Whether genuinely lifelike organization can originate, persist, and be detected inside a computer — 'virtual life' — remains open. Susan Stepney's 2025 overview in Philosophical Transactions of the Royal Society B synthesizes an engineering-focused definition of life (abstract requirements: autopoiesis, agency, and open-ended adaptation; generic designs; implementation mechanisms) and concludes that virtual artificial life satisfying all these requirements 'is yet to be exhibited', while the field still lacks good definitions and well-defined ways to detect and measure life in silico. Component-level progress — foundation-model search for lifelike simulations, spontaneously emerging self-replicating programs, new large-scale simulation engines, and evolutionary-activity statistics — leaves the integrated core untouched. The problem asks for definitions of life precise enough to apply to purely computational substrates, detection and measurement criteria for recognizing lifelike organization that is produced endogenously rather than supplied by programmer 'shortcuts', and the exhibition and validation of a complete living virtual being or living virtual system satisfying those definitions through its own endogenous mechanisms.

Background

Artificial life (ALife) is the research field that studies 'life as it could be' — living or life-like systems beyond the particular instance we know on Earth — in contrast to biology's study of 'life as we know it to be'. Its subject matter spans synthetic biology, xenobiology (biology built from non-standard or artificial biochemistry), soft and hard robotics, inorganic chemistry, and virtual life: organization that exists entirely inside a computer, in a computational substrate rather than a material one. The founding intuition, going back to von Neumann's self-reproducing automata, is that if the essential properties of life can be captured computationally, they should be able to arise and persist in silico.

Three ingredients must come together for that intuition to be realized, and all three are unsettled. First, definition. There is no consensus definition of life after decades of proposals. A recent framing (Stepney 2025) separates an engineering-style definition into levels: abstract requirements, generic designs that could realize them, and concrete implementation mechanisms. The requirements level takes the defining properties of life to be autopoiesis, agency, and open-ended adaptation. Autopoiesis (introduced by Varela, Maturana and Uribe) is the property of a system that continuously produces and maintains its own components and constructs its own boundary between self and environment. Agency is the capacity to interact with the environment in pursuit of the system's own goals through purposeful behaviour. Open-ended adaptation is continual adaptive novelty in response to an open, changing environment, without convergence to a fixed endpoint. On this view a 'living being' is an autopoietic system that has agency and adapts; a 'living system' is a collection of living beings open-endedly co-adapting with one another and with a non-living environment. These requirements are for 'full' life; systems meeting only some of them may be 'partially alive', and alternative definitional frameworks, such as the 'lyfe' of Bartlett and Wong (2020) — any system fulfilling four processes: dissipation, autocatalysis, homeostasis, and learning — continue to be debated for non-standard life (Wong et al. 2022).

Second, detection and measurement. Since any candidate virtual life is unprecedented, one needs criteria by which to recognize it if and when it appears. The standard tool for detecting ongoing adaptive innovation has been the evolutionary activity statistics of Bedau and colleagues: cumulative measures of new, persisting, functional 'components' that survive selection, computable from logged runs of digital evolutionary systems. But Stepney and Hickinbotham (2024) argue a fundamental limitation: a genuinely open-ended system will eventually move outside any fixed model of its behaviour, and hence outside any measure based on that model, so analysis must proceed from system-generic to system-specific measures and focus on mechanisms rather than a single quantification of open-endedness. For virtual systems one can additionally ask which life requirements are met by endogenous mechanisms and which are supplied externally by the programmer as 'shortcuts', which suggests quantifying 'how alive' a system is (Stepney 2025).

