This question came up when I was thinking about Māori and other indigenous “ways of knowing”, and I realized that this kind of knowledge is acquired mostly through trial and error. Now that doesn’t mean it isn’t science, for it is “science construed broadly” as I discuss in Faith versus Fact. And before we had the scientific toolkit now used by professional scientists, this is pretty much all we had. In his new book, Rationality, Steve Pinker begins with a long and absorbing disquisition on how the San people of Africa have developed a sophisticated way of tracking animals, and accomplishing other feats, through trial and error over the centuries. And sometimes it’s based on hypotheses, like “Wildebeests are bigger than other grazing animals, and would leave deeper tracks.”
So yes, “trial and error” is science, but it seems to me that modern science involves far more than trial and error, but rests more frequently on testing a priori hypotheses. We didn’t use trial and error to get the Webb Space Telescope to its position, for if we made one error, the mirror wouldn’t open. We also used the known laws of physics, which aren’t used in “indigenous knowledge.” Darwin didn’t develop the theory of evolution by natural selection using trial and error, but had an epiphany from reading Malthus and then worked out the consequences, including checking the results of breeders. Avery et al. didn’t discover that DNA was the hereditary material by trial and error, but by knowing that the constituents of bacteria were mostly proteins and nucleic acids, and seeing which of them, via lab experiments, could “transform” other bacteria. This kind of science, more than does indigenous knowledge, builds on itself, so that as hypotheses are confirmed, one after the other, they ultimately lead to an immensely sophisticated body of knowledge.
I came to these thoughts when I was thinking about the ability to navigate between islands used by ancient Polynesians. This is often touted as a form of indigenous “science”. Now this couldn’t have developed, I think, as an a priori set of hypotheses used to get from place A to place B. Rather, we must remember all the voyagers who didn’t make it compared to those who did. Those who did may have used methods that enabled them to succeed: keeping stars in one position, looking for seabirds, and so on. Over time, the successful voyagers incorporated one method after another so voyages became ever more successful. The methods” of those who died attempting the voyages were lost. This was trial and error, and it was a form of scientific knowledge, but it involved enormous wastage. It was the result of one successful trial piled atop another, with the methods that produced successful voyages incorporated into lore.
A similar method could be used to suss out a way through a bifurcating maze. Imagine that there is a Y maze, with each branch bifurcating. After three bifurcations you have 8 separate destinations. Further assume that all the destinations save one—for example, taking the left branch, then the right, and then the left—conclude in a pitfall trap with spikes at the bottome, while that one successful trio of choices leads to a cache of food. Your job is to get the food. You send 1000 people into the maze, and only about 125 (one out of eight) will emerge at the other end. You then ask them how they got there. They’ll all say, “Go left, right, and then right.” Everyone else would be dead. And so you have “knowledge” of how to solve this problem.
It must have been something like this that was the nucleus of Polynesian navigation—and much other indigenous knowledge. It seems to me—and I may be wrong—that modern science is far more than trial and error, and far more than the kinds of “knowledge” that advocates of, say mātauranga Māori tout as coequal to modern science. But teaching how modern science approaches problems is far more sophisticated than trial and error, even though indigenous knowledge is indeed sometimes “knowledge” or “truth.” Teaching mātauranga Māori in science classes is not possible so long as that method incorporates philosophy, mythology, legend, and morality, but its empirical methods—trial and error—could be taught. That would take one hour at best. But could you use trial and error as a way of understanding the cosmos that is coequal to the entire toolkit of modern science, which includes trial and error, but so much more?
I am not yet convinced that there’s a good reason to teach indigenous “ways of knowing” as coequal to science, though there is a rationale for teaching the methods that our predecessors used to gain knowledge. It’s just that this is not coequal to modern science.
When knowledge was based on trial and error, progress was slow and tedious, and sometimes didn’t happen. It took centuries of failed trial and error to figure out how bubonic plague or malaria was caused, and how to deal with it. The solutions came from hypotheses that were tested, not willy-nilly guesses that weren’t tested, but merely enacted—and failed, one after another. (A lot of my ancestors were burned as supposed remedies for the plague.)
