By simulating a mass extinction on a population of virtual robots, researchers have shown that these cataclysmic events are an important contributor to an organism’s ability to evolve, a finding that has implications to evolutionary biology, the business sector—and even artificial intelligence.


It’s no surprise that mass extinctions exert a tremendous influence on evolution. If it hadn’t been for the Cretaceous-Paleogene extinction event, for example, mammals would likely have never supplanted dinosaurs in many ecological niches. What’s less established, however, is whether or not mass extinctions produce consistent evolutionary outcomes with measurable effects. By using computer models, a research team from the University of Texas at Austin has presented compelling evidence in support of the hypothesis that repeated extinction events do in fact contribute to an increase in evolvability. The details of their research can now be found at PLOS ONE.

Evolvability, for the purposes of this study, was defined as “the capacity of an organism’s lineage to generate novel phenotypic traits.” In other words, it’s a measure of a species’s ability to produce an abundance of mutations over time. Darwinian principles suggest that evolvability is a good thing; it prevents organisms from getting stuck in an evolutionary rut. By regularly producing a variation of physical characteristics, organisms can adapt to environmental pressures and ever-changing conditions.

Playing Digital God

To determine if mass extinctions have an influence on evolvability, computer scientists Joel Lehman and Risto Mikkulainen connected neural networks to a simulation of robotic legs. Each robot was programmed with the goal of evolving the capacity to walk smoothly and stably. Robots were tasked with walking along a uniform 40 x 40 grid. The simulation began with the robot standing upright at the center of the grid, and it proceeded until the robot fell or 15 seconds had elapsed.

“This is a good example of how evolution produces great things in indirect, meandering ways.”

- Joel Lehman, computer scientist at the University of Texas at Austin

Like biological evolution, the robots were mutated over time, allowing them to exert a wide range of novel features. Some were detrimental, some were advantageous. After hundreds of generations, a diverse range of phenotypes emerged, allowing the simulated bots the opportunity to adapt and settle into various ecological niches.


So far so good. But in the effort to test their hypothesis, the researchers unleashed a mass extinction event by randomly killing off the robots in 90% of the niches. In the wake of these digital catastrophes, evolution was allowed to continue. After several generations, the lineages that survived were the most evolvable, exhibiting the greatest potential to produce new adaptations. What’s more, the surviving bots came up with better solutions to the task of walking compared to those that hadn’t experienced a mass extinction.

“Overall, results in the abstract model show that extinction events results in higher evolvability in the final population,” write the authors in the study.

Evolvability in the abstract model (credit Lehman et al. 2015)



“Focused destruction can lead to surprising outcomes,” noted Miikkulainen in a statement. “Sometimes you have to develop something that seems objectively worse in order to develop the tools you need to get better.”

To which co-author Lehman added: “This is a good example of how evolution produces great things in indirect, meandering ways.”

Surviving the Bottleneck

As the researchers note in their study, by “creating a survival bottleneck dependent upon unpredictable [physical] traits,” extinction events may indirectly favor lineages capable of diversifying quickly across the space of such characteristics.

The researchers conclude their study thusly:

[The] conclusions of this study may provide insight into similar phenomena in other domains, such as creative destruction in business and the way wildfires help renew ecosystems. In such cases, the temporary effect of a mechanism is superficially destructive to particular participants in a process, while the ultimate longer-term effect is to make the process as a whole more innovative and robust. This deceptive pattern seems to be often exploited by evolutionary systems in general, and thus it may be useful to consider other seemingly destructive forces in the evolution of economies, products, and ideas in the same light.

This research could also be used to produce robots more capable of overcoming adversity, including rescue bots or those sent to explore distant planets. On the scary side, the experiment also suggests that an overt attempt to eliminate dangerous AI could backfire in the sense that the most adaptable agents would be the ones left over.

A couple of quick caveats to the study. Obviously, simulation is not reality, and the experiment most assuredly suffers from a “reality gap.” Secondly, the researchers were actually harsher on their robots than nature is on biological species; as they themselves say the study: “[It] is unlikely that natural systems experience extinctions as extreme as in the ER model, i.e. imposed every 300 generations with only five distinct survivors.” Looking to the future, the researchers hope to refine their experiment, including the modeling of more complex geographies, and the ability to study evolution across multiple spatial scales, namely global-scale extinctions versus local-scale extinctions.



Read the entire study at PLOS ONE: “Extinction Events Can Accelerate Evolution”.

Email the author at and follow him at @dvorsky. Top image by Joel Lehman.