Plants as Intelligent and Information Processing Organisms

Plants as Intelligent and Information Processing Organisms #

Plants are everywhere in our daily lives, in our offices and gardens, on our plates or between the joints of the pavement tiles. Yet, they are very different from us. In contrast to animals, they do not move, have a more modular morphology and synthesise their food from water, sunlight and nutrients they absorb (autotrophy). Moreover, plants are found in nearly every ecosystem on the planet and often thrive in those ecosystems. Without plants, animals in their current form would not exist on Earth since they are at the basis of each food chain. However, because plants are so different from animals, plants are often not considered as intelligent. Mostly, people think that plants “just grow” and undergo changes in their environment. An increasing body of research is proving otherwise, indicating that plants exhibit emergent intelligence (, (). Plant Intelligence: An Overview. BioScience, 66. biw048. https://doi.org/10.1093/biosci/biw048 ; , ) (). The Intelligent Behavior of Plants. Trends in Plant Science, 21(4). 286–294. https://doi.org/10.1016/j.tplants.2015.11.009 .

Are Plants Intelligent? #

Plants are at the basis of nearly all ecological systems on Earth (, ) (). Vegetation of the Earth and Ecological Systems of the Geo-biosphere (3). Springer-Verlag. https://doi.org/10.1007/978-3-642-96859-4 . Humans and, more generally speaking, all animal life depends upon plants either directly or indirectly. Despite their inability to move, which is also called sessile, plants thrive in almost all habitats found on Earth. This sessile nature has forced plants to develop a wide range of strategies to cope with environmental changes. Mangrove forests thrive despite growing in saltwater and being flooded twice a day due to the tide ( & , ) & (). Biology of mangroves and mangrove Ecosystems. In Advances in Marine Biology. (pp. 81–251). Academic Press. https://doi.org/10.1016/S0065-2881(01)40003-4 . Cacti have to store water over more extended periods of dehydration and rapidly refill reserves when water is available ( & , & (). Plant Physiology, Fifth Edition (Fifth edition). Sinauer Associates, Inc.. ; & , ) & (). Hydraulic Conductivity and Anatomy for Lateral Roots of Agave deserti During Root Growth and Drought-induced Abscission. Journal of Experimental Botany, 43(11). 1441–1449. https://doi.org/10.1093/jxb/43.11.1441 . Crops are subject to intense sunlight and heat during the day and low temperatures at night, even in summer.

To survive in highly variable conditions such as sunlight intensity, temperature fluctuations and water availability, plants need to continuously sense their environment through a wide range of sensors. These sensors are spread all over the plant. Based on these sensory signals, plants adapt their physiology accordingly ( & , ) & (). Plant Physiology, Fifth Edition (Fifth edition). Sinauer Associates, Inc.. . As a result, they can measure many more variables than humans, including electrical fields, chemical gradients and temperature (, ) (). Plant Sensing and Communication. University of Chicago Press. Retrieved from https://www.degruyter.com/document/doi/10.7208/9780226264844/html .

Unlike animal-like organisms that have central sensory processing in a brain-like structure, plants are distributed organisms. Plants lack specialised organs identifiable in animals, such as the brain, lungs and digestive system. Instead, they have a more modular structure that avoids high-specialisation. From an evolutionary perspective, this structure is vital since plants are sessile and can easily fall victim to predation. Loss of a vital organ would mean the death of the plant; by limiting specialisation, plants can continue to grow ( & , ) & (). The revolutionary genius of plants: a new understanding of plant intelligence and behavior (First Atria books hardcover edition). Atria Books, an imprint of Simon & Schuster, Inc. . Indeed, mowing the lawn does not stop the grass from growing. Planting a cutting of most plants also results in a new plant due to the rapid development of roots in the cutting. These characteristics make plants very distinct from animals.

