Quizizz Is Great, But I Would Only Use Its AI To Assess Super, Super Simple Student Answers

I have published many posts over the years extolling Quizizz and its many features.

And I absolutely love using its Artificial Intelligence tools to create materials, like interactive videos, online flashcards and review games.  In fact, its so good in these areas that I have hardly ever had to edit what its AI has produced.

Yesterday, I tried out its AI tool for assessing student responses – when you create quizzes with “open-ended” questions, you have the option of having AI assess those answers.

I was decidedly unimpressed.

On the surface, its a well-designed tool.  You can write your own criteria for the AI to use in its assessment, or you can have the AI develop its own criteria, which you can also then edit, if you want.

I created a quiz using simple Aesop Fables, and users had to write down the moral of the fable.

I asked students in one of my IB Theory of Knowledge classes to experiment with it.  All the English-proficient students got a 100%.  None of the English Language Learners in the class, who are all advanced ELLs, got that score.

In reviewing the answers, the ELLs did, indeed, answer the questions correctly, but didn’t use the key words that were in the AI criteria.

I have major ethical and pedagogical reservations about using AI to assess student written work (of course, multiple choice is different), though I’m open to considering its use for very simple written answers in game formats (see How I’m Using The Groovelit Game Now & How I Hope To Use It In The Future).

I’m sure Quizizz will refine its AI in the not-so-distant future so that it, too, can work for simple answers in a game format, and in other ways to make it particularly useful to ELLs (like Groovelit is doing).

If you have fewer reservations than I do about using AI to assess student work, I would strongly recommend you take the time to carefully review student answers.  But keep in mind that it won’t be very energizing for students to initially hear from AI that their correct response is wrong.