Recently my brother Josh sent me a link to a fascinating article in Quanta Magazine about neuroevolution, a subdiscipline within the field of artificial intelligence. Like other approaches to AI, neuroevolution is all about creating mathematical algorithms, but whereas traditional approaches attempt to create algorithms that solve problems efficiently, neuroevolution seeks to create algorithms that maximize novelty and diversity and then tries to figure out what they might be good for.
This is exactly how evolution works in the natural world. Bioevolutionary processes don’t happen for the purpose of solving the survival problems that species encounter in their environments. They happen because they happen, and every once in a while they just so happen to solve a problem along the way.
A key concept in neuroevolution is the steppingstone principle. In natural evolution, morphological features arrived at through random genetic mutations may not only be useful in themselves but may also serve as steppingstones toward solutions to other problems. For example, biologists believe that feathers were first put to use as insulation before they were used for flight (although they did not evolve foreither purpose).
Neuroevolution uses the steppingstone principle in a similar way to solve problems by not trying to solve them. An example given in the aforementioned Quanta article concerns a maze that wheeled robots were tasked to find their way out of. Both traditional and neuroevolutionary approaches were used to evolve algorithms for this purpose. But whereas the traditional approach entailed trying a bunch of sensible strategies and then retaining and “breeding” the most effective ones for multiple generations, the neuroevolutionary approach simply went for maximum diversity of escape strategies, selecting for novelty rather than effectiveness.
Each approach was tried 40 times. Traditional AI succeeded in evolving a robot that escaped the maze three times. Neuroevolution succeeded 39 times. The reason? The traditional approach was too focused on early success, going all in for promising escape strategies that often led to dead ends. By casting a much wider net, neuroevolution traded early partial success for ultimate total success.
In reading about neuroevolution, I couldn’t help but wonder if the steppingstone principle might not also apply to running, and if so, how. My hunch is that it does. Artificial intelligence is really artificial learning. Biological evolution can be thought of as species learning—learning to adapt to the environment. And training for distance running can also be thought of as a form of learning—learning how to run better. It’s from this perspective that applying the steppingstone principle to running begins to make sense.
To suggest that the steppingstone principle does apply to running is to suggest that not trying to get better at running is an effective way to get better at running. Clearly, this can only be true to a certain extent. Running is without question the most effective way to get better at running. More than that, specific run training methods, such as the 80/20 rule, are known to work optimally to maximize running performance. These best practices are the products of a multigenerational, global process of trial and error that looks a lot like traditional AI, where different techniques have been tested and then either discarded if they proved ineffective or retained if they proved effective.
You need only compare the performance level of today’s top runners to the performance level of the top runners from 80 or 90 years ago to know that this approach to solving the problem of maximizing running performance has worked exceptionally well. But it is plausible that it has also resulted in a dead-end effect similar to the one I described in relation to the wheeled robots in the maze. A runner who relies entirely on proven best practices to seek improvement does not expose his or her body to a lot of novel challenges, and as neuroevolution has shown, novelty and diversity are rich sources of new learning.
How might a runner incorporate novelty in a sensible way into his or her efforts to become a better runner? Perhaps the least risky way to do so is to run in a variety of environments. Have you ever done a long run on a technical trail after an extended period of training only on the roads and/or on nontechnical trails and then woken up the next morning feeling sore in muscles you never knew you had? That’s novelty at work. When you run on different types of terrain and in different conditions, your neuromuscular system is forced to explore new ways of getting the job of running done, and the resulting discoveries might make you a better runner in any environment.
Non-running activities can take this effect even further. We know that activities such as strength training and dynamic stretching can improve running performance by enhancing some of the underlying physical qualities, such as muscular endurance, that contribute to running performance. But I suspect that such activities and others may also improve running performance by exposing the body to less familiar movement patterns that, in effect, add new tools to the toolbox the body draws from to push back performance limits in running.
Supposing my suspicion is correct, this way of incorporating the steppingstone principle into your running could be exploited by continuously mixing up the strength and mobility exercises you do and perhaps also by dabbling in stuff like snowboarding, surfing, and basketball. It’s not as crazy as it might sound. There’s quite a bit of research showing that early specialization in a single sport is bad for long-term development. Youth athletes who lock in on one sport before high school are more likely to get injured and burn out. I think there’s a little bit of the steppingstone principle at work in this phenomenon as well, and while adult runners who want to realize their full performance potential most certainly should specialize in the sport, there’s good reason not to go too far in the direction of specialization at any age.
Again, all of this is highly speculative. But I’m confident it can do no harm to your running and may do it some good to continuously run in a variety of environments, to constantly vary the strength and mobility exercises you do, and to dabble in activities like climbing or line dancing or horseback riding or yard work or kayaking or whatever floats your boat, because becoming the best runner you can be is not that different from escaping a maze designed for wheeled robots.