.png)
.png)
How to Design AI-Resistant Science Assignments
During a recent online biochemistry class, I showed students a structural formula of glucose and asked them to work out its molecular formula. The atoms were all shown, so they needed to count the carbons, hydrogens and oxygens.
One student answered, “C six, H twelve, zero six.”
I noticed the “zero.” Had she counted six oxygen atoms, or was she reading a formula without connecting the O to oxygen? I wondered whether she was reading an AI response, although a spoken slip could not tell me that.
What I could say was that hearing the right numbers had not shown me how she arrived at them. We could have moved on with the question apparently answered and the skill still unchecked.
Counting atoms is hardly an elaborate assessment. But it was the practice I wanted students to have. This is the difficulty I am trying to work through with AI: how to keep useful science activities while finding out whether students can do the work those activities are meant to teach.
What makes a science assignment AI-resistant?
In his Edutopia article, “Is Your Assignment Vulnerable to AI?”, Andrew Miller recommends looking at the thinking a task requires rather than searching for something AI cannot complete.
For science teachers, I think that means considering both the question and how students will respond to it. A lab prompt might require excellent reasoning, but a student completing it at home can still give the data and instructions to a chatbot. A short follow-up during the lesson gives us another opportunity to assess their understanding.
That is how I am using AI-resistant science assignments here: tasks that make it harder to outsource all the intended thinking, with a manageable individual check built into the lesson.
The follow-up does not have to be difficult. For glucose, I could ask students to mark the oxygen atoms they counted, then repeat the task with an unfamiliar structure.
Keep useful practice, but be careful what it proves
Students still need to convert units, rearrange equations, read structures and learn to write conclusions. I would not replace all of that with projects simply because AI can answer the practice questions.
A graph prepared at home can still be useful. Asking students to interpret part of it in the next lesson helps me decide what they understand and what needs more teaching. The preparation and the individual check serve different purposes.
I discussed that distinction in my post on what completed student work can actually show us. It matters when an assignment carries a substantial grade: a finished product may need to be considered alongside evidence from students’ work during class.
It also helps to explain why the practice matters. If a student can search for glucose’s formula in seconds, counting its atoms may seem pointless. Reading an unfamiliar molecule is a reason to learn the skill, and students deserve to hear that reason.
Use REAL to make one useful change
The REAL Science Task Framework is a short check for an assignment you already have. You can use one part without rebuilding the whole task.
Require relevant evidence
This is about which evidence students select.
Rather than asking them to “use the results,” ask them to mark the readings, observations or graph points that support their answer. They might compare two values or identify a result that needs further investigation.
Be fair about the evidence available. A group with missing measurements may need a clearly labelled class data set to practise the analysis. Requiring their own results regardless can leave them stuck—or tempted to invent what is missing.
Expect evidence-based reasoning
This is about what students do with that evidence.
Two numbers copied into a conclusion do not explain why the comparison matters. Ask students to connect the values to the scientific idea, consider another explanation or state what the evidence cannot establish.
The difference becomes clear in lab evaluations. A student moves from “human error” to “we may have spilled some liquid,” but still has not explained the effect. If the liquid was lost after its mass was measured and before its volume was measured, would the calculated density be too high or too low? Why?
Students may need a worked example before they can answer that independently. A more demanding prompt is only useful if we have taught the reasoning it requires.
Ask students to reveal their process
Choose a point where students have to make a decision: selecting measurements, choosing graph axes or revising a claim. Ask for a brief explanation there.
In group work, include an individual response. Clear lab roles and thoughtful grouping help students contribute, but the student who collected the apparatus still needs a chance to interpret the results.
My AI-resistant assessment strategies post gives more examples of these checks.
With spoken answers, let students refer to their work and give them time. A sketch or annotation may tell you more than an immediate verbal response.
Link the task to real conditions and limitations
Give students a condition that affects their choice.
If a mixture contains water and dissolved copper(II) sulfate, recovering the salt alone is a different problem from recovering both substances.
Ask students where each substance will be collected and how their proposed method accomplishes that.
The condition should help assess the science.
A long scenario about costs, local needs and environmental impacts might suit a decision-making project; it would be unnecessary reading if you only need to check a separation technique.
A before-and-after example for an online science lesson
Suppose students have learned how temperature affects enzyme activity. You provide the following illustrative data. The same enzyme and substrate amounts were used, with the same reaction time and measurement method.
Each temperature was tested once.
Temperature | Product formed after 60 seconds (arbitrary units) |
20°C | 8 |
35°C | 18 |
55°C | 4 |
A familiar assignment might ask:
Describe the effect of temperature on enzyme activity. Explain your answer.
A student could submit a general paragraph about increasing temperature and denaturation without examining the table.
Here is a revised task:
Mark the two comparisons you would use to describe the results. State the values.
Use your knowledge of enzyme activity to suggest why product formation increased and then decreased. Distinguish what the data show from your explanation of the pattern.
Another student writes, “The enzyme’s optimum temperature is exactly 35°C.” Do these results establish that? Explain.
Choose one additional temperature to test and explain what you hope to learn from it.
The table supports a narrower claim: 35°C produced the most product among the temperatures tested. It does not establish the exact optimum. The low result at 55°C is consistent with reduced activity following denaturation, but these measurements alone do not prove that explanation. With only one trial at each temperature, repeat measurements also matter.
For an online lesson, I would give students a few minutes to annotate the table and write their claim before discussing it. They could work on a printed copy or in a document, depending on the tools available.
Then give everyone a short individual follow-up while the class is still together:
Could you predict the exact amount of product at 40°C from this table? Explain what you can and cannot say.
Allow a minute of quiet thinking, then have students submit their responses before showing any. Discuss a few contrasting answers afterward. Everyone has attempted the interpretation, and you have something more useful than one volunteer’s answer.
This uses all four parts of REAL: students select data, explain the pattern, leave a record of their reasoning and deal with the limits of the investigation. It is still a short data-analysis activity.
A live online response cannot establish that no outside help was used. For a major assessment, you would need to consider the conditions more carefully.
For practice, these responses can help you decide whether the class understands the difference between a result and a claim that goes beyond it.
Put AI expectations beside the assignment
Tell students what help they may use before they begin. For this enzyme activity, if independent interpretation is the goal, the direction could be:
Use the class notes. Complete the comparisons and explanations yourself. Do not use AI to generate, rewrite or check your responses.
The boundary may differ for another lesson.
My AI guidelines for the science classroom cover examples involving observations, data and lab reports.
If you want adaptable materials, my AI Classroom Policy Toolkit on TPT includes science-specific handouts, permission-level displays, agreements and disclosure forms. You can choose the wording that fits your course and introduce it with the included presentation.
Leave time for the check
If you want students to interpret their lab results before leaving, plan time for it.
It cannot always fit into the two minutes after cleanup.
For the next assignment, decide which part you need each student to attempt independently. Replace a less useful question if necessary, rather than adding another page. Then use their responses to choose what to teach next.
If several students select the correct numbers but cannot explain the comparison, they need help with reasoning. If they explain the pattern but claim an exact optimum from three temperatures, they need practice judging the limits of evidence. Those are different teaching needs, and a completed paragraph can conceal both.
What is one question you ask to find out whether students understand an answer they have submitted? Share an example in the comments.
Comments