The concept of plant intelligence has been advanced by Trewavas as a potentially useful framework to guide those seeking to understand plant growth and development. In this short critique, the validity of this concept is critically assessed. Central to this critique is the proposition that the concept of the individual, to which intelligence and behaviour are intimately linked, cannot usefully be applied to plants. It is argued that the adaptive responses of plants are best appreciated if the importance of the autonomy of the individual organs is acknowledged. Although Trewavas does acknowledge the autonomy of organs by describing an individual plant as being ‘a democratic confederation’, that terminology implies a complexity to the interaction between organs which would demand a cogitative ability beyond that actually demonstrated in plants. It may be more appropriate to consider a plant as operating normally as a simple economic federation of many specialized economies (organs and cells). Occasionally, there can be a dramatic, and sometimes complex, reshaping of the economic balances, with the result that the fate of some or many of the individual cells will change. However, such major changes in growth and development are driven by a few simple events in an individual organ and cells. These driving events are more akin to small local revolutions in individual states than they are to democratic decisions in a sophisticated confederation.

Received: 9 September 2003; Returned for revision: 24 October 2003; Accepted: 8 January 2003


In a stimulating, thought‐provoking and wonderfully wide‐ranging article, Trewavas has argued that ‘the use of the term intelligence with regard to plant behaviour will lead to a better understanding of the complexity of plant signal transduction, the discrimination and sensitivity with which plants construct images of their environment and raise critical questions how plants respond at the whole plant level’ (Trewavas, 2003). Trewavas argued that evidence of learning and memory in plants, both considered necessary components of a truly intelligent system, has largely been ignored with the result that there is little appreciation of intelligence in plants. While there is a danger that any discussion about intelligence in plants becomes distracted by semantics, beneath any such linguistic debate there are important issues that need to be appreciated by anyone attempting to understand how plants work.


A key issue in discussing intelligence is the concept of individuality. The word individual leads us to the first semantic trap. The word in English derives from Latin, in‐ meaning not and dividuus meaning divisible. However, dictionary definitions (e.g. Chambers, 1959) stretch this definition to encompass meanings from subsisting as one to having an independent existence. Animals are highly individualistic in the sense of subsisting as one. Each animal has an individual identity which is a consequence of the overlaying of a very inflexible pattern of growth and development with an extremely variable behaviour (both ‘hard wired’ and learned). (This short article focuses on comparisons of unitary ‘higher’ animals and plants while recognizing that equally important comparisons might be drawn between ‘lower’ and colonial animals and plants.) In contrast, the concept of the individual in the case of a plant is much harder to define. It is easy to think of a very young seedling (the commonest experimental object in plant science) as being an individual, but which is the individual bracken plant in a huge clonal patch of bracken? Are all Cox’s Orange apple trees the same individual? Maybe, along with the injudicious use of statistics and the excessive constraints that tight experimental control impose on studies of plant functioning (Trewavas, 2003), we need to add that a blinkered use of ‘individual’ young seedlings as model systems may also be misleading. Interestingly, the widespread adoption, in the 19th century, of the very young seedling as the model for much experimental work coincided with the exploration of the proposal that, as in animals, certain plant organs could control the growth and development of other parts of the individual (for some history, see Went and Thimann, 1937). In a very young seedling with only two, distinct elongating organs, each possessing an apical region pioneering the exploration of their own very different environments, it is maybe reasonable to consider the seedling as an individual. However, once the two young axes lose their monopolies when other organs grow and develop, the sense of what is the individual becomes less clear. Consequently, while there are some general principles governing the ways in which plants grow and develop, principles that may be most easily identified by studying very young seedlings, the experimentalist needs to be conscious of the need to explain how similar processes might be controlled in more complex, multi‐organ plants. A 4‐d‐old arabidopsis seedling might look like an individual but a strawberry plant with attached runners, a coppiced willow tree or a shallot plant present difficulties for those seeking to identify the individual. Those experimentalists who confine their studies to young seedlings are not confronted with one of the key differences between animals and plants. In animals, growth and development is the enlargement and modification of the individual. But, in plants, growth and development also includes the adding of new members, or the discarding of some old members, of what Trewavas calls the ‘democratic confederation’. This key difference is least apparent in young seedlings; but even in young seedlings the main organs can be grown in isolation, hence in the short term such organs largely depend on their neighbours in the ‘confederation’ as suppliers (McIntyre, 2001) rather than governors. The ‘confederation’ has evolved with economic dependence being the main driving force behind specialization, hence the seedling is really an economic union rather than democratic confederation. (Plant biologists might more usefully read Adam Smith than Thomas Paine to appreciate their experimental subjects.) The capacity of the seedling, and more evidently the capacity of the mature plant, is largely the summed capacity of the component parts, with many of those capacities being interdependent largely in ‘economic’ terms. As will now be argued, any ‘intelligence’ that might be ascribed to ‘the plant’ could only reside in organs, tissues or cells because the concept of the plant as an individual is a misleading one.


