top of page

School assessments in the AI age


You can't outrun Gen-AI, but well designed assessments can still be authentic and valid.

Can you blame us teachers for wishing GPT would just go the way of COVID lock-downs? After decades of ratcheting up expectations, a receding tide of support, and shrinking time to recover, teachers are understandably feeling a raft of GPT-triggered emotions. From exhaustion to hope, excitement to dread, confidence to disempowerment - and for some of us it's all of those in the same day!


Paradoxically, the pathway to good gen-AI assessments is not at all that new. It is the same pathway to good assessments; period. It was 1988 when Grant Wiggins and Jay McTighe first developed Understanding by Design. It might be old, but it encourages better assessments because it helps teachers to put the end, first. Here it is in three steps:

  1. Articulate the knowledge, understanding and skills this unit is seeking to develop. It will be a hard wrestle, so allow good time for this step.

  2. Consider the evidence you could collect to measure student attainment of the above. (two aspects to this)

  3. Plan the learning and instruction sequence.

  4. [I am adding 4. Make the resources. Have you even noticed how tempting it is to start with making cool resources - yep, I have done that!]

I think all teachers have experienced units of work where the first job has not been done very well (or even at all). Then, sometime towards the end of the assigned teaching weeks, the process of bartering what is going to be "on the test" begins. We know that rarely represents the finest teaching.


That initial grappling with the key learning we are aiming for, leading to explicit and clear outcomes, helps enormously with assessment - and especially assessment in the world of AI. It allows assessment to perform its FIRST function; to inform subsequent teaching and learning. Only with the knowledge of where our students are on the novice-to-expert continuum, is it possible to implement Load Reduction Instruction (LRI). LRI is possibly one of the best evidenced teaching strategies, but it relies heavily on assessing-as-we-go to know when to "take the bumpers off" to prevent expert-reversal effect (I can't recommend this short pamphlet enough) .


Clarity about the intended learning targets also empowers assessment to perform its SECOND critical function in learning; affirming the student's role in the learning process through accountability.

Here's where the matrix meets reality, because this is the type of assessment (summative) where gen-AI has most disrupted our usual operations. If summative assessment is more about recognising student's progression towards the intended goals and less about deciding high-stakes outcomes like (grading classes or ranking students), we can think more creatively around how we might allow students to demonstrate their progress. Opportunities for students to demonstrate their mastery alone, in an isolated environment like a test, will probably always be appropriate. However, if we reduce all our learning targets to those a one-size-fits-all-pencil-and-paper-test can evidence, then I suggest we have not set ourselves learning goals worthy of the young minds we are nurturing.


There is a danger of gen-AI (and our responses to it) pushing the pendulum too far towards tests alone, that we no longer value the skills of collaboration, critical thinking, communication, innovation, creativity, competence with modern tools, etc. We might say we do, but our students know that what we truly value - we assess. "Will it be on the report, Sir?"

Here are some thoughts as you grapple with the gnarly issue of authentic project-style assessment in this AI age:

  1. Students generally want to learn. Start with this assumption and design for the great, awesome majority of our students who do.

  2. Focus designing your assessment instrument to collect the evidence you detailed in step 2 of your Understanding by Design planning. It might just be different from previous years.

  3. No assessment tool has ever been better than just, "OK". Resign yourself to the graininess of any instrument to measure understanding and don't let your school pretend they have more precision than they do (see point 7 below).

  4. Offer your students choice in a project-assessment. (autonomy is one of the big 3 which contribute to happy humans).

  5. Allow them to be creative. Students are more likely to outsource (to AI) the effort on tasks they find boring busywork. We can detect those task easily if we take an honest look. Design a task that excites or interests most of your students. Fun is a great safeguard against disengagement.

  6. Given them an authentic audience. No one likes creating for a vacuum. Authentic is something like research on zoo animals that zoo keepers would be the audience for.(see my favourite project, Zoo of Poo -you are welcome to join us). Don't forget your students' parents as a potential audience.

  7. Think about reducing the "high-stakes"ness of your assessments tasks. Do you need to rank students from 1 to 223? Are there other ways of ability grouping without ranking next year's classes? My experience is that the learning cost of ranking (students and/or classes) is huge. Do an honest opportunity-cost analysis, think of how assessment could be different if it didn't have to serve that purpose.

  8. Model appropriate use of gen-AI yourself as a teacher. Share with them where and how you use it, from time to time. This makes #9 more natural.

  9. Be explicit: On the task instructions, outline what is and is not appropriate gen-AI use. Don't worry that you can't be a perfect police officer. Teaching is always a tension between being an administrator and inspiring young minds.

  10. Where you do need to authenticate that it's that student's work, think about the power of conversation (see my other post on this).

  11. If you need an "authentication test" it might have these 3 questions; (a) explain your [project] to a student in Year [n-3], (b) in doing your [project] you encountered [challenges/ odd results/ things which didn't work as you wanted], outline how you were able to pivot or learn from these and describe some of your responses to these challenges, (c) Which part of your [project] are you most proud of and why?

Change is hard at first, messy in the middle and gorgeous at the end - Robin Sharma


Our students want us along with them as their world shifts. Many are likely anxious about the world of work that awaits them in this new age (or doesn't await them). You, their teacher, are their lighthouse and their anchor. Teachers have a special and unique bond with young people that no one else in the world shares. Have enough of a dabble with AI yourself so that you can laugh and learn, together.

Roger Kennett




53 views0 comments

Recent Posts

See All

Comments


bottom of page