These are just some random thoughts to inspire others to think about the issue of indigenous knowledge versus modern science
41 thoughts on “Is learning through trial and error “science”?”
I’m currently reading The Beginning of Infinity by David Deutsch and he devotes the first chapter “The Reach of Explanations” to exactly this issue of what constitutes science. He gives examples of steps in the early progress of gaining scientific knowledge and how the method has been refined to what it is now and how it now produces such rapid advances in knowledge of the workings of the universe. He shows how methods such as trial and error are not sufficient. It’s 33 pages of well argued and closely reasoned text that I can’t do justice to here. Well worth reading for the first chapter alone.
Thanks for the recommendation; I’ll have a read.
And you already have a copy of it!
May I draw your attention to a talk he gave a long time ago. I second the recommendation. See it here (~16mins):
Even if in ancient times people learned how to do big things like building pyramids, sewage systems, was it science if it was not written down. Some of these things done by people thousands of years ago were then lost and had to be discovered all over again years later. Sometimes it took modern science to figure out how they did it many years before. It seems that learning and science can get mixed together and confused. If you are teaching a class in how they did things years ago I would call it history.
The method will vary with the subject (experimental methods can hardly ever be used in geology, for example), but certainly, learning from trial and error, whether in a planned, sometimes hypothesis-driven fashion or inadvertently or haphazardly is an integral part of science — only NOT learning from trial and error wouldn’t be. Someone here recounted that cats were removed from an island because it was believed that they were a threat to endemic birds, then it turned out the cats had been protecting the birds from rats whose uncontrolled proliferation turned out far worse than the birds had ever been. An attempt to help the birds had been made, it trned out to be an error, the lesson was learned.
I was thinking something similar. In a lot of cases, when developing or testing reagents for example, there is a need to get a sense of an effective dose. So a series of experiments that are essentially trial and error experiments need to be run, test over a five log dose range, and then fill in the relevant details, fiddle with the correct timing to see a phenomenon etc. These are pretty easy studies to do in cell culture or in some animal models, but painfully slow and expensive as clinical trials. This is pretty evident in the difference in dosage and the timing of the first and second shots of the Pfizer and Moderna covid vaccines (which are otherwise very similar). The Moderna dose is three times higher (and a half dose, which is still more is used in the booster). The shots were initially scheduled 3 and 4 weeks apart respectively. These were just based in essence on best guesses at the time and were then locked in by large trials that worked well enough for approval. So even here there is a trial and error component and the perfect is not really the enemy of the good, otherwise we’d still be testing best doses and intervals with no vaccine in sight
Perhaps the further we stray from “pure” science (or maths) the more trial and error (or “test and verify” per ThyroidPlanet below) comes in.
You may be thinking of Marion Island, in the sub-Antarctic, where a number of countries (including South Africa) have established weather stations. In 1949, 5 domestic cats were introduced in an effort to control house mice, and this number grew to more than 3400 feral cats in the late 1970’s, which killed an estimated 450 000 birds (mostly Burrowing Petrels) per year. Various measures were then introduced to try and eradicate the feral cat population, amongst others hunting, poison as well as biological control (not recommended reading for cat lovers!)…
Just to remark that “trial and error” methods are used in forefront engineering. Darwinian evolution is a classic “trial and error” process that optimises the “fit” of an animal to its ecological niche. Similar “genetic algorithm” processes are used to optimise engineering solutions. So, for example, the aerodynamics of a car might have resulted from a “trial and error” process both at the computer-aided-design stage and at the wind-tunnel-testing stage.
A small language quibble :
I think the more accurate and less … shop-talk ..(?)… phrase would be “test and verify”.
I have to be careful when using “trial and error” – if in the right company, sure. But if its getting serious, “test and verify”
I know every reader and PCC(E) understands, but I thought it’d be worth noting this small thing.
