an example, see the predictions below. (Based on ‘Intuition of Perplexity’ class material from Dan Jurasfky, Stanford University.)
I always order a pizza with cheese and ______
- Mushrooms (10%) probability
- Pepperoni (10%) probability
- Anchovies (1% probability)
- Fried rice (0.01% probability)
I saw a ______
- Cat (0.01% probability)
- Bird (0.01% probability)
- Crash (0.01% probability)
This natural language processing concept made me realize why certain user interfaces are more perplexing than others: the abundance of like choices.
See the screens below. Perplexing. They have about 30 – 45 choices; all of them with a ‘similar’ probability of being selected.
The average branching in a game of chess has been computed to be 31-35. So yes, navigating interfaces like the ones above can be a little bit of a chess game.
So, what does a low perplexity UI look like? Below is an example I like of how a low perplexity user interface could be realized.
In the past, I heard someone refer to a software we worked on as ‘obvious’ to use; because at every given screen, it was clear what to do next. This stuck with me. ‘Obvious to use’ (instead of easy to use) is a subtle but powerful user interface attribute. When I design user interfaces I make sure that there are few choices at any given time; obvious choices. Creating obvious to use interfaces takes precedence over nice graphics and fancy interactions. If the core concepts are not obvious to use, a sophisticated and beautiful UI design is only lipstick on a pig.