‘Intelligence is not a term commonly used when plants are discussed’ (Trewavas, 2003). Maybe this is because plants are clearly not intelligent in the sense that most people understand the term ‘intelligent’. The word intelligence comes from Latin (from intellegere to discern, comprehend and, literally, ‘choose between’, from inter‐ and legere, to choose). The key words in this definition are discern, comprehend and choose, all of which are terms that are meaningful in the context of human behaviour. These terms, and the concept of intelligence, were adopted by those participating in the development of the English language to describe actions, and to express thoughts, about their own behaviour. The terms discern, comprehend and choose each imply a considerable degree of mental processing of more basic sensory information. There is rather little evidence that plants (or maybe more accurately plant cells) do anything other than rudimentary processing of sensed information and that alone should caution us against adopting the term ‘intelligence’ when discussing the abilities of plants. Trewavas moves to firmer ground by adopting Stenhouse’s view (Stenhouse, 1974) that intelligence is ‘adaptive variable behaviour within the lifetime of an individual’, hence directing attention at the term behaviour. This is convenient for Trewavas’s thesis because many biologists use the term behaviour simply to denote ‘a response to a stimulus’ and that allows the term to be applied rather easily to plants. Plants do indeed show responses to stimuli and clearly those responses are largely adaptive, hence one comes to a point that Stenhouse’s definition of intelligence could be applied to plants. But has this semantic exercise helped us identify new, productive ways of observing, studying, manipulating or understanding plants? Several generations of scientists have never doubted that all organisms have evolved to show some ability to adapt to their environmental circumstances. A bacterium can monitor its environment and instigate developmental processes appropriate to the prevailing circumstances, but is that intelligence? Such simple adaptation behaviour might be bacterial intelligence but is clearly not animal intelligence. Hence we have reached a definition of intelligence that has no meaning unless combined with another word or used in a precise context in a sentence. This is not a new problem, as shown by the debate about whether machines can be intelligent. The term machine intelligence can mean something other than the intelligence of a machine. We can certainly start using the term plant intelligence but only if it is agreed that the term has nothing to do with intelligence (the ability to discern, comprehend and choose) in the more widely accepted sense. (To add to the opportunity for confusion it should be pointed out that the term ‘plant intelligence’ is already in use to mean the utilization of data about the performance of factories to optimize performance—but plants could be considered as factories, so maybe that is OK; http://www.gefanucautomation.com/ plantintelligence/default.asp.) However, why do we need a new term? If the discussion of a plant’s adaptive responses is less ambiguous than a discussion of the plant’s intelligence, why adopt the new term? Indeed, might not the adoption of the term plant intelligence begin to distort, rather than clarify, our perception of the way in which plants function? To illustrate this danger, some of the key terms that are part of Trewavas’s view of plant intelligence will now be placed under scrutiny.


Trewavas (1999, 2003) considers learning to require two faculties. The organism must be capable of having a goal, and the organism must be able to assess its current behaviour and adjust that behaviour to make the attainment of the goal more likely. As part of the latter ability, he considers the need for an error‐detection mechanism—a method for judging current state against a reference state (goal). Each of these elements needs to be considered from the perspective of the adaptive responses of plants.