I don’t think Polynesian navigation or San tracking can be fully characterized as “trial and error”. Although it may have begun as such, they clearly developed principles that could be applied to new situations. The San, for example, knowing that they themselves and animals they observed carefully must drink water, could very well have established the principle that “animals need water”. Animals differ greatly in how often and how much water they need, but the San could use this generalization to anticipate the movements of animals which they had not carefully observed. Recognizing the similarities between themselves and, at least, other mammals, could lead to a form of “insight” learning– “Mammal X has similar needs/senses as I do, which allows me to anticipate its behavior.”
Similarly for the Polynesians, knowing when storms occur, which way the winds blow, the directions birds fly from, how long they stay on the island, how these all change over the seasons, inferring that land plants must come from the land– all of these “estimates” of the way the world is could lead to reasonable surmises about navigation. And the Polynesians (or at least the Micronesians) created “charts” out of sticks that aided navigation– it was not written, but it was not just oral tradition.
And this could lead to a long discussion, but I’ll just remark that most of modern science is “estimation”, not “hypothesis testing” (to use the statistical terminology).
Imre Lakatos’s ‘methodologies of scientific research programmes’ seems to be a better fit than the stereotypic Popperian falsification of hypotheses. Here, a ‘research programme’ means a sequence of successive theories, each one more refined than its predecessor to fit in with evidence. This isn’t exactly trial & error so much as what you allude to, estimation and re-estimations.
I have the utmost respect for indigenous knowledge. I would not venture into the African bushveld nor venture to climb Everest without a trusted local guide. .But with the greatest respect to these wonderful people, trial and error is only a small component of the scientific method as I understand it.
‘Trial and error’ is indeed not enough to be characterised as ‘science’. But what is?
Karl Popper had a go at answering this question around 60 years ago in ‘The Logic of Scientific Discovery’. He got a fair bit of pushback from Kuhn, Lakatos, and most notoriously Feyerabend, whose counter-attack was titled ‘Against Method’, and whose slogan was ‘Anything Goes’.
The issue seems to me to be not how hypotheses are generated, where indeed ‘anything goes’, but how they are then tested, provisionally accepted, and modified or supplanted in the light of subsequent experimental or other evidence. This process demands something much more rigorous, objective and reproducible than simple ‘trial and error’. The protocols of modern science are no doubt capable of improvement, but they get a lot closer to the ideal than either undisciplined trial-and-error or unsystematic folklore.
There’s a risk of taking “trial and error” too literally. The trials aren’t random and they don’t always result in an error. Those sailing Polynesians didn’t want to die. Some may have been in a desperate search for food and water so taking great risks was reasonable. Some may have found islands by accident when they were blown off course, or made a navigation error, but they didn’t set out with that in mind. While “trial and error” CAN describe making random changes to a system and recording the results, that’s not really how it works in practice. The trial is usually done with some thinking behind it (ie, an hypothesis) and with some expectation of success.
As I see it, modern science is still trial and error. It certainly involves trials that often fail. Isn’t it really just trial and error with more sophistication and a greater understanding from a meta perspective? Science obviously doesn’t proceed in some sort of lockstep fashion. Experiments are still trials that may not work. Scientists still operate based on theories, many of which future scientists will look back on as crackpottery.
Frequently, a scenario might crop up where a “quick and dirty” experiment may be precisely the thing to when lost one afternoon in lab. The counterpart to “an afternoon in the library can save a week of experiments”.
Nobody will care what Popper or Khun wrote about anything in this case because if a dark spot show up it means e.g. your sample is good and you should proceed, or if not, you gotta start over…
Couldn’t resist the legal allusion … who is it, Ken Kucek? He’d have words for this “trial and error” phrase….
An excellent point very well made Paul.
I also think the question of what exactly constitutes a form of science is a tricky semantic one. Whether indigenous “ways of knowing” should be taught alongside modern science, less so.