Due to this very different behaviour from animals, plants are not often considered intelligent organisms (, ) (). The Intelligent Behavior of Plants. Trends in Plant Science, 21(4). 286–294. https://doi.org/10.1016/j.tplants.2015.11.009 . This is an attribute reserved only for animal-like organisms such as parrots, octopuses (, & al., ) , , & (). How intelligent is a cephalopod? Lessons from comparative cognition. Biological Reviews, n/a(n/a). https://doi.org/10.1111/brv.12651 , and humans. However, some researchers are challenging this view (, & al., , , & (). Consciousness Facilitates Plant Behavior. Trends in Plant Science, 25(3). 216–217. https://doi.org/10.1016/j.tplants.2019.12.015 ; , & al., ) , , & (). Plants are intelligent, here’s how. Annals of Botany, 125(1). 11–28. https://doi.org/10.1093/aob/mcz155 . As a result, the question arises: “what is intelligence?”. There is no single definition available. Yet, based on  () (). Intelligence Emerging: Adaptivity and Search in Evolving Neural Systems. MIT Press. and ( & , ) & (). A Collection of Definitions of Intelligence. Frontiers in Artificial Intelligence and Applications, 157. 17–24. Retrieved from http://arxiv.org/abs/0706.3639 , there are three properties that are common aspects of intelligent behaviour in many definitions: (i) intelligence is a property that an individual agent has as it interacts with its environment(s); (ii) intelligence is related to an agent’s ability to succeed or profit with respect to some goal or objective; (iii) intelligence depends on how the agent is adaptive to different objectives and environments. &  () & (). A Collection of Definitions of Intelligence. Frontiers in Artificial Intelligence and Applications, 157. 17–24. Retrieved from http://arxiv.org/abs/0706.3639 summarise this as:

General Intelligence Intelligence measures an agent’s ability to achieve goals in a wide range of environments ( & , ) & (). A Collection of Definitions of Intelligence. Frontiers in Artificial Intelligence and Applications, 157. 17–24. Retrieved from http://arxiv.org/abs/0706.3639 .

In this dissertation, we will use a more specific definition that is encompassed in the one above:

General Intelligence, Alternative Definition Intelligence is a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience (, ) (). Mainstream science on intelligence: An editorial with 52 signatories, history, and bibliography. Intelligence, 24(1). 13–23. https://doi.org/10.1016/S0160-2896(97)90011-8 .

This definition mentions some key aspects necessary for intelligence, such as problem-solving capabilities, memory, planning and more. Recent research is also identifying such behaviour in plants.

, & al. () , , , & (). Evidence for Light Wavelength-Specific Photoelectrophysiological Signaling and Memory of Excess Light Episodes in Arabidopsis. The Plant Cell, 22(7). 2201–2218. https://doi.org/10.1105/tpc.109.069302 showed that plants possess memory of previous light incidents, which is used for the optimisation of future light acclimation and optimisation responses. Combined with other research, this has led &  () & (). Secret life of plants: from memory to intelligence. Plant Signaling & Behavior, 5(11). 1391–1394. to conclude that plants can store and use information of light sum, intensity and day length for several days or more to anticipate changes that might appear in the near future in the environment. These examples illustrate that plants have typical learning (habituation, priming) and complexly integrated store/recall systems of memory ( & , ) & (). Plant memory: a tentative model. Plant Biology (Stuttgart, Germany), 15(1). 1–12. https://doi.org/10.1111/j.1438-8677.2012.00674.x . Mimosa pudica (also called sensitive plant) plants have also been used to demonstrate the learning capabilities of plants. These plants can fold their leaves, a defensive mechanism. Yet, this should only occur when there is imminent danger because photosynthesis is reduced, and leaf folding also requires energy. , & al. () , , & (). Experience teaches plants to learn faster and forget slower in environments where it matters. Oecologia, 175(1). 63–72. https://doi.org/10.1007/s00442-013-2873-7 showed that dropping a plant initially triggers the plant to fold the leaves, yet is ignored after a few incidents. In later work, , & al. () , , , & (). Learning by Association in Plants. Scientific Reports, 6. 38427. https://doi.org/10.1038/srep38427 even demonstrate associative learning in plants. The researchers introduced a neutral environmental cue along with the same direction of the incident light. They showed that this cue was used by plants to predict the future light source’s location, affecting the growth direction. These case studies underpin that plants can store and recall past events despite lacking a central brain-like structure. , & al. () Baluška, F., Gagliano, M. & Witzany, G. (). Memory and Learning in Plants. Springer International Publishing. https://doi.org/10.1007/978-3-319-75596-0 assembled an overview of recent work on memory and learning in plants.