Trewavas defines a goal as a plant’s fitness objective. The main problem with the concept of a single plant possessing a goal is that each mature plant is made up of ever‐changing components (roots, leaves, flowers, etc.), each of which occupy their own temporally and spatially variable environments. When a plant is at its most individualistic, as a seed, there are clear goals—dispersal, survival and germination at the right time. However, shortly after germination each extending meristematic region, such as a root tip or an apical bud, may encounter local conditions that are very different from those encountered by other organs on the same plant exploring other localities. Furthermore, the conditions experienced by one generation of organs are likely to differ from those experienced by later generations. The need to optimize the performance of each organ in its own variable, unpredictable environment has given rise to a selective process that has favoured specimens with sensing and response capacities, which are largely devolved to the organ or sub‐organ level. This devolution of sensing and response to functionally specialized organs results in most goals being themselves functional and local. Thus it is hardly surprising that the great majority of sensing and response events in plants are localized to the cell, tissue or organ, in that order. Furthermore, because most organs are multifunctional and grow in heterogeneous environments, any overall goals must be flexible and necessarily based on a conservative compromise. To cope with the variability that an organ may encounter, it is likely that evolution will have selected for variants that can make approximately the optimum number of structures (be they organs, cells or cell components), but where there is some variable functional capacity. The variable functional capacity of each unit enables the overall current and short‐term functional needs to be met. For example, the gas exchange capacity of a leaf is in part determined by the number and the aperture of the stomata, and varying the latter is the most effective way of giving operational control to the leaf in a varied environment. Furthermore, the environmental conditions that a developing leaf experiences might differ considerably from the conditions that a mature leaf experiences. So for an organ there may be one or more developmental goals and some operational goals. The end result should provide sufficient functional flexibility to cope efficiently with whatever changing environment it may have to operate in. Another level of control might have evolved to cope with rarer, more extreme conditions that push organs beyond the limits of their operational controls. Such ‘emergency’ shutdown routines at best protect all the members of the economic union and at worst ensure the survival of sufficient members to enable a successful recovery programme to be initiated. The graded responses of plants to increasingly severe water stress would be an example of such ‘emergency’ recovery routines (Hsiao, 1973). Even in these circumstances, the response of ‘the plant’ is actually the sum of the responses that take place in cells, tissues and organs. In such circumstances, each responding cell need have no information as to the location or scale of the problem elsewhere, simply that a programmed response is needed to contribute to the fitness of the union.

In conclusion to this consideration as to whether a plant can possess a goal, the most important means by which the economic union that is the plant achieves the fundamental goal common to all living creatures (survivorship and reproduction) is to possess both a flexible way of controlling the number and types of organs (organs can be initiated and discarded) and allowing each organ considerable operational autonomy. Most of the developmental and operational control of the organs is devolved to local sensing, so that the development and functioning of an organ can match the local conditions and hence can contribute most effectively to the union. Selection in plants has favoured the evolution of flexible modular organisms where the final structure is just one of many different possible outcomes. The developmental flexibility which enhances plant fitness is in contrast to the adaptive flexibility that behaviour gives to animals.

Error detection

As outlined in Box 1, a simple on/off switch linked to a sensor is all that is needed for ‘error detection’. The examples of oscillations that Trewavas takes as evidence of ‘trial‐and‐error learning’ could equally be used as evidence for simple automaton behaviour. Examples of homeostatic mechanisms in plants and animals, which are readily modelled in terms of a sensor and a reaction to sensor state, provide robust, if simple control systems, sufficient for many purposes. The fact that many examples of oscillatory behaviour occur in ephemeral organs also causes one to doubt whether any benefit would accrue from the possession of an ability to learn. For example, a root undergoing gravitropism is actually a different root at the start and end of the process. If a root is exposed to an extreme gravitational stimulus twice, 12 h apart (which would be an event so rare in nature that one cannot imagine selection optimizing a mechanism that included a learning stage), the second stimulus will initiate a new response in cells that have not necessarily experienced the first stimulation, and hence they cannot usefully have learned. This is in contrast to an animal where learned behaviour resides at the organism level – the hand has no memory of being burned, but the brain controlling the hand might possess the appropriate knowledge after an adequate previous learning experience.


Trewavas considers the evidence that plants make choices, or decisions, in reaching his conclusion that plants have a kind of intelligence. But can plants really make choices in any meaningful way? A thermostat can ‘make a decision’ (Box 1) but few would claim that it makes a real decision in any sense of free will. There would seem to be a confusion here between approaching a state with a limited number (often two) of predetermined outcomes and real decision making. A more demanding criterion for choice or decision making is whether there are several variable outcomes, e.g. action or inaction, and different ways of taking action. When ‘choices’ are made in plants they are made at the organ (or cell) level and not at the plant level. Even for the most fundamental ‘choice’ that ‘a plant’ might make—to initiate flowering—one could argue that the switch to reproductive development in the apex is a simple two‐state system. It is also surely significant that the control of flowering is not determined centrally by ‘the plant’, indeed the photoperiodic sensing can occur in many leaves or even within a small part of one leaf—a very robust system (Zeevaart, 1976). The plant does not make a choice to flower, instead a leaf influences the predetermined fate of some cells in the apex. There is no choice made by ‘the plant’ rather events occur within the plant that dramatically change the balance within the economic union, with consequent changes to the fate of some members of the union.


