Link to paper

The full paper is available here.

You can also find the paper on PapersWithCode here.

Abstract

• Computational models can be applied to the biological world.
• View should be observer-dependent and pragmatic.
• Living systems can perform multiple functions in the same place at the same time (polycomputing).
• Understanding of meso-scale events can be improved with an observer-centered framework.
• Overloading of different functions on the same hardware is an important design principle.

Paper Content

Introduction

• Feynman argued that manipulating matter and energy at small scales could lead to vast technological progress
• Biology expands the adaptive function of an existing system by overloading mechanisms with multiple functions, which is called polycomputing
• Examples of polycomputing in biology include spider webs that serve as auditory sensors and prey capture devices, and holographic memory storage in the brain
• Examples of polycomputing in computer and materials engineering are emerging
• Polycomputing is defined as the ability of a material to provide the results of more than one computation in the same place at the same time
• The notion of “computer” needs to be expanded to include massively parallel, stochastic, emergent, and evolutionary systems
• The question of what constitutes a “living thing” is being redefined due to new chimeric, synthetic, and bioengineering techniques
• Polycomputing helps us understand, predict, and control new classes of evolved and designed materials

Current debates: dissolving dichotomous thinking

• Researchers look to nature for inspiration when technological progress slows
• Examples include photosynthesis and flapping wings
• Difficulty of packing more compute into microchips challenges researchers to seek new paths by considering how computation is embedded in living systems
• Debate about whether organisms are machines has a long history and has become more intense in recent years
• Machines capable of computation include robots and physical computers, but exclude simple mechanical devices

Structure function mapping and polycomputing.

• 1-to-1 mapping between structure and function is difficult to tease apart
• Advances in biology and computer science are demolishing long held dichotomies
• Machines can support different functions at different spatial scales and times
• Investigating polycomputing enables new questions for biology and a quest for engineers to understand how to pack more functionality into the same machine

Assumed distinction counterexamples

Software/hardware

• Physical materials can compute and learn
• Tape-less von Neumann self replicators
• Digital/analog circuits can exploit electromagnetic properties of the circuit’s substrate
• AI-designed organisms
• Intentional stance for automaton/free agent
• Brain/body computational metamaterials

Body/environment

• Cells in multicellular bodies have an environment
• AI technologies can pass verbal, visual, and physical Turing tests
• Artefacts can be created or evolved by human-created evolutionary algorithms

Dichotomous thinking in the life sciences.

• Biology does not support dichotomous categories.
• Evolution and developmental biology exhibit gradualism.
• There is no clear line between cognitive humans and quiescent oocytes.
• Categories represent poles of a spectrum of diverse properties.
• Interoperability of life enables chimeras of all levels of organization.
• Cognitive and logistical limitations lead to departmental, funding, and publication distinctions.
• Cognitive neuroscience techniques can be used to understand collective intelligence of cells.

Dichotomous thinking in computer science.

• Advances in computational sciences challenge the boundary between body and brain.
• Physical computing and morphological computing are technologies that are eating away at this distinction.
• Mechanical computing uses optical, mechanical, or quantum phenomena to perform calculations.

Polycomputing in bodies and brains.

• Polycomputing can be shuttled between biological bodies and brains and machine bodies and brains.
• Polycomputation requires storage of multiple computations at the same place and time, but at different peaks in the frequency spectrum.
• Digital circuitry and biological nervous systems cannot perform polycomputation as they traffic in electrons and chemicals/ions, not vibration.
• Neurons may communicate mechanically, which may contain vibrational components, suggesting nervous systems may be able to polycompute.
• Computer science shows how taking a causal stance at higher levels enables progress.
• Boolean networks can compute different functions simultaneously depending on the level of analysis chosen by an observer.
• Computational metaphors applied to brains have assumed tissues, cells, and other biological systems that are not brains do not compute.
• Studies of basal cognition in plants, single somatic cells, microbes, and tissues suggest the brain/body dichotomy is being dismantled.

Learning from superposed systems in engineering

• Computational density can be increased by computing and storing multiple results in the same place at the same time.
• Quantum computing can perform multiple computations simultaneously.
• Non-quantum forms of computational superposition suggest more computation can be packed into smaller spaces.
• Holographic data storage stores data dispersed across the storage medium.
• Physical reservoir computing can extract desired computations from inert materials.