Moreover, plants can also communicate with each other. Since the discovery of phenolic compound accumulation in poplar when nearby trees are damaged ( & , ) & (). Rapid Changes in Tree Leaf Chemistry Induced by Damage: Evidence for Communication Between Plants. Science, 221(4607). 277–279. https://doi.org/10.1126/science.221.4607.277 , there has been ever-increasing evidence that plants communicate. Main methods for communication appear to be airborne volatile organic compound (VOC) but others have demonstrated that root-root interaction when experiencing drought also occurs (, & al., , & (). Plant communication. Plant Signaling & Behavior, 7(2). 222–226. https://doi.org/10.4161/psb.18765 ; , & al., ) , , & (). Root exudate signals in plant–plant interactions. Plant, Cell & Environment, 44(4). 1044–1058. https://doi.org/10.1111/pce.13892 ; as well as plant-to-plant competition ( & , ) & (). Plant Physiology, Fifth Edition (Fifth edition). Sinauer Associates, Inc.. ; or even acoustic signals (, (). Green symphonies: a call for studies on acoustic communication in plants. Behavioral Ecology, 24(4). 789–796. https://doi.org/10.1093/beheco/ars206 ; , & al., ) , , , , , , & (). Plants emit informative airborne sounds under stress. bioRxiv. 507590. https://doi.org/10.1101/507590 . Communication is not limited to hormonal or electrical cues; hydraulic or even electrical field communication is also reported (, & al., , & (). Ionic Current Changes Associated with the Gravity-Induced Bending Response in Roots of Zea mays L. 1. Plant Physiology, 100(3). 1417–1426. https://doi.org/10.1104/pp.100.3.1417 ; , & al., ) , & (). Swarm intelligence in plant roots. Trends in Ecology & Evolution, 25(12). 682–683. https://doi.org/10.1016/j.tree.2010.09.003 .

These examples illustrate the complexity and highly optimised behaviour of plants for the challenging conditions in which they develop. These behaviours can be interpreted as emergent intelligence. While not all aspects of the above definitions of intelligence are already observed, there is abundant evidence that plants are more than passive organisms. More research and improved experimental designs will be necessary to investigate this emergent intelligence further.

The ongoing research on plant intelligence and related properties has led some researchers to go even further and attribute emergent consciousness to plants (, & al., , , & (). Consciousness Facilitates Plant Behavior. Trends in Plant Science, 25(3). 216–217. https://doi.org/10.1016/j.tplants.2019.12.015 ; , & al., , , & (). Plants are intelligent, here’s how. Annals of Botany, 125(1). 11–28. https://doi.org/10.1093/aob/mcz155 ; , ) (). Awareness and integrated information theory identify plant meristems as sites of conscious activity. Protoplasma, 258(3). 673–679. https://doi.org/10.1007/s00709-021-01633-1 . This has led to a hotly debated topic with part of the community trying to debunk the consciousness hypothesis (, & al., , , , & (). Debunking a myth: plant consciousness. Protoplasma, 258(3). 459–476. https://doi.org/10.1007/s00709-020-01579-w ; , & al., , & (). Anesthetics and plants: no pain, no brain, and therefore no consciousness. Protoplasma, 258(2). 239–248. https://doi.org/10.1007/s00709-020-01550-9 ; , & al., ) , , , , , , & (). Plants Neither Possess nor Require Consciousness. Trends in Plant Science, 24(8). 677–687. https://doi.org/10.1016/j.tplants.2019.05.008 , while other support it vigorously (, & al., , , & (). Consciousness Facilitates Plant Behavior. Trends in Plant Science, 25(3). 216–217. https://doi.org/10.1016/j.tplants.2019.12.015 ; , & al., , , & (). Plants are intelligent, here’s how. Annals of Botany, 125(1). 11–28. https://doi.org/10.1093/aob/mcz155 ; , ) (). Awareness and integrated information theory identify plant meristems as sites of conscious activity. Protoplasma, 258(3). 673–679. https://doi.org/10.1007/s00709-021-01633-1 .