Does a plant build up a spatial map of its environment? To answer this question one needs to think more carefully about the type of plant, the type of environment and type of spatial map that might provide a fitness benefit. Once again, the favoured experimental subject of those investigating a sensing system in plants, the very young seedling grown in a very controlled environment, might be misleadingly simple. It is very clear that the construction and maintenance of an accurate spatial map of a large plant such as a tree would demand huge processing power. Thousands of leaves and roots, each having a need to sense their local environment to serve their own needs, would produce very large amounts of data. The data would be changing by the minute as the leaves fluttered in a breeze and the roots explored their very heterogeneous environment. This example indicates that there is a need for great caution when scaling up knowledge from young seedlings to more mature, more complex plants. It might be more profitable to regard a tree not as a big seedling but as a collection of thousands of seedlings—each potential cutting is a potential new plant, hence the concept has some validity. This consideration should make us concentrate on the organizational level at which spatial mapping could be meaningful—the organ level. In the main, each operational or developing organ can, at best, sense a few parameters that sufficiently characterize their local environment with respect to the function of that organ. Each organ has some extra information derived from the overall integrative state of the plant (much of which might be considered to be similar in nature to consequences of independent, competing economic activities in a human confederation). This extra information is available from the summed activities of all members of the federation of organs, but such exotic information is rarely complex. Some extra specific information might also be available to certain cells from chemical signals emanating from non‐adjacent cells. Although Trewavas, as Canny (1985) before him, argues that the pathways of inter‐organ transport could allow complex chemical signals to flow between organs, it is unlikely that the complexity of the signalling molecules implies a complexity of message. (It should also be noted that the existence of signalling systems is not the issue at stake, rather it is whether complex signalling is a good indicator of the existence of an ‘intelligent’ adaptive response.) The complexity of signalling molecules is only necessary to ensure that the right recipient receives the simple message. A signalling molecule is akin to a postcard bearing a tick or a cross as the message, with the much more complex information of the address ensuring that the simple message reaches the right recipient (Firn, 1985). [The reader interested in the complexities of hormonal signalling in plants is referred to Weyers and Paterson (2001).] It might also be significant that some of the best examples of long‐distance signalling (e.g. flowering or tuberization) are actually ones where spatial information is not important—many leaves can detect day‐length changes and it matters not which one does. Are there any examples of spatial information derived by plants that is not used mainly locally?

In conclusion, there is no convincing evidence that plants possess the ability to convey sufficient information from sensors to any processing units to enable a meaningful spatial map to be constructed. The only map that a plant as a whole makes of its environment is recorded in its morphology, anatomy and chemistry. It is a map of past events, not of future plans.


There can be no doubt that the developmental history of a plant can determine the outcome of many subsequent sensing events. More accurately, the developmental history of many organs (or cells) can determine the response that results from subsequent stimulation. Is that memory or is it simply a consequence of developmental progression? A simple washing machine controller possesses ‘memory’ due to the equivalent process of developmental progression. Developmental progression at its simplest is a linear sequence of events, with one or more steps under the control of some sensor (Box 1). Once the sequence is initiated, progress is determined by the state of some or all of the controlling sensors. Events occurring at an early stage can influence the outcome of later events and the system can thus have ‘a memory’ of earlier events. However, a crucial distinguishing characteristic of memory in animals is that the processes are not linear and they are not sequence‐dependent.

Assuming that ‘plant memory’ is little more than developmental progression, it is clear once again that developmental progression would be a property of organs or cells and not plants. This fact reveals why it is unlikely that true memory could have evolved in plants. Organs, for example, need to cope with the environment in which they find themselves and, being essentially ephemeral, they could benefit little by retaining and using information of past events. A root growing in a heterogeneous environment cannot use an experience of its current state to predict its future state. Water (or potassium or nitrate, etc.) which is in short supply now may be in ample supply later. The memory of animals is aimed at using the experience of one organ to the advantage of others—a hand that is burned can provide a lesson to the whole organism so that in future other limbs can avoid experiencing the same problem. Where is the equivalent collective organism experience (= memory) of benefit in a plant’s growth and development?