I’m not sure about this – you could regard science as a more sophisticated form of trial and error, where the trials are informed by hypotheses, but in many cases the hypotheses themselves are based on ‘guesses’ from informal observation. Presumably trial and error was similarly based on prior knowledge or informed guesses. So maybe not ‘science’ in the full sense, but scientific.
The way I learned it, “trial and error” was what science evolved to correct. The adage “try it for yourself and see” allows a lot of false positives which tend to become entrenched over time and get passed down. There’s no theory or explanation behind them. There’s no way to separate the valuable from the worthless. There’s nobody challenging some of the basic premises, let alone a committee of skeptics. And, as you point out, there’s no way to build upon successes. It’s been argued that modern science could not have happened without the invention of the printing press.
What Māori knowledge seems to be is good technology.
” “try it for yourself and see” allows a lot of false positives which tend to become entrenched over time and get passed down. There’s no theory or explanation behind them.”
Agreed – I would add that application of mathematics would make a difference in this regard – when theory or explanation doesn’t.
E.g. “plug n’ chug” application of a search algorithm – fancy words for “try and see” – can turn up unambiguous results, that are amenable to verification in a way that experimentation with Nature either cannot, or would be inefficient or impractical. Not that it would be sufficient for discovery – just that mathematics combined with experiment will … I don’t know,.. work better than in isolation?…,
This is hard to discuss on a tiny screen but a great exercise! Really interesting…
I think early navigation anywhere was brutal – lits of lost life.
… (system ate my edit)…
I think a key factor in this enormous topic is :
In the process of repeating successful patterns over and over again to get good results, when did the idea to ask a critical question, or to pose a problem in a way it could be solved, occur. When were two knowns put together to learn a third new piece of information.
And is that a fair assessment – repeating patterns over and over again.
I think science differs from trial and error in that it is systematic and predictive.
One of the things I used to study is wood engineering. With trial and error, you can build a mast or a bridge that will fulfill it’s purpose. There will be some rules you learn, about how long to dry the wood, what species work best, and ideal dimensions. People have done this in many cultures without needing a written language. Such techniques change very slowly. How do they do this? I can tell you in a single word…. (tradition)
A several centuries ago, some people decided to start testing and recording the physical characteristics of different types of wood. By the time I started studying it, it had become a huge series of tables and formulae.
That lets one take a proposed design, and predict if a desired wood type will hold up to the application, or whether it would need additional support, as well as what size gaps to leave to allow for swelling, etc. I guess that is science.
If you maroon me on some island, but leave me the right books, I could take whatever odd species of wood grows there, and take some measurements to enter a new wood data set into the tables. I could then start making all sorts of things, Gilligan’s Island style, confident that my designs will take the predicted loads. That has to be science.
A technology in some ways analogous to Polynesian navigation might be the architecture/stonemasonry used in building the extraordinary medieval
cathedrals. The relatively few collapses—famously Beauvais, towers of Ely and Winchester—aren’t frequent enough to characterize the technology as
“trial and error”. There was a body of knowledge, known to the master masons of the time, which underlay their design, planning, and construction, and which must be comparable to modern Physics and civil engineering, It did not have an intellectual structure like modern science, however, and presumably resembled traditional craft secrets, passed down through a largely oral tradition from masters to apprentices.
Yet modern buildings still fail once in a while. New vulnerabilities, and techniques to deal with them, continue to be invented. Modern science differs from what came before but “trial and error” is still an apt description of a lot of what passes for science and engineering today. The idea that we no longer rely on trial and error is just wrong, IMHO. Trial and error always involved hypotheses and modern science still tries out new ideas. Just to name one example, drug researchers come up with possible mechanisms for disease and do experiments to test them. If that isn’t “trial and error”, then I don’t get what the phrase means.
I’d like to point out a distinction-with-a-difference between invention and discovery. That’s all.
I think a critical question is whether something new was discovered – where is the discovery?
