I have a list of statements that make up a definition. In this case it's for healthcare so its attributes about a person. Currently, we have a checkbox to allow someone with the knowledge to define the truth of a statement.

[ ] Patient is 65 years old or older

This gives me a "true" or "unset".


Users can skip over the unset state as something they don't care about. However, now we're adding an AI element that parses background information and will return a known false, a known true or an unknown/unset.

[ ] Patient is 65 years old or older - AI "No" patient's date of birth is 50 years before today.

The catch is that sometimes the AI gets it wrong so a human needs to be able to override the choice.


What's the best way to display all these states and still allow the user to override it? I doubt any users will actually set these to strictly false but they should still get the option. Users WILL frequently override the AI's unknown with a true.

I'm considering the ability to cycle between states for each click and then creating checkbox states that look negative, look neutral and look positive.

cycle of checkbox states: from 'unset' to 'set' (green) to 'negative' (red) and back to 'unset'

Is that intuitive? I expect users will receive training on this system but I wonder if there's a better way to go about it? Anyone have examples of this done well? I considered a "dropdown" or select style interaction but that actually seems more mentally complex.

  • So, when the users come, or are presented with this attribute list, it's been run through the AI, so is the users job to scan to see if there's results that differ from the AI suggested output? And how many attributes are there?
    – Mike M
    Apr 7, 2020 at 23:52
  • Correct the items have already been processed. However we're not very good yet so probably 50% are still unset (ai didn't know with enough certainty). The length varies between 3 items and maybe 50 but average is 7 - 10 items. Apr 8, 2020 at 0:17
  • I know you have an answer to this already, but consider making the age question a mandatory DOB field. The patient might be 64 today and 65 next week, and the system can track this better than the user creating the profile should the patient return 2 weeks from now.
    – Izquierdo
    Aug 11, 2020 at 14:57
  • @StacyH All data is added to the system by the Electronic Medical Record (patient's source of truth). This UI is intended to allow a user to check the "results" of an AI. DOB is just an easy example as most statements are way more complex. Aug 11, 2020 at 15:03

1 Answer 1


Using the 'checkbox' style will confuse a known interaction: that of a single selection checkbox being binary.

The control is standing in for an icon as well, and without a label, icons can confuse users.

If you have a grid of items, you can have an assessment row w/ select controls, and it's only 2 clicks to change an item, and the label is persistent and visible:

enter image description here

You have this way:

  • A selection control signifying what's editable
  • An icon and color change for two visual distinctions
  • An explicit label indicating the current status
  • Yeah, the select list seems like a good option. The layout already has a column header to make it easier to scan. Apr 8, 2020 at 15:21
  • Mike, may I ask: which tool did you use drawing this nice UI-mock?
    – hc_dev
    Apr 18, 2020 at 9:07
  • 1
    this is balsamiq. I'm using the desktop tool, but it's built in here when you post or answer. There's a little 'UI Wireframe' icon built into the posting editor, right next to the Add Image icon. That will launch a browser version of balsamiq that will post to your answer.
    – Mike M
    Apr 18, 2020 at 11:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.