Third, origination. Existing systems each realize fragments. Digital organisms in Tierra and Avida (two classic platforms in which populations of self-replicating computer programs mutate and evolve) evolve in artificial chemistries (abstract computational systems of interacting, transforming 'molecules' such as programs or strings); computational autopoiesis implements self-producing bounded organization; neuro-evolution (evolving neural-network controllers by mutation and selection) and quality-diversity search (collecting many behaviourally distinct solutions rather than a single best one) produce adaptive behaviours and morphologies; and self-replicating programs have been shown to emerge from random interaction in simple programming-language substrates (Agüera y Arcas et al. 2024). Bedau et al. (2000) distilled fourteen open problems of ALife, including 'Achieve the transition to life in an artificial chemistry in silico'; a subsequent review (Aguilar et al. 2014) reported that all fourteen remain open. Recent work accelerates the search side: vision-language foundation models (large neural networks trained on image-text data, used as automatic judges of interestingness) now search simulation spaces for 'life-forms' and open-ended dynamics across substrates such as Lenia (a continuous-valued cellular automaton), Boids (an agent-based flocking simulation), and neural cellular automata (grid systems whose update rules are neural networks) (Kumar et al. 2025), and new engines target large-scale open-ended evolutionary simulation (Tensen et al. 2026). Yet none of these satisfies all the requirements at once; Stepney (2025) concludes that virtual ALife satisfying all these requirements is yet to be exhibited, with at best 'partially alive' systems, and that the obstacles include fragile symbolic self-maintenance mechanisms, the absence of a common design framework for composing life's processes, and virtual worlds that are 'at best in a relatively simple world, at worst in a virtual vacuum'.

A further complication is the possibility question. Rosen argued, from his definition of an organism as a material system 'closed to efficient causation' (every component's causal organization is produced within the system itself), that software running on hardware is a machine and therefore not alive — so virtual life is impossible. Counterarguments from the autopoiesis and semantic-closure traditions (semantic closure being a system's capacity to build internal signs that acquire meaning and causal power for the system itself) keep this debate unresolved, and Stepney (2025) assumes virtual life is possible for the purposes of the overview while noting the argument 'still rages'. The Royal Society theme issue 'Origins of life: the possible and the actual' (Solé, Kempes and Stepney 2025) places the virtual-life question at the 'possible' extreme of origins-of-life research: whether lifelike organization can arise in a computational medium, as opposed to a simulation of material life, depends on the chosen definition of life, and a requirement-based definition does not exclude it.

Problem Statement

The problem is to make virtual life real and recognizable. It has three jointly linked parts, all drawn from how the primary sources state the gap. (1) Definitions: formulate definitions of life and of its requirements — such as autopoiesis, agency, and open-ended adaptation, stated at the requirements level and separated from design and implementation levels — precise enough to be applied to purely computational systems. These definitions retain the full content of all three requirements in the sense characterized in the Background, including open-ended adaptation as continual adaptive novelty responding to an open, changing environment without convergence to a fixed endpoint; a system meeting only some of the requirements is 'partially alive' in the primary source's terms and does not thereby satisfy part (3). These definitions are to be fixed and justified on their own terms, independently of the system exhibited under part (3), with each requirement's definition tied explicitly to the Background's characterization of that requirement (autopoiesis, agency, open-ended adaptation), so that the exhibited system is measured against the definitions rather than certified by them. (2) Detection and measurement: develop well-defined criteria, measures, and protocols by which life-like organization can be recognized and its degree assessed in an evolving computational system, including the discrimination between properties produced endogenously by the system and those supplied externally by programmer 'shortcuts'. These criteria are to be stated in a form that others can apply to a candidate system and its logged runs, rather than reverse-engineered to certify the particular system exhibited under part (3). (3) Origination: exhibit, and validate against those criteria, a virtual system — 'a complete living virtual being, or living virtual system' — that satisfies the requirements-level definition of life through its own endogenous mechanisms, thereby determining whether genuinely lifelike organization can originate and persist inside a computer. The parts form one scientific objective: the definitions and detection criteria are what make the exhibited system recognizable as alive, so an answer that supplies the system without the recognition apparatus (or vice versa) does not resolve the problem as stated.