If there is a discovery, it is science. It is irrelevant how a discovery was made, but odds are that to know a discovery is true, it takes more than a couple low-level quick-checks. A new island found with new plants, to take the example given.
If there is no discovery, it could still be an impressive array of technology or engineering – as readers note the masonry of medieval architecture, or early seafaring construction. A new way to build such things is improved, refined tech or engineering. It isn’t a discovery because we already knew about ships or cathedrals and how to make them.
Do you think there should be a distinction made between the science of discovery (truths about the world, e.g. why the sky is blue or space curved) vs the science of manmade construction (e.g. building a dam, basketweaving, bridge engineering, navigation)?
If the knowns are used to find something previously unknown, it is a discovery.
So, lemee think this out :
jellyfish aequorein, not new. Jellyfish aequorein built into known protein, that (was) new (at the time), but the Discovery with a capital D was that a new construct of protein plus aequorein let us view biological processes in cells and find medicines for diseases.
But at this point, that is well-travelled territory. So new protein constructs with aequorein are immensely useful, and necessary, but we already know it can let us watch cells in action, find medicines, understand biology, along with a host of other chromophores.
… so I guess no, there is no distinction in principle…. besides a bridge being large and a protein construct being small. Still “human-made”… still nothing to say that we (rather, “they”) will never find a completely new way to do .. those things… bootstrapping from what is already known….
But at some point, someone discovered The Wheel, The Chain, The Bridge (like in the previous fascinating discussion here). A host of factors play a role in whether it is useful or not…. blah blah… we need to do this at a pub! Too much poking on a tiny screen!…
In late medieval times, they knew about cathedrals, and about load-bearing columns (discovered in antiquity)—but wasn’t the invention/discovery of the flying buttress something new? The master masons who came up with this idea must have understood things related to later Newtonian mechanics and to what we call today force vectors and vector quantities —and used
this knowledge to devise an architectural innovation. Would that not be science broadly (or not so broadly) construed? As a more general point, the historic connection between science (including scientific discovery) and technology is often under-appreciated.
“wasn’t the invention/discovery of the flying buttress something new?”
It is an invention.
“Would that not be science broadly (or not so broadly) construed? ”
The distinction I am making is that discovery finds what was there all along. DNA, its hereditary nature, its structure, … that has been true since the earliest life forms.
The flying buttress – as important as it is – was willed into existence by use of principles of Nature that had been already discovered..(I think)…
That’s my sense too. The discovery/invention distinction is too blurry to delineate useful categories and becomes semantics. Thanks.
Thank you – but, if I may :
DNA is not an invention. It has been there for billions of years and it was found. DNA, blood types, viruses, gravity, etc. were discovered – the cover was thrown off of it.
Flying buttresses, Beethoven’s ninth, microscopes, etc. were invented. They were not there the whole time.
It seems a clear distinction to me.
Any process that reveals underlying causes and improves our predictive and explanatory power would I categorize as science. How such a process did its work seems to me not important though there are a set of best practices how to do science.
The end product of science is a function that tells us of any sentence humans can think of whether it is true or false. If we do science right, as time progresses, the set of true sentences increases, but also the function will continuously change and will be implemented using different, more cohesive scientific sources. Science of the past doesn’t have to be science now.
Some Māori-knowledge was (probably) science in the past, but do we still use original Māori-knowledge to build this function or has it been outpaced by other science?
It would be fascinating to know if the Maiori described something akin to the longitude problem – where an accurate timekeeping technique would aid navigation.
That is not really how their system worked. They had good boat handling skills, and very capable seacraft. But their navigation was more formulaic, and designed to get them in a rough area, where they would look to birds, clouds,scents in the air, and swell patterns to find clues to the exact location of land. I am confident that even if they headed out randomly, or got lost, they would be able to locate land if it was there, anywhere near their route.