Are Plants Conscious? #

Often, the debate about plant consciousness revolves around a primary form of consciousness, simply put: “a first-person point of view” (, ) (). What is it like to be a bat?. Readings in philosophy of psychology, 1. 159––168. . While both pro and contra groups have compelling arguments in favour and against this hypothesis, we believe a paradigm shift is needed from a very anthropomorphic view on consciousness to a more generalised version.  () (). Awareness and integrated information theory identify plant meristems as sites of conscious activity. Protoplasma, 258(3). 673–679. https://doi.org/10.1007/s00709-021-01633-1 provided the first steps in this direction, by adopting the word awareness in place of consciousness. However, the experimental evidence to support or reject the consciousness hypothesis in plants is mainly lacking.  () (). Intelligence without neurons: a Turing Test for plants?. Protoplasma, 258(3). 455–458. https://doi.org/10.1007/s00709-021-01642-0 identified that most research focuses on analysing literature rather than hard experimental analysis.

Two causal interaction diagrams.
Figure 1.1 Two causal interaction diagrams and their causally effective information 𝛷 expressed in bit. Diagram (a) has a heterogeneous structure, resulting in a much higher value for 𝛷 than (b), which has a uniform structure. Figure inspired by Tononi (2004).

What is needed is a set of measurable criteria that enable researchers to test their hypotheses.  () (). Awareness and integrated information theory identify plant meristems as sites of conscious activity. Protoplasma, 258(3). 673–679. https://doi.org/10.1007/s00709-021-01633-1 proposes the integrated information theory (IIT) for consciousness. IIT was initially proposed in the field of neuroscience and psychology by  () (). An information integration theory of consciousness. BMC Neuroscience, 5(1). 42. https://doi.org/10.1186/1471-2202-5-42 . As the name suggests, it uses integrated information that is defined through a network of elements or nodes, illustrated in Figure 1.1. Linkages modify the behaviour of each node, thus representing the intrinsic information, which cannot be measured directly (, & al., ) , , & (). Integrated information theory: from consciousness to its physical substrate. Nature Reviews Neuroscience, 17(7). 450–461. https://doi.org/10.1038/nrn.2016.44 . Though this theory seems interesting at first glance, several issues can be identified when transitioning to physical media. Plants are composed of interconnected networks that indicate modularity. However, we should also consider other inanimate structures such as mass-damper systems. They are also composed of interconnected networks and fit the framework in a similar sense as plants. We can compute the IIT for these, yet attributing consciousness to mass-damper systems is far fetched. Similarly, in a simulated environment, the IIT can be computed for neural networks. Yet, it is also difficult to attribute consciousness to current state-of-the-art neural networks (, ) (). Intelligence without neurons: a Turing Test for plants?. Protoplasma, 258(3). 455–458. https://doi.org/10.1007/s00709-021-01642-0 .

Instead,  () (). Intelligence without neurons: a Turing Test for plants?. Protoplasma, 258(3). 455–458. https://doi.org/10.1007/s00709-021-01642-0 proposes that a Turing test for plants might be needed. The essence of a classic Turing test is that the observer investigates a system on its ability to think. To this end, the observer asks questions as if the system were intelligent. The performance of the system (also called black box) is compared to a true thinker (i.e. a human). If the observer is unable to identify which is the black box and which is the human, the system passes the test (, ) (). Computing Machinery and Intelligence. Mind, LIX(236). 433–460. https://doi.org/10.1093/mind/LIX.236.433 . Thus instead of maintaining the status quo and mainly performing literature reviews, it would be more interesting to devise a non-verbal Turing test. Though this might not be trivial, as  () (). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3). 417–424. https://doi.org/10.1017/S0140525X00005756 demonstrated using the Chinese room argument. The setup in the Chinese room argument is similar to that of the Turing test, but for a system that is perceived to understand Chinese. The system processes Chinese characters at the input and output according to a computer programme. If the programme is sufficiently advanced, it could pass the Turing test, convincing the human operator that the programme understands Chinese. However, since the system is following a predetermined programme, it cannot be considered as true intelligent behaviour (, ) (). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3). 417–424. https://doi.org/10.1017/S0140525X00005756 .

Plants as Computational Resource #

A general computational resource is depicted in Figure 1.2. Inputs are presented to a processing unit, which produces the desired output. Computers are prime examples: they receive inputs from, for instance, the keyboard, process these signals to generate the corresponding display command and depict the pressed letter on the screen. Humans are also information processing entities. We continuously receive sensory inputs that are processed by our brain and act accordingly.

