Synthetic living machines

Fields and Levin (2022) describe Xenobots as the result of removing skin cells from a frog embryo. In a new environment the cells reboot their multicellularity and form self-motile proto-organisms.³ Levin (2023) adds that the motion comes from cilia normally used to spread mucus over frog skin, and that the forms perform varied spontaneous behaviours.⁵ Abramson and Levin (2021) describe them as made of epithelial and/or muscle cells, able to navigate their surroundings and interact in swarms.¹ Levin (2023) also notes that in the animal these cells form a passive outer layer, and that the novel behaviours are normally suppressed by signals from other cells.⁷

The most discussed behaviour is kinematic self-replication. Levin (2023) reports that Xenobots build copies of themselves by rearranging loose cells supplied in the medium.⁵ The same author describes the biobots corralling cells into piles that self-compact and form the next generation, which repeats the cycle.⁸ Bongard and Levin (2023) link this to von Neumann's idea of machines that assemble copies from material in their environment. Here the material is dissociated cells.⁶ Levin (2023) says this style of replication is not, to the authors' knowledge, used by any other species. Xenobots reached it within 48 hours of creation, without a history of selection as Xenobots.⁵

The authors read these results as evidence about the cells, not about added engineering. Bongard and Levin (2023) stress that Xenobots are not genetically modified and that standard frog cells implement the new functions.⁶ Levin (2023) argues that the novel behaviours are hidden in the body by instructive signalling and appear when cells are liberated.⁷ Fields and Levin (2022) say the cells reuse the hardware of the wild-type frog genome in new ways. They add that evolution yields reconfigurable hardware able to support diverse goal states, not only fixed setpoints.³ They also hold that such agential materials need strategies different from those used for passive materials.⁴ The held passages contain no dissent from this reading.

Abramson and Levin (2021) call for an empirical framework to measure sensing, decision-making, learning and problem-solving in such beings, and say it is imperative to understand their degree and type of intelligence.² Gumuskaya and colleagues (2024) extended the work to adult human tracheal cells, called Anthrobots. Their shape and behaviour resemble those of Xenobots despite highly divergent genomes, age and tissue origin.⁹ The authors take this to show generic laws of morphogenesis acting alongside species-specific genomic information. They also report differences. Linear motion was more consistent in Anthrobots (80%) than in Xenobots (67%).⁹