Can Plants Learn? Read the Replication, Too.

What would count as an association?

The proposed association was between airflow and light. The useful question was whether a cue that had predicted light during training could later influence growth when the light was absent. Simply growing towards a current light source would not answer that question.[1]

That distinction is why the comparison groups matter. A plant’s eventual direction could reflect several processes. The experiment must separate a response to the proposed learned relationship from a response already present without that training. A familiar word such as “learning” is not a substitute for the operational test.

2016: the reported pea-maze result

Gagliano and colleagues trained pea seedlings in Y-shaped mazes, pairing fans and blue light on the same or opposite arms. During the test the fan appeared without the light. Most tested seedlings grew towards the arm the training predicted, while the controls grew towards the arm where light had last appeared. The authors interpreted this as associative learning.[1]

The reported outcome was the arm into which a seedling grew. It was not a recording of thought, a human-style memory or an account of what the plant felt. Those would be additional claims requiring additional evidence.

2020: a result that did not repeat

Kasey Markel used more plants and an independent observer who scored the mazes without knowing the treatments. The trained and control groups did not show the reported learning difference. Control plants also did not consistently grow towards the last light position. The replication used a different pea cultivar, and substantial numbers were excluded because they reached the maze decision too early or too late.[2]

That gives two things to examine: the absent effect and the behaviour of the test itself. A negative result is informative, but how strongly it rules out a proposed effect depends on whether the experiment could reliably have detected that effect under its conditions.

The objection, and the response to it

Gagliano and colleagues argued that light had not worked as a reliable unconditioned stimulus in the replication. They suggested differences in plant spacing and background light might account for the weaker directional response, making the setup unsuitable for the learning test. Their correspondence introduced no new experimental data.[3]

Markel replied that the proposed setup differences did not adequately explain the outcome. He acknowledged that weaker growth towards the last light reduced the ability to detect learning, but challenged the expectation of a perfectly consistent response and compared the lighting arrangements. He also made clear that an unsuccessful replication does not rule out plant associative learning in principle.[4]

Keep the question open at the right scale

Our reading of this exchange is that the pea-maze claim needs reproducible support before it can bear a broad conclusion. “Not reproduced here” and “impossible in plants” are different statements. So are “reported once” and “established reliably”. The dispute is useful because it makes the conditions for a convincing test more explicit.

The named papers concern a particular associative-learning task. They do not settle every question about priming, persistent changes in cellular response, habituation or other processes sometimes called memory. Nor would a supported learning effect, by itself, establish subjective experience.[4]

At the Practice this is a reason to keep scientific evidence and fictional correspondence visibly separate. Felix’s voice is written by the service; it is not a translation of a plant’s testimony. Nothing in the maze exchange creates a household need for plant psychotherapy.

What to notice in the evidence

• Did the comparison groups behave in the way the test required?
• Were outcomes scored without knowing which treatment each plant received?
• How many plants were excluded, why were they excluded, and were the conditions sufficiently described to repeat the work?

Three questions to take to the paper

1. 01

   Read the claim and the method together

   Find the exact outcome used to define the proposed learning. Ask what result would count against the interpretation.

2. 02

   Read both sides of the replication exchange

   The original paper, replication, objection and response each answer a different part of the question. A headline rarely contains all four.

3. 03

   Ask what would resolve the disagreement

   A useful next test would make the competing explanations distinguishable and report its controls, exclusions and scoring rules clearly.

What remains uncertain

This dossier does not announce that plants can never learn, or that the 2016 finding has been independently established. It describes the named evidence and our limited interpretation of it. It is not a complete survey of research published under every definition of plant learning.

The scope of this dossier

The original work and parts of the later exchange involve Australian researchers; that location does not change the standard of evidence. A houseplant in an Australian room is also not equivalent to a pea seedling grown in a controlled maze.