Diagram depicting a general computational resource.
Figure 1.2 Diagram depicting a general computational resource. This diagram is not only a good model of computers, but also for humans and possibly plants.

While both a computer and the human brain are computational resources, they have widely different properties. On the one hand, humans have no trouble driving a car, even in conditions not experienced before, while this is a much more challenging problem for computers. While on the other hand, humans are not good at raw number crushing. For instance, computing the square root of a random number is non-trivial for us, yet computers are very efficient at this. Consequently, they each excel at different tasks. The same holds for less conventional computational resources.

artificial neural network (ANN), for instance, are loosely inspired by the brain, consisting of a network of elementary processing elements, similar to how IIT works. ANN are universal approximators (, & al., ) , & (). Multilayer feedforward networks are universal approximators. Neural Networks, 2(5). 359–366. https://doi.org/10.1016/0893-6080(89)90020-8 , so they can also be used as a computational resource. Research on ANN is a very active part of computer science.

One specific type of computing paradigm that is very relevant here is reservoir computing (, & al., ) , & (). An overview of reservoir computing: theory, applications and implementations. Proceedings of the 15th European Symposium on Artificial Neural Networks. p. 471-482 2007. 471–482. Retrieved from http://hdl.handle.net/1854/LU-416607 . It uses a randomly initialised ANN to perform computations. Figure 1.3 visualises a simple reservoir. Input is fed into the reservoir that transforms this information into new information that depends on the current input and past inputs through the recurrent connections present in the network. It is easiest to consider the system in discrete time, though the extension to continuous time is easily made. The input then consists of a series of sensor values. At each time point, a single value is put into the network. The arcs (arrows) from the input nodes to the nodes in the reservoir determine how the input modifies the reservoir state. The state values, represented as nodes, of the reservoir not only depend upon the input but also upon the previous reservoir state. The new node state (or value) is determined by the weighted sum of the previous states that influence a particular node. The darker an arc, the stronger this node influences the other. Finally, based on a (partial) observation of the reservoir, the output is determined. This output is thus also a weighted sum of node values. Though this simple readout system is observed to perform well for specific tasks, it is not advised to use reservoir computing for general-purpose machine learning (, & al., ) , , , , , & (). Backpropagation algorithms and Reservoir Computing in Recurrent Neural Networks for the forecasting of complex spatiotemporal dynamics. Neural Networks, 126. 191–217. https://doi.org/10.1016/j.neunet.2020.02.016 . However, physical reservoir computing (PRC) is still relevant for solving tasks related to the physical body that is used for computation (, ) (). Physical reservoir computing—an introductory perspective. Japanese Journal of Applied Physics, 59(6). 060501. https://doi.org/10.35848/1347-4065/ab8d4f . A typical example is the ability to use signals generated by a compliant body to steer locomotion (, & al., ) , , , & (). Effect of compliance on morphological control of dynamic locomotion with HyQ. Autonomous Robots, 45(3). 421–434. https://doi.org/10.1007/s10514-021-09974-9 .

General architecture of a RNN in reservoir computing.
Figure 1.3 General architecture of a RNN in reservoir computing.

This separation of computing and readout has inspired a wide range of physical implementations. Physical reservoir computing outsources the network to a substrate. The reservoir can be a robot body (, & al., ) , , & (). Locomotion without a brain: physical reservoir computing in tensegrity structures. Artificial Life, 19(1). 35–66. https://doi.org/10.1162/ARTL_a_00080 , a plant, a photonic circuit (, & al., , , , , , & (). Photonics for artificial intelligence and neuromorphic computing. Nature Photonics, 15(2). 102–114. https://doi.org/10.1038/s41566-020-00754-y ; , & al., ) , , , , , , , & (). Experimental demonstration of reservoir computing on a silicon photonics chip. Nature Communications, 5. 3541. https://doi.org/10.1038/ncomms4541 or a water bucket ( & , ) & (). Pattern Recognition in a Bucket. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-39432-7_63 . The idea behind (physical) reservoir computing is explained in depth in Introduction to Machine Learning and Reservoir Computing.