An example, from my family. I am not claiming that these stereotypes are universal, or apply to anyone else. Anyway, If my Dad wants to tell me about a spot on the ranch where he saw a Mountain Lion or something, he might tell me it is on a plateau, in a spot just over 4.5 miles ENE if Ossier peak. My sister would tell me-“From the Fox Creek main gate, go down and cross the river at the Amish camp site, then go upriver until you get to where the creek flows into the river. Head up the gulch for about a mile and a half until it ends, then climb to the plateau. The spot is then on your left, about 100 yards.
For a Polynesian on the sea, those directions would include going for a certain amount of time towards the direction where a certain star rises from the horizon, or, after a certain number of days, turn so that the swells are on the port quarter. Ideally, the route would have some fixed landmarks, like discolored water from shoals, or some island that is not the final destination. That system works best when the weather is consistent most of the time, which is the case in some parts of the Pacific. It can also be seasonal. Obviously, overcast skies or storms, even distant ones, can disrupt the process.
Navigational ability and techniques in the Pacific differed tremendously. Some groups had large territories covering multiple island groups well beyond sight of each other, and viewed the spatial relationship between the islands in a more or less conventional way.
I guess it becomes science when the navigator from a tradition where they always go from island A to B,C,D and so on, figures out how to go direct from A to G, then to B, to save time and to expect more favourable winds and current.
Indigenous ‘ways of knowing’ include both supernatural elements – the role of ancestors and spirits in guiding or influencing daily life and the ‘need’ to appease these spirits – and practical stuff such as how to navigate beyond the horizon, how to catch a fish, which plants are edible and so on. Clearly the former is well outside science. With the practical stuff, some of it will have been learned as pure trial and error/accidental discovery and then passed on from person to person. For example at some point in the distant history of Homo perhaps someone accidentally left some food in or close to a fire and discovered that the food was actually made more palatable and digestible as a result. Such discoveries don’t seem to warrant consideration as science. Other practical knowledge, however, is likely to have been based on the application of reason; knowledge gained from past experience may have lead to principles on which simple hypotheses are based that could be tried out to develop new ways to catch food, make shelter and so on. That approach can probably be characterised as science broadly construed.
Modern science takes an analogous approach in using past ideas and successes to suggest new ideas and testing them by experiment. It has proven to be so hugely successful in advancing our understanding of the world, though, because it has formalised basic principles and approaches that greatly reduce the possibility of self-deception such as controlled experimentation (ensuring confounding variables are eliminated or accounted for), replication, statistical analysis, peer review, double-blind trials and so on. One major way in which modern science differs from pre-Enlightenment science and much indigenous knowledge is that it explicitly allows and encourages the knowledge of elders and authorities to be challenged and replaced. This means that it continues to advance where ‘other ways of knowing’ tend to be much more static.
I think it should be instructive to see which indigenous ‘knowledge’ the educators in favour of these proposals would actually include in the curriculum. I have a feeling that they would eschew the more supernatural parts — while avoiding giving credit to the scientific process that we have used to discard of those ideas.
I think the distinction between science and technology/engineering matters. Science relentlessly pursues the question of *why* the world is as it is, rather than being satisfied with description, no matter how useful or reliable the description. Aside – and the *why* can’t be an unexaminable spiritual answer.
As a behavioural scientist, I see my animals learning *what works* all the time (engineering? Knowledge anyway), but I don’t think they’re speculating about *why*. And they certainly don’t test their ideas about *why* by trying alternatives to see if they *don’t* work (science).
So trial and error can discover what works, but I don’t think it even attempts to ask why it does. It’s part of science, but science is more.
(Aside – Jonathan’s last point above is vital and well-made. Because science is both public and challengeable, it’s fundamentally different from protected guild secrets.)
I agree with your general point, but I don’t know if “why” is the right word. I would call it more like “how,” even though that’s not quite accurate either. Asking “why” in science always leads to an infinite regress which may or may not satisfy you, depending on your depth of curiosity. There was a good interview with Richard Feynman on this question near the end of his life, should be easy to find if so inclined.