Source-linked AI summary
Plant-Inspired AI: Plants as Inspiration for Novel Problem Formulations, and Two Case Studies
Deepayan Sanyal, Joel Michelson, Carla E. Cao, Adam B. Roddy, Maithilee Kunda
TL;DR
Existing AI problem formulations leave some computational problems suggested by complex plant behavior unaddressed. The paper abstracts two plant phenomena into new formulations and preliminary tasks, then discusses their broader applications and limitations. It concludes that plant-inspired formulations expose distinctive challenges involving distributed information gathering, growth, resource allocation, and heterogeneous components.
Problem
Existing AI problem formulations such as supervised learning, tree search, and constraint satisfaction do not address some problems suggested by complex plant behavior.
Method
The paper abstracts leaf mimicry and coordinated root-shoot growth into two AI problem formulations and develops preliminary computational task formulations.
Results
The two formulations identify unaddressed computational tasks and suggest applications beyond plants, including collective mimicry and root-shoot collaboration problems.
Takeaways & Limitations
Plant behavior can serve as inspiration for AI formulations involving distributed information gathering, growth, coordinated resource acquisition, and resource allocation.
Takeaways & Limitations
The paper notes that no established method currently evaluates problem-formulation quality, which is ultimately validated retrospectively through adoption and usefulness.
Abstract
from arXiv · showhide
Artificial Intelligence (AI) has long been inspired by studies of biological intelligence. Reinforcement learning, for instance, drew inspiration from studies involving animal learning and is now a powerful paradigm for solving many real-world problems. Recently, plant biologists have uncovered a wide range of complex behaviors in plants that enable them to flexibly adapt to variable environments. Here, we argue that such behavior can motivate new AI frameworks encompassing a range of problems overlooked by existing problem-solving frameworks such as supervised learning, tree search, and constraint satisfaction. We illustrate this idea with two examples of intelligent problem-solving in plants: (1) leaf mimicry in Boquila trifoliolata, a vine capable of altering its leaves' morphology to resemble those of multiple host trees simultaneously; and (2) coordinated root-shoot growth, wherein plants allocate resources across organ systems exploring distinct environments. While leaf mimicry is highly specific to Boquila, coordination of root-shoot growth is shared across most plants. For both examples, we capture underlying computational principles and identify problems fitting these frameworks that are currently unaddressed by AI. Finally, we outline preliminary task formulations and discuss how these formulations may be applied to non-plant problems.
1. Introduction
The paper argues that plant behavior can inspire new AI problem formulations, extending the tradition of deriving useful abstractions from biological intelligence. It develops two plant-inspired formulations and identifies computational tasks that existing AI formulations do not address.
- Problem formulations are mathematical abstractions that capture useful structure across classes of problems.The paper presents formulation design as a crucial part of AI research because formulations can generalize beyond their original motivating problems.
- Biological intelligence has inspired powerful AI formulations, including reinforcement learning and shortest-path problems.The paper links reinforcement learning to animal trial-and-error learning and shortest-path problems to navigating a maze.
- Plant behavior offers a source of AI formulations involving different means of perception, communication, and decision-making.Plants lack centralized nervous systems and coordinate resource acquisition and growth through distributed chemical processes.
- Leaf mimicry in Boquila trifoliolata motivates a formulation whose key characteristics include substrate plasticity and relative localization.The formulation is inspired by a vine that can alter leaf morphology to resemble multiple host trees.
- Cooperation between many root tips and shoots motivates a formulation characterized by resource allocation and heterogeneous roles.This formulation draws on coordination between organ systems that acquire resources in distinct environments.
- The paper identifies specific computational tasks in both case studies that remain unaddressed by existing AI problem formulations.It also discusses limitations in generating the formulations and directions for future work.
2. Related Work: Computational Modeling of Plant Behavior
Prior work models, explains, or replicates plant behavior using approaches including predictive processing, active inference, mathematical growth models, optimization, and distributed computation. This paper instead uses plant behavior to identify broader classes of AI problems.
- Plant-behavior research includes predictive-processing and active-inference accounts of perception, sampling, and adaptive behavior.These approaches propose computational mechanisms that could underlie complex plant behavior.
- Mathematical models have represented plant behavior, growth, tropisms, stimulus integration, and root-shoot resource allocation.The cited work includes input-output models, perceptron-like growth models, and models of balances between root and shoot growth.
- Plant phenomena have also inspired metaheuristic optimization methods for complex optimization problems.A cited review covers 13 examples of plant-inspired metaheuristic optimization methods.
- Other research treats plant organs and cellular units as distributed computational systems or semi-autonomous modules.These perspectives emphasize information processing and collective control of plant growth and development.
- Unlike prior work focused on explaining, modeling, or replicating plant behavior, this paper uses plants to identify new AI problem classes.Its formulations abstract away biological details and target broader computational problems, including perception-morphology and resource-allocation-morphology interdependence.
3. Case Study 1: Leaf Mimicry in Boquila trifoliolata
Boquila trifoliolata’s leaf mimicry motivates a problem formulation in which distributed cells must perceive, communicate, and grow into a target shape under irreversible structural constraints.
- 3. Case Study 1: Leaf Mimicry in Boquila trifoliolata: Boquila trifoliolata can alter leaf morphology to mimic multiple host species simultaneously, reducing herbivory.The mechanism underlying this behavior remains unclear.
- 3.1 Collective mimicry as a Problem Formulation: The collective mimicry formulation asks a cell graph G to maximize similarity M(G, T) to a target graph T.Cells can create new cells and signal information about target features to neighbors and future selves.
- 3.2 Key Problem Characteristics: The formulation makes substrate plasticity central: new cells become feature detectors and communicators, while growth decisions irreversibly alter future information and action possibilities.Perception and morphology therefore cannot be treated as distinct solution components.
