Placing “self-replicating” and any kind of “bots” in the same sentence immediately raises red flags, conjuring the image of reducing the surface of the world to gray goo. But that is not a concern here, for reasons that will become clear. There is a lot to unpack here, so let’s start with what xenobots are. They are biological machines, little “robots” assembled from living cells. In this case the source cells are embryonic pluripotent stem cells taken from the frog species Xenopus laevis. Researchers at the Allen Discovery Center at Tufts University have been experimenting with assembling these cells into functional biological machines, and have now added self-replication to their list of abilities.

Further, these xenobots replicate in a unique way, by what is known as kinematic self-replication. This is the first instance of this type of replication at the cell or organism level. The researchers point out that life has many ways of replicating itself: “fission, budding, fragmentation, spore formation, vegetative propagation, parthenogenesis, sexual reproduction, hermaphroditism, and viral propagation.” However, all these forms of self-replication have one thing in common – they happen through growth within or on the organism itself. By contrast, kinematic self-replication occurs entirely outside the organism itself, through the assemblage of external source material.

This process has been known at the molecular level, where molecules (like proteins) can guide the assemblage of identical molecules using external resources. However, this process is entirely unknown at the cellular level or above.

In the case of xenobots, the researchers placed them in an environment with lots of individual stem cells. The xenobots spontaneously gathered these stem cells into copies of themselves. However, these copies were not able to replicate themselves, so the process ended after one or a very limited number of generations. In a new study, the researchers set out to design an optimal xenobot that could sustain many generations of self-replication. They did not do this the old-fashioned way, through extensive trial and error. Rather, they used an AI simulation, which calculated for literally months, testing billions of possible configurations. It came up with a simple shape – a sphere with a mouth, looking incredibly like a Pac-Man. These xenobots are comprised of about 3,000 cells. The researchers assembled their xenobot Pac-Men and when placed in an environment with available stem cells they spontaneously herded them into spheres and then into copies of themselves. These copies were also able to make more copies of themselves, and so-on for many generations.

There are no concerns that these living Pac-Men will go out of control and wreak havoc. They are fragile and would not survive outside the lab. They also need a supply of specific stem cells in order to assemble further copies of themselves. They would be useless and die in “the wild”.

There are some interesting features of these xenobots worth pointing out, in addition to their ability to replicate through kinematic self-replication. First, the xenobots did not have to be genetically engineered or programed in any way to be able to self-replicate. This was a spontaneous ability once they were assembled into the right shape. Also, these stem cells were destined (if left within their embryos) to become skin cells. This highlights the fact that genes are not the only things that determine how cells develop. The local environment is also critical. Taken from the environment of the embryo, with the presence of other cells and chemical gradients, the xenobot cells do not develop into skin cells but rather behave very differently.

This research also highlights the incredible power of AI and computer simulations. Obviously it would have been practically impossible to design and test billions of configurations of cells with actual cells. However, a computer simulation was able to hit upon a working design in several months. It’s hard to estimate how long the research would have taken using actual cells, but I think it’s safe to say that the AI was able to perform decades even centuries of research in a few months. This is fulfilling one of the core promises of AI-aided research that has been touted for years. Sometimes incredible-sounding claims for future technology are not hype.

What are these xenobots useful for? While an interesting question, I would point out that answering that question is not essential to concluding that the research is useful. Basic science research is about increasing our knowledge of how the world works. Historically such increased knowledge tends to have incredible downstream benefits, often times not anticipated by the original research.

That being said, the researchers did explore this question, simulating further if xenobots could be made to do useful work while they are replicating. They looked at the task of microcircuit assembly, and showed that the random motions of the xenobots could do useful work, such as moving wires and closing circuits. Further they showed that a self-replicating system could have “exponential utility”. In other words, a small initial investment in creating just a few xenobots could have exponential effect due to self-replication, as long as sufficient feedstock was available. They further point out that feedstock can be made exponentially available also, as each Xenopus laevis produces thousands of eggs per day, some of which can be used for feedstock and some to produce further egg-laying frogs.

Also, the xenobots represent a unique biological system, which may have many applications in research itself. They are an ideal tool for studying self-replicating systems themselves. Such a system may also be made easy to customize quickly to produce specific effects. Time will tell, but again if history is any guide, new biological laboratory systems tend to be extremely useful for future research and technology. But as I said, there is a lot to learn from the tiny xenobots, even if they do not prove directly useful themselves.