The main goal of this dissertation is to validate the use of plants for computation experimentally. By computation, we imply plant-centric types of computation instead of general-purpose computation, achievable using a conventional computer. Though , & al. () , , , , , , & (). Computers from Plants We Never Made: Speculations. InStepney, S. & Adamatzky, A. (Eds.), Inspired by Nature: Essays Presented to Julian F. Miller on the Occasion of his 60th Birthday. (pp. 357–387). Springer International Publishing. https://doi.org/10.1007/978-3-319-67997-6_17 have proposed to build a fully functional computer from plants, they will not replace computers in everyday life, but they can replace some conventional systems in tasks inherently related to plants, such as yield optimisation. , & al. () , , , , , , & (). Computers from Plants We Never Made: Speculations. InStepney, S. & Adamatzky, A. (Eds.), Inspired by Nature: Essays Presented to Julian F. Miller on the Occasion of his 60th Birthday. (pp. 357–387). Springer International Publishing. https://doi.org/10.1007/978-3-319-67997-6_17 describe several input-output relationships, but none seems to be practically scalable using currently available technology. Limiting the scope to tasks that are related to plants is probably relevant in practice. This hypothesis is based on similar work in robotics. In compliant robotics, there is considerable interest in outsourcing part of the control loop for locomotion to the body. It has been observed that sensory information from this robot body is highly suitable for locomotion tasks and terrain classification (, & al., , , , & (). Effect of compliance on morphological control of dynamic locomotion with HyQ. Autonomous Robots, 45(3). 421–434. https://doi.org/10.1007/s10514-021-09974-9 ; , & al., ) , , , & (). Terrain Classification for a Quadruped Robot. https://doi.org/10.1109/ICMLA.2013.39 .

Examining the computational properties of plants is not only interesting from a fundamental point of view but can also cause a shift of paradigm in phenotyping. Phenotyping is, loosely speaking, the characterisation of a plant’s traits. Reservoir computing can provide a new framework for plant physiological studies. Most studies focus on the interplay of one or two environmental variables on a plant, while a plant’s responses are the integrated sum of many more variables (, & al., ) , , , , , , & (). Pampered inside, pestered outside? Differences and similarities between plants growing in controlled conditions and in the field. New Phytologist, 212(4). 838–855. https://doi.org/10.1111/nph.14243 . After all, the environments in which plants grow are subject to constant change ( & , & (). Dynamic non-photochemical quenching in plants: from molecular mechanism to productivity. The Plant Journal, 101(4). 885–896. https://doi.org/10.1111/tpj.14601 ; , ) (). Plants and Microclimate: A Quantitative Approach to Environmental Plant Physiology (3). Cambridge University Press. https://doi.org/10.1017/CBO9780511845727 . These changes are not only the result of variable weather conditions but also of biological nature. Animals continuously interact with plants, resulting in, for instance, pollination, predation and even fertilisation. Moreover, the soil is also subject to constant changes and interactions due to parasitic and symbiotic relationships between the roots and nematodes, fungi and bacteria. So, instead of focussing on the relationship between one or two controlled variables and the plant’s responses, reservoir computing can provide a more general framework: the plant is a computational resource that processes these input signals in a certain manner and behaves accordingly.

Additionally, in the long-term, industrial applications in horticulture and agriculture are also possible. For instance, it is crucial to continuously monitor the plants’ conditions in greenhouses for optimal yields. Currently, growers rely on manual observation and known set-points of optimal growth due to the long-term experience of a farmer with a particular crop. This high degree of specialisation has led to ever-increasing yield yet also poses difficulties when switching cultivars. Growers need to be able to respond to shifts in the market rapidly to maximise profits. However, switching cultivars in this classical approach is not easy. It can take several growing seasons before production is optimised, partially due to the lack of detailed information on a plants’ state. Using the computation properties of plants can solve this issue because plants become active participants in optimising environmental conditions.

Obtaining information on plants’ state and using this in a control loop, as described above, can yield to highly autonomous plant systems. In such systems, plants can take over many responsibilities of the farmer, including irrigation, fertiliser application and climate control.

Research Outline #

As mentioned in the previous , the main goal is to validate computing with plants by means of a PRC-inspired setup. The tasks can be roughly divided into the following objectives: (i) investigate suitable sensing technologies to assess the plant’s state; (ii) use sensor data to evaluate PRC with plants; and (iii) investigate potentially interesting biological tasks.

