Before AI compares, define what you measured

A fictional plant-measurement exercise shows why clear definitions, missing-data labels and human judgment belong beside an AI answer.

Original paper-cut illustration of two green plants in navy and teal pots beside an open blank notebook and an unmarked amber ruler on a warm classroom desk.
Original conceptual illustration.
Watch the 42-second measurement-note lesson

The comparison begins before the prompt

Imagine a fictional classroom with two potted plants. One learner measures from the tabletop to the highest leaf. Another measures from the soil surface. Both write “height” in a notebook. An AI tool can turn those notes into a neat comparison, but the same column heading now hides two different measurements.

The first useful question is not which plant is taller. It is whether the observations describe the same thing. This teaching exercise uses no real study data and reports no measured result. Its purpose is to practice defining evidence before asking a system to summarize it.

An observable trait is not its explanation

The National Human Genome Research Institute defines phenotype in terms of observable traits and explains that both genetic makeup and environmental factors influence it. Its glossary also cautions against assuming a simple one-to-one relationship between genotype and phenotype. This is an educational reference, not a new finding from Ben's research.

That distinction gives our classroom example a useful boundary. A height observation describes a trait at a particular time. It does not, by itself, explain why the plants differ. A confident sentence about superior genes would go beyond the evidence available in this exercise.

Write a measurement note

Before collecting anything, agree on a short definition: for this exercise, height means the vertical distance from the soil surface to the highest point of the plant in its usual position, recorded in centimeters. Note the observation date, the tool and any relevant conditions. The definition is a classroom convention, not a universal botanical measurement standard.

Give each plant a stable label. Keep the original observations separate from later comments. If a measurement was not taken, write “not observed,” not zero. If the starting point is unclear, label it unclear instead of quietly guessing. These choices make the limits visible to both a person and an AI tool.

Now ask the tool to organize the notes using only the supplied observations. Require it to preserve units, dates and uncertainty labels, and to flag incompatible definitions. Do not ask it to fill the gaps with plausible values. The useful output may be a question for the observer, not a ranking.

Review the transformation

Compare the organized version against the original notes, one observation at a time. Did centimeters stay centimeters? Did an absent measurement become a number? Did “observed today” become a claim about growth over a week? A polished paragraph should not change the meaning of the evidence.

If the observations used different starting points, do not invent a conversion from missing information. Retain that limitation and decide whether a new, consistently defined observation is needed. Even a consistent comparison would not isolate the effects of light, water, genetics or other factors. Explaining causes would require a different question and an appropriate study design.

This is an original teaching activity and a suggested review habit. It is not a validated evaluation of an AI model, a genomics experiment or health advice.

Keep the definition with the contribution

For a small knowledge-portfolio entry, save the measurement definition, the original notes, the AI-organized draft and your review. Name who chose the definition and who checked the transformation. That preserves human contributions that can disappear when only the final paragraph survives.

The connection to AI governance is practical: an instruction to organize evidence is not permission to invent evidence. The connection to attribution is equally concrete: the observer's choices and the model's wording are different contributions. Recording them does not establish ownership or grant reuse rights.

BenHarvey.ai separates its published genomics foundations from its current governance and attribution proposals. Today's exercise makes no new research-performance claim. Try it with harmless classroom observations: before asking AI to compare, write one sentence saying exactly what was measured.

More from the blog RSS feed