It’s time to rethink assessment tasks in the era of AI

This seems like a very simple question, but it is not immediately obvious. In my own pondering over the last year or so, this is what I can boil it down to: 

1. To check the learning of knowledge  
2. To provide practice for skills  

The big problem we are all facing at the moment is that AI can be easily used to by students regurgitate knowledge.

 

As we now know, students can get AI to produce very well-written essays, as it is very good at synthesising information from across the internet because this is what it excels most at.

 

However, oneo of the problems for us as teachers is that, when students do this, they don't usually read carefully what AI writes in these essays, so they aren’t learning anything.   

 

As a result, for many of us teachers, we are quickly learning that assigning long-from essays that are completed in students own time is no longer a true way of assessing their knowledge of a topic.

 

Aside from the fact that the use of AI is a breach of academic integrity (since students are passing off this work as their own) the biggest problem is that students aren’t learning during the completion of their assessment tasks.  

And this is the real problem. Assessment tasks were introduced as a way for students to demonstrate what they have learned in their studies.

 

100 years ago, knowledge exams were a great litmus test for checking their basic knowledge of facts.

 

Then, essays were introduced to allow them to demonstrate that they had invested extended thought into the topic in order to create an overall argument about it.

 

However, since AI can do this in their stead, students can submit an assessment without thinking about the content at all. Students are not learning.  

 

This is where I go back to a core question: why do we teach history to teenagers in high school?   

 

Ultimately, the reason we teach History is because we want students to become critical thinkers in their daily lives: to use their knowledge and skills to respond to opinions they encounter in the world.  

Exams, therefore, are still the best way of assessing knowledge, as it requires students to store the information in their head, as they don’t have access to AI during exam time.  

 

However, a traditional knowledge exam (with multi choice questions and short response) may not be the best or only way to assess it, as it doesn’t make them critical thinkers.

 

Instead, there are other ways we could achieve this:

If we want students to be critical thinkers in real-world scenarios, then that is what we should be trying to assess: the ability for students to use their historical knowledge and skills to response to complicated real-world opinions.

 

For example, in an exam, you could provide students with a section of a real-world news article, blog post, or other opinion piece, and the students are required use their knowledge to respond to the strengths and weaknesses of the opinions presented.

 

As a result, it is the knowledge and the skills that they have in their head that will help them succeed.   

 

In this instance, AI won’t help them on exam day. However, AI could actually become a valuable tool to help them prepare and revise for this task, but it won’t do the work for them in the exam.

As an additional thought, this could also be done in combination with a multi-week research task that culminates in an exam scenario.

 

In this way, they can use AI to complete a traditional research task (using AI to find sources, evaluate them, and even write an essay response), as long as the task requires them to explore two sides of a controversial topic.

 

Once they have completed the essay response, the expectation is that they have read the essay AI produced in order to deeply understand the arguments within.   

 

Then, they are given an opinion piece in an exam (as in the previous paragraph) that requires them to use what they have learnt in the research, to engage with both sides of the debate and reach a conclusion.

 

This way, the students need to have confidence in the knowledge and skills they developed during the research process, and the summary that AI provided, to ensure that it didn't leave anything out in preparation for the unseen article.

 

Alternatively, they could use the same AI-assisted research process, then complete a knowledge exam.   

 

What this would teach the students is that AI can produce information for them, but it is often over-simplified and limited in its key details.

 

Therefore, they must constantly review and improve AI outputs to ensure that they are fully prepared for the exam.   

 

It must be said that students can complete the same task without relying on AI at all.

 

The research process, writing of an essay, and exam revision in preparation for an unseen question, can all be done in the traditional way.

 

The only benefit is that AI can make these processes much quicker.  

 

Either way, the assessment is focusing on the knowledge and skills we want them to develop for the real world.

 

AI is just one tool that helps them achieve this. For most students, this will come from careful reading, critiquing, and remembering what their research essay said.

 

Ultimately, this might be a better test of their knowledge than a traditional ‘conduct some research and write a historical essay’ task.  