Before we can dive into the details of reservoir computing with plants, one needs to get acquainted with the basics of machine learning, and ecophysiology and phenotyping. These subjects are covered in Introduction to Machine Learning and Reservoir Computing and Introduction to Plant Ecophysiology, Phenotyping and Phenomics respectively. Essential concepts are introduced from both fields, making the rest of this book accessible to experts from either discipline. Afterwards, in Reservoir Computing with Plants, we discuss how we can make use of physical reservoir computing to study the computational properties of plants. In Reervoir Computing with a Snapshot Hyperspectral Camera, a first attempt is made to study the plant behaviour under varying conditions with high temporal resolution using a hyperspectral camera. While this line of research was inconclusive, valuable information on the limitations of hyperspectral cameras was obtained for such experiments. Consequently, an alternative sensing technology was used for which a versatile data logging system was developed, Gloxinia (Development of a Sensor-Platform for Measuring Dynamic Plant Properties). This sensor platform was used in the experiments of Experimental Demonstration of a Plants as Computing Resource for Physical Reservoir Computing where several tasks are used to evaluate the computational properties of plants. Concluding remarks and future research are discussed in Discussion and Future Perspectives.

Visualisation of the dependencies between publications and the research subjects.
Figure 1.4 Visualisation of the dependencies between publications and the research subjects.

A list of all published work is included below. Most of these publications have been (partially) integrated into this book. Here, we approach each of these studies from a reservoir computing point of view. However, this is not the point of view in some of our publications. Figure 1.4 visualises the interconnections between different publications and research subjects.

In our initial attempt to investigate PRC with plants, we worked with a hyperspectral camera as the plant observing sensor. However, the results were inconclusive. Based on the data collected from this sensor, there was no significant difference between the background and plant observations. Nonetheless, we published these results in , & al. () , , , , & (). Limitations of snapshot hyperspectral cameras to monitor plant response dynamics in stress-free conditions. Computers and Electronics in Agriculture, 179. 105825. https://doi.org/10.1016/j.compag.2020.105825 , where we analysed the extractability of useful information on dynamic plant properties. Consequently, we shifted our focus to different sensing technologies. To that end, we developed Gloxinia, a custom sensing platform optimised for our future experiments. At the core of our experimental design is the rapid and continuous observation of plant properties such as leaf length and thickness. Such measurement systems are not very common, and possible implementations are costly. As a result, we designed a custom sensing platform with the experimental design in mind. We developed an open research platform named Gloxinia that is (i) cost-effective, (ii) accurate, (iii) modular and (iv) versatile. Our experimental validation study of PRC with plants is currently under review.

All of the main publications focussed on PRC experiments with physical substrates. Yet, we can also approach plant PRC using simulation models (also called digital twins) of plants. We published preliminary work on this in , & al. () , & (). Development of a quantitative comparison tool for plant models. Retrieved from http://hdl.handle.net/1854/LU-8677699 . The focus here is on using reservoir computing as a means to compare plant models.

Several side-track projects were also published as a conference paper or a journal article. , & al. () , , , , , , , , , & (). Solis: A Smart Interactive System for Houseplants Caring. IEEE. https://doi.org/10.1109/ICOT.2018.8705915 is based on a student project. Students developed a smart house plant monitoring system further developed by researchers in the Department of Industrial Systems and Product Design. , & al. () , , , , , , , & (). MIRRA: A Modular and Cost-Effective Microclimate Monitoring System for Real-Time Remote Applications. Sensors, 21(13). 4615. https://doi.org/10.3390/s21134615 is also the result of preliminary work by master students. This work discusses the development of a microclimate system, MIRRA.

We experimented with several extensions to linear regression such as group least absolute shrinkage and selection operator (LASSO) when analysing the data. This led to a contribution to an open-source software library Pyglmnet, published in , & al. () , , , , , , , , , , , , , , , , , , , , & (). Pyglmnet : Python implementation of elastic-net regularized generalized linear models. JOURNAL OF OPEN SOURCE SOFTWARE, 5(47). https://doi.org/10.21105/joss.01959 .