Biology is massively overloaded: polycomputing

• Analyzing natural systems to determine polycomputation is challenging
• Traditional engineering design principles are not effective
• Changes to optimize one function can interfere with another
• AI methods can design polycomputable technologies
• Evolutionary algorithm used to design granular metamaterial
• Many biological functions can be analyzed as computations
• Evolutionary algorithm can be used to evolve metamaterials
• Metamaterials can act as AND and XOR gates simultaneously

Table 2: examples of biological polycomputing at diverse scales multiple computations in the same biological hardware reference

• Mitochondria act as micro-lenses in photoreceptors
• Proteins act in multiple conformations
• Gene regulatory networks have multiple memories/behaviors
• Chemical networks perform neural network tasks
• RNA encodes enzyme and protein functions
• DNA has more than one active reading frame
• Ion channels are also transcription factors
• DNA transcription factors work in DNA replication machinery

Polycomputing and the range of phenotypic possibility.

• Xenobots are proto-organisms made from frog skin cells
• They can self-assemble and move autonomously
• They can even self-replicate
• Evolution teaches us how to make systems with cells that can be coerced into different shapes and behaviors
• Evolution creates generic problem-solving machines
• Evolution exploits the competencies of cells to create adaptive outcomes

Evolving polycomputing.

• Evolution is slower without a multiscale competency architecture
• Cells can make an eye in aberrant locations and still provide vision
• Competency of substrate in regulative development and remodeling can neutralize lethal side effects of mutations
• Monocomputational developmental architecture would prevent exploration of beneficial effects of mutations
• Polycomputing architecture allows exploration of consequences of mutations
• Cells and tissues allow exploration of other effects of mutations on phenotype
• Machines harness laws of physics, computation, etc. to produce specific, useful outcomes
• Evolution exploits fact that life is a machine by making changes to material, control algorithm, and environment
• Polycomputing architecture produces plasticity, robustness, and novelty

A new approach to identifying and harnessing computational capabilities in vivo and in silico.

• GRNs are a formalism used to study gene regulation
• Evolution enriches for learning capacity in biological networks
• Biological networks can offer multiple different types of memory and computations
• Evolution searches for ways to exploit existing components as features of the environment
• Brain stores multiple memories in the same neuronal real-estate

Conceptual transitions

• Overcome human cognitive bias
• Resist temptation to cleave phenomena apart
• Look for non-intuitive phenomena
• Seek gradients to move from “obvious” approximations to more accurate reflections of reality

Directions of conceptual travel.

• Tasks can be performed in parallel or serially
• Human thinking is mostly serial
• Traditional parallelism assumes multiple processes are coincident in time but not in space
• Superposition is the performance of multiple functions in the same place at the same time
• Modular processes are usually preferred in engineering, but natural phenomena are often continua

Practical implications for ai/robotics.

• Learning how biological systems polycompute can create more efficient AI technologies and robots
• Polycomputing technology may be more compatible with biological components, creating biohybrids
• Polycomputing may lead to machines that perform multiple functions in the same place at the same time
• Polycomputing may provide a solution to catastrophic interference
• Gradual computing in biology is important to track the emergence of specific features
• Boot-up of a biologically-embodied intelligence involves polycomputing
• Transition from analog device to computer is a shift in the relative payoffs for two different formalisms
• Recognizing the degree of agency in potential interaction partners is important
• Polycomputing systems may be capable of acceding to multiple requests simultaneously

Becoming a computer

• Computers are not just physical machines, but a theoretical construct known as a Turing Machine
• Turing Machines can be embodied by organisms, fluids, algorithms, and other physical systems
• It can be difficult to tell when a system transitions from “just physical materials” to a computer
• Biological Turing Machines may not have a localized tape
• Systems can change over time, making it difficult to determine when it becomes a computer
• Polycomputing systems can provide different results to different observers, making it difficult to determine if it is a computer

Conclusion

• Biological systems are computers, but not the same as today’s computers
• Biological systems are powerful polycomputing devices
• Blanket pronouncements about what living or non-living machines can do are worthless
• Control of multiple scales of biological organization is essential
• Computational frameworks are needed to facilitate control
• Synthetic, evolved, and hybrid systems far outstrip our ability to predict
• Abolishing absolutist categories and objective views of computation is necessary
• Biological systems are capable of polycomputing
• Biological components are massively overloaded towards polycomputing
• Biological systems are capable of second-order polycomputing
• Biological systems are capable of modeling each other’s computational behavior
• Biological systems are capable of optimizing their behavior based on expected reward
• Multiple computational models can be true of a biological system at the same time
• There is no one-to-one mapping between biological form and function