In recent decades, the proliferation of computational systems around the world has led to a vast increase in the information processing capacities of the planet. This form of planetary computation has revealed the Earth to itself, as in the case of climate change, a kind of planetary-scale evolution rendered visible only via simulations. It would be remiss to tie the emergence of computation to the emergence of computers, however. As biologist Michael Levin and others point out, computation is as old as life itself. Genes, proteins, cells all engage in the ongoing computing of information

. Planetary computation, thus, does not mark a radical break from the “natural” evolution of the world, but rather a continuation of the informatic evolution of the planet

The core role of information in the evolutionary history of Earth was famously posited in John Maynard Smith and Eörs Szathmáry’s seminal 1995 book The Major Transitions in Evolution

. For the authors, evolution has been characterized by a series of major transitions, each of which involved a fundamental change in both the unit of the individual (such as the evolution of multicellularity), and the information processing capacity of the system (such as the development of inheritable genetic code).

This theory framed evolution not as a continuous linear development, but rather as a series of discrete phase changes, each of which radically altered the possibility space for evolution to come

. While the first transition marks the shift from a “prebiotic soup” to protocells (also known as the “origin of life”), the authors present the most recent transition as the introduction of natural language, which enabled societies to pass information across much broader scales of space and time.

What if a new major transition was underway? The transition from natural to machinic language over the past century may mark a step change in complexity capable of re-imagining the evolutionary dynamics of the planet

. Just as planetary computation has disclosed its own impacts on the planet, might it also be capable of examining its own emergence as a process of life undergoing perpetual adaptation – a process that researchers call open-ended evolution

The Xenobiology Multiplex, or Xenoplex, sets out to explore these questions. This speculative experimental design leverages the most recent transition (planetary computation) in order to elucidate the first transition (origins of life). By simulating origins of life in a chemical-computational array, the project seeks ultimately to understand the mechanisms driving open-ended evolution on and of the planet. As the biosphere and technosphere increasingly constitute each other, an evolutionary theory that encompasses both biotic and abiotic systems is paramount. This project thus traces the evolution of evolution itself

, from biochemical origins to the computational present.

Experimental Design

The Xenoplex acts as an evolutionary time-compression device. Much like particle accelerators enabled the simulation of particles from the Big Bang, the Xenoplex enables us to run highly compressed micro-evolutionary histories in parallel.

The range of possible chemical constituents and temperatures from the “prebiotic soup” form the initial seed conditions for each box in a large array. Crucially, these represent not just the chemical conditions of early Hadean Earth, but also the biosignatures observed on exoplanets via new telescopes, such as the James Webb. This experiment allows us to examine universal principles of open-ended evolution across both origins of biotic life and astrobiology, paralleling attempts to create grand unified theories that accommodate both particle physics and astrophysics. This is crucial for exploring the vast space of counterfactual ‘tapes of life’

, beyond the perceptual doors of our version of evolution to experiment on unrealized biological trajectories. Going outside of our known story is the only way to survey the interplay of chance and necessity

This interplay is studied by computational means. Open-ended evolutionary algorithms are designed to explore the possibility space of environmental transformations that may have led to the origins of life. Chemical and temperature gradients are believed to form a crucial precondition; such disequilibria harbors the thermodynamic free energy required for self-organizing behavior to occur. Environmental complexity poses the first problem to solve for emerging life

. The solutions to this optimization problem for dissipating free energy gradients can be explored using parallelized gradient descent algorithms, allowing different parameterizations of the system to be rapidly searched and selected for in the arrays.

Rather than being a purely computational simulation, the Xenoplex empirically tests chemical environments. This avoids the dimensionality reduction of artificial chemistry simulations, and instead leverages the intrinsic reactivity, and creativity of chemical systems. This yields a form of recursive transfer learning, where chemical knowledge guides computational knowledge and vice-versa. This is crucial, as designed systems remain incapable of the ongoing open-ended evolution exhibited by natural and cultural systems. Solving this problem has long been a hallmark of Artificial Life (ALife), a broad field of study attempting to recreate lifelike processes using computational methods as simulations

. While an exact metric for measuring open-endedness remains elusive, generally such processes are characterized by the ongoing generation of (1) adaptive novelty, or (2) complexity.