References

  1. Susan Stepney (2025), 'Towards origins of virtual artificial life: an overview', Philosophical Transactions of the Royal Society B: Biological Sciences 380(1936):20240298. DOI: 10.1098/rstb.2024.0298. URL: https://doi.org/10.1098/rstb.2024.0298
  2. Wendy Aguilar, Guillermo Santamaría-Bonfil, Tom Froese, Carlos Gershenson (2014), 'The Past, Present, and Future of Artificial Life', Frontiers in Robotics and AI 1:8. DOI: 10.3389/frobt.2014.00008. URL: https://doi.org/10.3389/frobt.2014.00008
  3. Mark A. Bedau, John S. McCaskill, Norman H. Packard, Steen Rasmussen, Chris Adami, David G. Green, Takashi Ikegami, Kunihiko Kaneko, Thomas S. Ray (2000), 'Open Problems in Artificial Life', Artificial Life 6(4):363-376. DOI: 10.1162/106454600300103683. URL: https://doi.org/10.1162/106454600300103683
  4. Ricard Solé, Christopher Kempes, Susan Stepney (2025), 'Origins of life: the possible and the actual', Philosophical Transactions of the Royal Society B: Biological Sciences 380(1936):20240281. DOI: 10.1098/rstb.2024.0281. URL: https://doi.org/10.1098/rstb.2024.0281
  5. Alan Dorin, Susan Stepney (2025), 'To Engineer an Angel, First Validate the Devil: Analyzing the "Could Be" in Artificial Life's "Life as-It-Could-Be"', Artificial Life 31(4):397-400. DOI: 10.1162/ARTL.e.452. URL: https://doi.org/10.1162/ARTL.e.452
  6. Akarsh Kumar, Chris Lu, Louis Kirsch, Yujin Tang, Kenneth O. Stanley, Phillip Isola, David Ha (2025), 'Automating the Search for Artificial Life With Foundation Models', Artificial Life 31(3):368-396. DOI: 10.1162/ARTL.a.8. URL: https://doi.org/10.1162/ARTL.a.8 (also arXiv:2412.17799, https://arxiv.org/abs/2412.17799)
  7. Blaise Agüera y Arcas, Jyrki Alakuijala, James Evans, Ben Laurie, Alexander Mordvintsev, Eyvind Niklasson, Ettore Randazzo, Luca Versari (2024), 'Computational Life: How Well-formed, Self-replicating Programs Emerge from Simple Interaction', arXiv:2406.19108. URL: https://arxiv.org/abs/2406.19108
  8. Susan Stepney, Simon Hickinbotham (2024), 'On the Open-Endedness of Detecting Open-Endedness', Artificial Life 30(3):390-416. DOI: 10.1162/artl_a_00399. URL: https://doi.org/10.1162/artl_a_00399
  9. Théo de Pinho, Lana Sinapayen (2026), 'A speciation simulation that partly passes open-endedness tests', arXiv:2603.01701. URL: https://arxiv.org/abs/2603.01701
  10. Mark Tensen, Ciaran Regan, Bert Wang-Chak Chan, Mizuki Oka, Kenneth O. Stanley, Grisha Szep (2026), 'Microcosmos: Reimagining Artificial Life for the GPU Era', arXiv:2607.02954. URL: https://arxiv.org/abs/2607.02954
  11. Stuart Bartlett, Michael L. Wong (2020), 'Defining Lyfe in the Universe: From Three Privileged Functions to Four Pillars', Life 10(4):42. DOI: 10.3390/life10040042. URL: https://doi.org/10.3390/life10040042
  12. Michael L. Wong, Stuart Bartlett, Sihe Chen, Louisa Tierney (2022), 'Searching for Life, Mindful of Lyfe's Possibilities', Life 12(6):783. DOI: 10.3390/life12060783. URL: https://doi.org/10.3390/life12060783
  13. Wolfgang Banzhaf, Bert Baumgaertner, Guillaume Beslon, René Doursat, James A. Foster, Barry McMullin, Vinicius Veloso de Melo, Thomas Miconi, Lee Spector, Susan Stepney, Roger White (2016), 'Defining and simulating open-ended novelty: requirements, guidelines, and challenges', Theory in Biosciences 135(3):131-161. DOI: 10.1007/s12064-016-0229-7. URL: https://doi.org/10.1007/s12064-016-0229-7