Adaptive behaviour takes many forms in different organisms. By using terms and concepts of learning, memory and intelligence that derive from human experience one can build up a view of each type of adaptive behaviour in other organisms, but maybe only at the expense of clarity. Indeed, there is a danger that the adoption of concepts and terms from higher animals distracts attention away from the key adaptive processes evolved by organisms that have not been selected to learn, memorize and think. In this article it has been argued that the ‘intelligence’ perceived in the adaptive response of a plant might best be considered to be the sum of the collective adaptive responses of its cells. A general theme running through this article is that the human sense of the individual easily distorts the ability of humans to appreciate how a plant operates. Even in the previous sentence the two simple words ‘a plant’ feel, to the author, inappropriate because these words emphasize the individual entity. Plant population biologists have adopted terms such as genet and ramet (Harper, 1977) because they appreciate that descriptors beyond the concept of the individual are needed to understand plant populations. However, many plant biologists studying the functioning of cells or organs don’t look to plant population biologists for inspiration but to animal biologists. Such plant biologists might benefit from shifting their focus to other types of plants rather than other types of organisms. An appreciation may then develop that clonal plants are simply demonstrating an extreme form of developmental adaptation, but it is a form of adaptation that can be seen in simpler terms in a 4‐d‐old arabidopsis seedling. In clonal plants, the ramet is a device to provide an efficient division of labour in a heterogenious, patchy environment (Stuefer et al., 1996; Hutchings and Wijesinghe, 1997; Jónsdóttir and Watson, 1997) and that is simply an extension of the strategy of allowing organs an autonomy to optimize their localized performance. The extent to which those semi‐independent units interact, and how they interact, is easier to understand if the overall strategy of plant development is comprehended.

Finally, the author, during the preparation of this article, has become very aware that human language guides and greatly limits our thoughts when trying to appreciate the functioning of plants. It might ultimately be unproductive to debate whether a plant has memory, whether a plant can learn or whether a plant can possess a spatial map, because the key terms come from an organism that is highly individual, an organism that finds it hard to appreciate the faculties of organisms that lack individuality and all that that implies. Our language lacks appropriate words. However, if new words are needed to describe how plants function, maybe we should invent new ones rather than trying to redefine existing ones.


Tony Trewavas is thanked for many years of provocative and stimulating comment. John Raven is thanked for sharing his wisdom and knowledge.


. Ashby’s law and the pursuit of plant hormones: a critique of accepted dogmas, using the concept of variety.
Australian Journal of Plant Physiology
. Ashby’s Law of Requisite Variety and its applicability to hormonal control.
Australian Journal of Plant Physiology
GeddieW, ed.
. Camber’s twentieth century dictionary. Edinburgh: W & R Cambers.
.Population biology of plants. London: Academic Press.
. Plant responses to water stress.
Annual Review of Plant Physiology
HutchingsMJ, Wijesinghe DK.
. Patchy habitats, division of labour and growth dividends in clonal plants.
Trends in Ecology and Evolution
JónsdóttirIS, Watson MA.
. Extensive physiological integration: an adaptive trait in resource‐poor environments? In: de Kroon H, van Groenendael J, eds. The ecology and evolution of clonal plants. Leiden: Backhuys Publishers,
. Control of plant development by limiting factors: a nutritional perspective
Physiologia Plantarum
. The evolution of intelligence—a general theory and some of its implications. London: George Allen and Unwin.
StueferJF, de Kroon H, During HJ.
. Exploitation of environmental heterogeneity by spatial division of labour in a clonal plant.
Functional Ecology
. How plants learn.
Proceedings of the National Academy of Science
. Plant intelligence: mindless mastery.
. Aspects of plant intelligence.
Annals of Botany
WentFW, Thimann KV.
.Phytohormones. New York: MacMillan.
WeyersJDB, Paterson NW.
. Plant hormones and the control of physiological processes.
New Phytologist
. Physiology of flower formation.
Annual Review of Plant Physiology

Author notes

Department of Biology, University of York, PO Box 373, York YO10 5YW, UK