If “life” were to emerge in the Xenoplex, how would we be able to recognize it as such? First, each reactor box must be carefully constructed such that the box itself interferes minimally with the simulated reactions. This self-contamination, or leakage of life, is epitomized by the iconic Miller-Urey experiments that studied the origins of life in the 1950s, where their use of borosilicate glassware inadvertently helped catalyze pre-biotic reactions


Second, the dynamics within each box must be quantified. Every reactor is connected to a mass spectrometer which samples the chemical space of each box. Fixed objective functions and equilibria are likely unrealistic, heuristic limitations that nature neither cares for nor obeys. Thus, the Xenoplex leverages a variety of metrics.

One possibility is the molecular assembly index, which was developed by astrobiologist Sara Walker with chemist Lee Cronin, as a measure of chemical complexity

. Derived from their concept of “assembly theory,” the molecular assembly index can be experimentally measured using conventional mass spectrometry, which produces chemical spectra where more peaks suggest more complexity and thereby more lifelike-processes.

Autocatalytic behavior (measured through copy number) and novelty (measured through diversity of chemical species) can also be measured, enabling a more open-ended evolution experiment to be designed with various nested goals that are dynamically adjusted. As these new specialized properties (e.g. self-replication) are acquired by populations of these boxes, the conditions of both the chemical gradient and computational gradient descent get propagated to nearby boxes, thus optimizing different properties in parallel.


The outcomes of the Xenoplex remain highly uncertain. However, five possible scenarios may emerge:

Scenario One: Null Result

A null result, where measures of lifelikeness plateau before self-replicating order emerges in any of the reactors. 

This implies the development of life requires deeper interactions with planetary processes, as microbiologist Lynn Margulis and chemist James Lovelock have argued in the Gaia hypothesis. Thus, life relies on tightly coupled feedback loops across multiple scales that are not easily reproduced in a lab


Scenario Two: Great Perceptual Filter

Life is observed only from prebiotic terrestrial simulations, but not exobiotic ones.

This could imply that life on Earth is universally unique, but a more compelling explanation is that the array still lacks the requisite technological sophistication for observing, simulating, or evolving non-canonical life forms. A great perceptual filter, as coined by Sara Walker, prevents the so-called “shadow biosphere” or xenosphere to be noticed

– akin to the invisibility of microbial life before microscopes were developed in the 1600s.

Scenario Three: Cosmoevolutionary Convergence

All of the successful boxes, including those with exoplanet initial conditions, are found to evolve some kind of biological life similar to Earth. 

This could illustrate cosmological-scale convergent evolution analogous to the terrestrial phenomenon of carcinisation where evolutionary trajectories often converge independently towards crab-like forms

. Alternatively, this scenario suggests theories of panspermia, where a common biotic ancestor “seeded” planets throughout the universe with life.

Scenario Four: Simulated Contact

A form of simulated contact in which life-like structures successfully evolve from experiments that are constrained by the exoplanet biosignatures.

This simulated contact may reveal forms of life that are entirely non-biotic – for example, convection cells that evolve the capability to detect, compute, and consume thermal information to replicate

. Such forms would likely require measurements of lifelikeness beyond chemical content for the algorithmic feedback loops; the emergence of this artificial tornado life would not be observed by a mass spectrometer.

This case might also represent a kind of xeno-xeno-life, where the lifeforms generated in the array are themselves a simulacrum constructed from putative biosignatures –  such that the alien life in the box no longer resembles those that actually exist on the exoplanet upon which it was modeled.

Scenario Five: Algorithmic Life

The most lifelike phenomenon that emerges is not found as a material substrate, but instead the algorithm itself that takes on a collective life of its own.

One of the hallmark goals of Artificial Life research is the discovery of the mechanisms driving open-ended evolution. By enabling various general characteristics of lifelikeness, such as novelty, complexity, and self-replication, to be selected in parallel, open-endedness might emerge through a kind of meta-learning process across all of the simulation boxes.

The discovery of such a mechanism would have deep ramifications for the development of artificial general intelligence as the next major evolutionary transition. Goal-oriented optimization and training tasks from machine learning are currently incapable of the exploratory dynamics we associate with evolution. A mechanistic understanding of such a process applied to neural networks would present a fundamental paradigm shift in how models train, learn, and replicate.

Regardless of the outcome, these simulations present an opportunity to decenter life away from a terrestrial, biocentric view. What is life if it is decoupled from the earth, or even from the realm of the biotic itself? Or conversely, what is life if it is so seamlessly integrated into its planetary realm that it is impossible to draw a separation between life and environment

? The Xenoplex does not promise a clean resolution, but rather further alienation. This is precisely the kind of otherworldly, uncomfortable zone that we must delve into for knowledge to expand itself.