- 3.2 Key Problem Characteristics: Relative localization requires each independently acting cell to coordinate with others to infer how its position corresponds to the host leaf’s layout.Global mimicry depends on aggregating local observations into a representation of the target.
- 3.3 A Computational Collective Mimicry Task: The computational task instantiates one cell population that grows and coordinates local decisions to mimic an unseen 28×28 MNIST digit.Each cell receives a scalar target-image observation and neighboring hidden states; at n = 8, the observation size is 41.
- 3.3 A Computational Collective Mimicry Task: The task emphasizes communication under impoverished imaging, irreversible commitment, and limitations absent from the setup, including feedback, relative target location, selective communication, and separate memory.Off-the-shelf multi-agent RL may not overcome the observability, communication, substrate-plasticity, and relative-localization challenges.
4. Case Study 2: Collaborative Growth of Roots and Shoots
The root-shoot collaboration framework models plant growth as decentralized coordination between heterogeneous cells that acquire complementary resources, exchange them, and balance expansion across root and shoot populations.
- Biological motivation: Roots and shoots acquire complementary resources from distinct environments, with roots obtaining water and minerals while shoots capture light and atmospheric CO2.Whole-plant growth depends on continuous exchange between these spatially separated systems.
- Problem formulation: The framework represents a growing cell complex whose cells sense local resources, maintain internal buffers, communicate with neighbors, and create new cells when requirements are met.The formalization includes environmental resource distributions, cell states, maintenance costs, actions, and a utility function.
- Key problem characteristics: Root and shoot cells have distinct resource-access constraints and growth requirements, so successful growth requires interdependent local policies and resource transfers.No single cell type can independently acquire all resources needed for continued expansion.
- Objective and benchmark: The benchmark maximizes plant size while penalizing root-shoot imbalance, using O = N/(1 + B), where N = Nroot + Nshoot and B = |Nroot−Nshoot|/|Nroot+Nshoot|.Root cells require Carbon, Water, and Nitrogen for creation, whereas shoot cells require Sunlight, Nitrogen, and Carbon.
- Challenge: Locally beneficial growth decisions can harm the whole complex because allocating resources to one population may reduce resources available to the other.Cells must learn when to store, transfer, or consume resources as the evolving structure changes.
- Scope boundary: The task omits spatial foraging and most morphological constraints, simplifying the benchmark but leaving these as directions for harder decentralized-growth tests.The cell complex need not decide where to grow in response to heterogeneous resources, and growth does not model support or overcrowding.
5. Related Work: Similar PFs in AI
Prior AI research provides related formalisms for collaborative decentralized decision-making, including extensions of MDPs and POMDPs with decentralized control and network structure.
- Decentralized decision-making: Decentralized MDP and POMDP extensions study collaborative decisions among groups of agents, including networked interactions and locality exploitation.Prior work also examines differences in computational complexity between decentralized and centralized control.
6. Discussion
The discussion positions the two plant-inspired problem formulations as distinct abstractions whose value depends on the structures they expose, their applicability, and their eventual usefulness to AI researchers.
- Problem formulations: The paper presents two distinct plant-inspired problem formulations with overlapping key problem characteristics but different challenges and mappings to applications.Their design constrains how they can be represented and solved.
- Future directions: Although both case studies use decentralized, cell-based computations and graph structures, plant-inspired formulations could also be continuous or physically embedded.Other organizational levels may inspire additional formulations.
- Evaluation: The paper notes that no established method evaluates the quality of a problem formulation; retrospective adoption and usefulness ultimately validate one.Potential design properties include broad expressiveness, distinctive structure, and support for effective algorithms.
- Applications: The formulations can describe non-plant applications including software-library conflict resolution, urban design, communications infrastructure, and renewable-energy transitions.The collective-mimicry and root-shoot abstractions correspond to different application structures.
- Exposed structure: Collective mimicry abstracts distributed information gathering and growth, while root-shoot collaboration abstracts coordinated acquisition, allocation, and heterogeneous roles.These characteristics are placed at the forefront as first-order problems.
7. Appendix
The appendix surveys plant behaviors that could motivate additional AI problem formulations, spanning resource management, adaptive growth, environmental sensing, and decision-making across species.
- Common plant behaviors: Plants pace metabolism by dividing starch reserves across the remaining hours of darkness and recalculating when night length changes.This behavior offers an example of arithmetic under changing temporal constraints.
- Common plant behaviors: Plants can sacrifice heavily shaded leaves and redirect resources toward vertical growth and defenses in valuable sunlit leaves.The modular defense strategy manages limited carbon stores.
- Common plant behaviors: Upper shoots alter leaf orientation when reflected-light spectra indicate lower-canopy shading, changing canopy architecture to improve light penetration.The response uses the red-to-far-red ratio as an environmental signal.
- Broad species groups: Fast-growing herbaceous plants choose predictable nutrient patches when resources are adequate but gamble on unpredictable patches when nutrient-starved.The strategy changes with resource availability and environmental uncertainty.
- Broad species groups: Tropical climbing vines first grow toward darkness to locate large hosts, then reverse direction toward light after contact.The behavior combines sequential environmental search with a strategy switch.
- Broad species groups: Mixed-mating plants adjust flower form according to conditions, producing large insect-pollinated flowers when favorable and small self-pollinating flowers otherwise.The reproductive strategy responds flexibly to environmental quality.
- Single species or small subsets: Dodder vines evaluate airborne host volatiles to trade off host nutritional quality against reachability before attachment.This is an example of cost-benefit selection in a rootless parasitic plant.
- Single species or small subsets: Venus flytraps use touch counts and timing thresholds to trigger closure, digestion, and proportional digestive-enzyme production.Different numbers of touches within specified intervals produce different responses.