It is a common mistake to think that modern methods of production are superior to what has come before, or that we are, in our time, at the pinnacle of knowledge. A lot of knowledge has been lost or forgotten over the years, and rediscovering old knowledge can be as powerful as producing entirely new knowledge.
An interesting example is Roman concrete. Researchers have found that it has excellent durability – much greater than current concrete – but have struggled understanding how that can be. In a recent report it turns out that the Romans employed a method that created an anti-fragile and self-healing kind of concrete through what the researchers call “hot mixing”.
The result of this method is a concrete that cracks in a specified way. The cracks that develop are then such that if they fill with water they heal. It is worth pausing to think about the ingenuity of this design: to predict that the material will crack, design the way the material fails and ensure that it fails in ways that can heal — that is a way of thinking that we can learn a lot from.
The conscious design of failure into strength is not just a great method for making concrete – it is a mental model that we can use to explore different other things: everything from large language models to cars. And this mental model is a re-discovery, not a new discovery.
The example of Roman concrete, then, suggests an interesting possibility: we should not just speak of innovation, but of re-invention of old models of construction, building, doing science and thinking. The careful study of history and earlier accomplishments can be as fruitful as plunging into the new.
The history of science is ripe with ideas that have fallen by the wayside just to be rediscovered and become essential later. Reading deeply into the history of our subjects of interest, and not reading history as an account of failure leading to today’s insights, but rather as a source of ideas worth re-thinking, is an undervalued research method.
The field of artificial intelligence has its own share of these examples. Neural networks were long thought dead ends, but proved to be enormously powerful tools for building machine learning systems. But what other things have been abandoned along the way that deserve re-discovery?
Science practice might do well to assign close study of what is considered a dead end on the path of discovery. Most of science is not a labyrinth with a single path to a center, but more of a maze, with many different paths to different centers. Doubling back in a maze and examining the forking paths more deeply can yield insights that are as deep as any that are found on the way ahead. Science-as-a-maze is a mental model that also allows us to escape the idea that progress is linear – which it clearly is not.
Scientific progress through remembering what was forgotten presents real opportunities, and arguably means that we should spend more time on understading the road not taken in many scientific fields. And maybe, just maybe, we will find curious insights that have been lost over time.