 

The same process can be done for a purely skills-based task. They can practice doing their skills in their research tasks (such as a Source Investigation) but then test them under exam conditions.

 

The obvious example here is to do a response-to-sources exam after completing a source investigation.   

 

These assessment tasks could work in tandem to be marked as one assessment task.

 

For example, the Source Investigation could require just two sources, while the third is provided by the teacher in the exam.

 

Alternatively, the students can do the questions and analysis of their sources during the research phase, but then hand-write their critical summary under exam conditions based upon the print-out of the source analysis from their research.  

Another idea is that students could use AI to create their own stimulus sheet of information for an unseen exam question or knowledge exam.

 

Such an exam wouldn’t just be a simple regurgitation of facts, but a high-level exam that asks about specific people, places, events, concepts, or arguments that require the students to draw upon a wealth of details on their stimulus to answer.

 

That way, they need to start critically assessing the information provided by AI as it builds their stimulus sheet in order to judge whether the information it provides is specific enough, broad enough, or even useful enough to provide the information needed to pass an unseen exam on the topic.  

 

There could be limitations set on the amount of revision material they can bring in, such as a page limit or certain format.

 

Realistically speaking, they could bring too much information in, and it would work against them, so the limitations would not have to be very strict.

 

The benefit of this process is that the students will closely read the information provided by AI and make a series of critical decisions about what to keep and what to remove.

 

This engagement is crucial learning process.   

But what about the times when an exam is not the culminating assessment item?

 

How do we handle a traditional research task that students don’t just fall back on a generic AI essay response?   

 

In short, AI allows us to dramatically increase teacher expectations of what students can now achieve in their assessments.

 

Real-world historians are currently using AI research tools to gather information from sources at a breadth and speed that they could not do previously.

 

That is a huge benefit to the profession.   

 

But this is also true of our students. AI can now quickly find high-quality sources with far more detailed information than was previously available.

 

Therefore, with AI, there is no excuse for generic research essays.  

 

Instead, with the help of AI, every student can easily:  

• Create a nuanced and precise research question with a highly focused topic  
• Only use academic sources for research tasks (no more Britannica or World History Encyclopedia would be acceptable beyond a ‘C’ response).  
• Help find the best primary sources that help answer the question  
• Have flawless referencing of sources and grammar  

 

What does this look like? Instead of a student doing research on the ‘causes of the rise of the Nazi Party’, they should be going far deeper.

 

For example, they can now ask, “What did middle-class women think about the Nazi Party and Hitler himself before 1933?”.

 

Then, as they research, they can now quickly get to first-hand accounts from these women from the time, thanks to AI.

 

The sources that will help them will come from archives and academics, not popular-level history websites.

 

And, thanks to AI, the students can quickly find different opinions among sources on this question.

 

The quality and depth of a student’s response will be far beyond what was ‘acceptable’ in the last few decades. 

We have to face the fact that students will be using AI extensively in their future careers.

 

Employers will expect them to use programs to produce information, reports, and new ideas, using AI tools as a baseline requirement of their employment.

 

As a result, we need to teach our students how to be the best at these things so that they have an edge in the employment market.  

 

Therefore, when teaching and assessing, we should give our students a wide scope to use AI, but to know that every student in the world is now using AI as well.

 

It becomes central to our teaching them to warn the students that at the moment, everyone is using it in a very over-simplified way.

 

If they want to be the best, achieve their dream careers, and stand out for the masses of basic AI users, they need to work hard to supercharge the outputs they’re getting from AI.  

 

So, how can we teach our students to produce the very best outcomes using all of the tools available to them?  

 

Students need to know that just asking ChatGPT to write an essay that reads like an old knowledge-based essay, the responses will all be very generic and low-level.

 

This is the new ‘C’ level, as this mimics what a basic use of AI will do in the workplace.

 

What we want is students that go beyond this standard and create something that is far more complex, detailed, and useful.

 

Students can even use AI to achieve this, but they need to be taught how. That is our job.