约翰·梅纳德·史密斯(John Maynard Smith)和艾尔什·萨斯玛里(Eors Szathmáry)于 1995 年出版的著作《进化之重大转变》(The Major Transitions in Evolution)著名地提出了信息在地球进化历史中的核心作用。作者们认为,进化的象征在于一系列重大、历史性的转变。每个转变都涉及个体单元(例如多细胞生物的进化)和系统信息处理能力的根本性变化(例如遗传密码的发展。 这套理论否认了进化的发展是直线、连续的,而将进化定义为一系列不连续的阶段转变。每个转变都彻底的改变了进化可能的空间,从而为未来的进化铺平了道路。第一个阶段性转变是从“原始汤”发展到原细胞的转变(也被称为“生命起源”)。而作者认为最近的转变可以视为自然语言的引入。自然语言使得社会能够在更广泛的时空尺度上传递信息。 那么,我们是否正在进行一种新的重大转变呢?过去一个世纪中,从自然语言到机器语言的转变可能标志着巨大的、飞跃性的阶段变化。这种转变能够让我们重新构想地球的进化动力。正如地球计算揭示了它对自身的影响一样,它是否也能够审视作为一个不断适应生命过程的自身?研究人员称之为无限进化。 异生物多元体系(Xenobiology Multiplex)或异元体(Xenoplex)旨在探索这些问题。这个推测性的实验设计利用了最近的转变(全球计算)来阐明第一个进化的转变(生命起源)。通过在化学计算阵列中模拟生命的起源,该项目最终旨在理解推动地球上和地球自身的无限进化的机制。随着生物圈和技术圈的相互构成越来越紧密,一种即包含了生物系统也包含了非生物系统的进化理论至关重要。因此,该项目追溯了进化本身的演变,囊括了生化的起源直到计算的当下。







那我们该如何确切识别在异元体中诞生的生命呢? 首先,每个反应器箱必须受得精心构建,以便减小箱子本身对模拟反应的干扰。这种自我污染,或生命的泄漏,出现于上世纪 50 年代研究生命起源的著名实验,【米勒-尤里】。在该实验中,他们使用的硼硅酸盐玻璃器皿无意中帮助催化了前生物的反应。 其次,反应器箱内部的动态必须被量化。每个反应器都连接着一个质谱仪,用于采取每个箱子内的化学样品。固定的目标函数和状态平衡是不现实的,启发式是自然既不关心也不遵守的。因此,异元体需要利用各种不一样的度量标准。 其中一种度量标准可以是分子组装指数。这是由天体生物学家萨拉·沃克和化学家李·克罗宁开发的一种衡量化学复杂性的指标。该指数是根据他们的“组装理论”导出的,可以使用常规质谱仪进行实验测量,质谱仪产生化学光谱,其中峰值越多,表示成分越复杂,也意味着更多类似于生命现象的存在。 还有一种可能是测量自催化行为(测量复制数量)和新颖性(测量化学物种的多样性)。这样以便设计更加开放性的进化实验,可以灵活调整嵌套的目标。当这些箱子内的种群获得了新型的专门属性  (如自复制能力),其化学梯度和计算梯度下降的条件会传播到附近的箱子,从而实现并行优化各种不同的属性。




零结果,即在任何反应器中生命样式的测量将抵达极限,因此无法观测自复制现象的出现。 这意味着生命的发展需要与行星自然现象进行更深入的互动,正如微生物学家林恩·马古利斯(Lynn Margulis)和化学家詹姆斯·洛夫洛克(James Lovelock)在盖亚假说中所论述的那样。因此可以总结,生命依赖着多尺度上紧密耦合的反馈循环,而这些循环在实验室中不容易复现。


生命迹象只有在前生命地球模拟中观察得到,而在外生生命模拟中没有洞察结果。 这可能意味着地球上的生命是绝对独特的,但更具有说服力的解释是,该阵列仍然缺乏观察、模拟或演化非规范生命形式所需的技术复杂性。这可能是一种“感知过滤器”(萨拉·沃克创造解说的现象)。技术极限阻止了我们观察到所谓的“阴影生物圈”或异相圈的能力。这个现象好比在 17 世纪发明显微镜之前,人类无法观察到微生物的存在。


所有成功的模拟箱子,包括那些含有外行星初始条件的箱子,都演化出了与地球相似的生物生命。 这可能意味着地球上的蟹化现象在宇宙尺度上趋同进化,其进化轨迹通常独立地趋于螃蟹形态。另一种解说暗示着泛种子理论,即一个共同的生物祖先在整个宇宙中为生命“播下了种子”。


