An inverted pendulum is a theoretical concept that vexes control system scientists. Think of it as an upside down pendulum whose center of mass is above its pivot point. It is unstable and will fall over unless it can be kept suspended upright by a control algorithm that moves the pivot point back under the center of mass—the same concept that keeps a person’s body standing.
Whether you realize it or not, there are a near-infinite number of such “keep the system standing” self-stabilizing homeostatic algorithms operating at every scale of biology—ranging from molecules, organelles, cells, organs and individuals to species and entire ecosystems—working in an interconnected fashion to keep you in good standing as a living person. This process is what keeps you healthy and helps you survive life’s many slings and arrows.
In my view, the inverted pendulum problem epitomizes the fundamental challenge of all living systems, which Hungarian biologist Erwin Bauer described as entities that are never in equilibrium and subsist only by expending their free energy. Some refer to Bauer’s description of living systems as the First Principle of Biology.
In college, I researched algorithmic solutions to the inverted pendulum problem to gain insights into biological homeostatic algorithms. I was struck by how human-designed efforts had failed to generate a robust solution to the inverted pendulum problem.
Seeing how difficult the problem was for control system theorists, I wondered if the algorithms could be self-designed by machine learning instead of by human design. While in medical school, I spent a year coding machine learning algorithms to see if artificial intelligence could self-correct its mistakes—just like the problem of keeping a person standing—in interpreting mammograms until it could outperform humans in breast cancer detection. In 1993, our lab’s work became one to the earliest papers published in the use of machine learning in modern medical applications. It was robust, but hardly perfect. It had also taken a year of trial and error. It seemed there was a much better way.
Then one day, while walking together down a familiar road, an old college friend mentioned his novel approach to solving the inverted pendulum problem. He was one of those brainiacs who had been in a molecular evolution postdoctoral fellowship with a Nobel Laureate just for fun. We had shared a deep passion for evolutionary theory at Harvard College where the field’s giants such as E.O. Wilson and Stephen Jay Gould roamed. Now my friend was proposing evolving a population of computing algorithms that competed to solve problems, such as the inverted pendulum problem. Think of it as an XPrize simulation run among machine learning algorithms.
The evolutionarily selected solution to the inverted pendulum problem which had vexed human-designed algorithms was slightly longer than a haiku.
What he had discovered stopped me in my tracks: The winning algorithm was just four lines long. That’s right. The evolutionarily selected solution to the inverted pendulum problem which had vexed human-designed algorithms was slightly longer than a haiku.
As I marveled over the algorithm’s elegant simplicity, I reflected on the larger implications of the process that generated it. We tend to think about genes or individuals as units of evolutionary selection, but should we also be thinking about natural algorithms as units of evolutionary selection? That would profoundly change the DNA of how we think about the natural sciences.
It would also change how we think about DNA itself. Replace the phrase “solving the pendulum problem” with “solving the problem of biological trait evolution.” Replace “thermodynamic processing unit” with “nucleotide.” Replace “four processing elements” with “four nucleotide symbols A, C, G and T.” Replace “six relationships among the processing elements” with “thermodynamic algorithms among the nucleotides.” In all of these cases, fundamental aspects of biology can be reinterpreted as evolutionary algorithms—greater than the sum of their parts—yielding new ways of understanding our world as well as improving it.
We now can reimagine the story of DNA as four nucleotides operating as a natural algorithm that performs the function of biological trait evolution more efficiently than any other set of molecules. While other competing natural algorithms could also perform the function of enabling biological trait evolution, the winner through Darwinian selection so far has been DNA, a simple natural algorithm that has established self-stability on the adaptive landscape of evolution.
More broadly, we can appreciate all phenomena, from the quantum to the astronomic to the social, as natural algorithms that perform particular functions more efficiently than other algorithms. We can also appreciate all biological pathways as evolutionarily selected natural algorithms that compete to perform functions that keep us healthy.
Drugs that address these symptoms often reduce the robustness of our natural algorithms—akin to how propping up children reduces their resilience down the road.
These homeostatic algorithms are nature’s greatest endowments to us, and they do their jobs well for about 40 years. Then they start to self-detune. The detuning leads to losses in bioresilience that manifest as features of aging, including those we subjectively feel and those that produce diseases. Drugs that address these symptoms often reduce the robustness of our natural algorithms—akin to how propping up children reduces their resilience down the road. On the other hand, working on solutions that increase the resilience of our bioalgorithms to promote healthy longevity is the road less travelled.
When it comes to aging, imagine the current health system as a pendulum that promotes atrophy, where to be in equilibrium is to be dead. The current system merely seeks to ameliorate the symptoms of aging, rather than preventing them from arising to begin with. Now imagine its inverted version, a healthcare system that promotes resilience where being in a continuous state of nonequilibrium—constantly changing and reprogramming our own biological algorithms in response to those changes—is to be alive.
That’s a notion worth standing up for.