模拟了一种近生物的首次接触。在这种遵循外行星生物标志的约束条件形式中,具有生命样式的结构通过实验得到成功的进化。 这种模拟出的接触可能揭示非生物的生命形式,例如进化出具有检测、计算和利用热信息进行自我复制能力的对流细胞。这种状况可能需要利用超越化学成分的生命特性测量,以便适应于反馈循环算法。因此,这种非自然生命龙卷风是不会被质谱仪所观测到的。 这种情况也可能代表某种异世异生命,即在阵列中生成的生命形式本身是由假设的生物标志构建的仿真体,以至于箱子内的外星生命不再临摹其所模拟的外星行星上实际存在的生命形式。


最具生命特性的现象并非以物质基质形式出现,而是算法本身获得了集体的生命力。 人工生命研究的一个重要目标是发现推动无限进化的机制。通过同时选择各种普遍的生命特征,如新奇性、复杂性和自我复制能力,无限进化有可能在所有模拟箱子里通过元学习过程逐渐的显现出来。 这种机制的发现对于即将成为下一个重大进化折点的通用人工智能产生了深远的影响。目前,机器学习中目标导向的优化以及其训练无法模拟进化的探索性动态。如果达到了对神经网络应用这种过程的机械理解,这将引发模型在训练、学习和复制方面的基本典范转移。

无论结果如何,这些模拟提供了一个独特的机会。实验可以让我们解脱以地球生物为核心的生命概念。如果生命与地球脱离,甚至与生物领域本身脱离,那生命到底是什么?反之,如果生命与其所处的行星环境融为一体,以至于无法将生命与环境分隔,那生命又究竟是什么呢? 异元体不仅无法保证清晰的结论,更是进一步加深了疏离。这个超凡、令人不安的领域,正是我们必须深入探索和了解的。这样,我们才能让知识得以真正的自我拓展。

Bongard, J. & Levin, M.1
Bettencourt, L. M., Lobo, J., Helbing, D., Kühnert, C., & West, G. B. 2
Growth, innovation, scaling, and the pace of life in cities. Proceedings of the national academy of sciences, 104(17), 7301-7306.
Wong, M. L., & Bartlett, S.3
Asymptotic burnout and homeostatic awakening: a possible solution to the Fermi paradox?. Journal of the Royal Society Interface, 19(190), 20220029.
Smith, J.M. & Szathmáry, E. 4
The Major Transitions in Evolution. Oxford University Press.
Gould, S. J., & Eldredge, N. 5
Szathmáry, E.6
Toward major evolutionary transitions theory 2.0. Proceedings of the National Academy of Sciences, 112(33), 10104-10111.
Adams, A., Zenil, H., Davies, P. C., & Walker, S. I.7
Stanley, K. O., & Lehman, J. 8
Hernández-Orozco, S., Hernández-Quiroz, F., & Zenil, H. 9
Pigliucci, M. 10
Is evolvability evolvable?. Nature Reviews Genetics, 9(1), 75-82.
Watson, R. A., & Szathmáry, E.11
How can evolution learn?. Trends in ecology & evolution, 31(2), 147-157.
Gould, S. J. 12
Monod, J. 13
Carroll, S. B. 14
Wong, M. L., Bartlett, S., Chen, S., & Tierney, L. 15
Kauffman, S. A.16
Investigations. Oxford University Press.
Criado-Reyes, J., Bizzarri, B.M., García-Ruiz, J.M., Moro-Martín, A., Kamitsukasa, F., Bokor, N., & Ugliengo, P.17
Marshall, S.M., Mathis, C., Carrick, E., al.18
Harvey, I.19
The circular logic of Gaia: fragility and fallacies, regulation and proofs. Artificial Life Conference Proceedings (pp. 90-97). MIT Press.
Chopra, A., & Lineweaver, C. H. 20
Davies, P. C. W., Benner, S. A., Cleland, C. E., Lineweaver, C. H., McKay, C. P., & Wolfe-Simon, F. 21
Signatures of a Shadow Biosphere. Astrobiology, 9(2), 241–249.
Borradaile, L. A. 22
Crustacea. Part II. Porcellanopagurus: An instance of carcinization. Natural History Report, Zoology, 3, 111–126.
Bartlett, S., Gao, A. K., & Yung, Y. L. 23
Janković, S., Katić, A., & Ćirković, M. M.24
Gaia as Solaris: an alternative default evolutionary trajectory. Origins of Life and Evolution of Biospheres, 52(1-3), 129-147.
Cleland, C. E. 25