* Briefly introducing the Sciencepolice-14 rules;
* Looking at some area of science where one of these guidelines was violated, and showing the damage done.
I eventually intend this talk to be part of a series of regular talks and discussions, but as it’s so central I’ve decided to make it a blog posting too.
All living things must choose suitable actions for all situations, in order to survive and hopefully thrive. (Here is the first of a couple of nice illustrations borrowed from Eichenbaum et al. 1999.)
For this, living things need to maintain a knowledge based control system to perceive and identify what’s going on, and then choose the right thing to do. Inside a cell, this is mainly in the nucleus.
Even single-celled creatures have this. Bacteria have a system of genes, which cause actions when they are turned on, and can turn each other on and off, as well as being turned on and off by outside infuences. Each gene is like a rule in a knowledge base.
This database (together with other systems inside the cell including enzymes), amounts to, and works as a model, of both the outside world and certain features of at least the creature itself. Here is a famous example of a model – the Tide Machine.
That mini representation of the world hovering above the rat’s head actually exists in the rat’s head. Here are some images of places being represented in a rat’s hippocampus. (The hippocampus is the brain’s Map Place; it deals with memory and maps, and also connects closely to the fear centre. The common feature behind fear, memory and maps is first-time learning, after just one incident – essential for noting every turning, and also after serious fear situations):
Chess playing programs do little else but this: for each possible action, predict the future it would lead to, decide how much you like it, and repeat for the full list of considered actions until the favourite is found.
All this constitutes the “Generate and Test” process using models… and for “models”, read “theories“, since they are used in the same way.We have now extracted from the basic essence of evolution and life, the implication of models, and in higher animals, theories:
Often in human societies, these models or theories merely had to generate behaviour required to survive and thrive. It didn’t matter if the theory involved gigantic mythical snakes whose movements had thrown up mountains, or giant dragons that regularly swallowed the moon, so long as they allowed the production of suitable actions at suitable times or situations. If all you needed to know was when to burn the vegetation back, or warn the emperor of an impending eclipse, or plant crops under suitable conditions, those theories, which your brain was designed to invent, bizarre though those theories may be, had done their job.
But judging a theory by your survival is a bit stark, and in science we value the ability of a theory to make a wide range of good predictions, not just the one or two important ones essential for survival. For a scientific theory, we want all its predictions to square with our observations. We value theories that contain detail that allows us to make all sorts of new predictions at the drop of a hat. (Perhaps this arose from a world where invention was becoming ever more vital.) Scientific theories/models are for predicting futures, and in science it is on this that we judge them.
Originally, a choice of action was good if it allowed you to survive. And later, in animals with emotions, as a useful intermediate, to be happy,
But a good theory has to make a wide range of good predictions (or at least good explanations).
At first, simple survival was enough. But good theories must explain lots.
Because our minds seek models, when we see something that surprises us, i.e. something not predicted by our current model, we ask “Why?”, which is a request for an enhanced model that explains the observation, or for an explanation of the steps through which someone else’s model would have predicted the observation.
When we create expert systems, we ask human experts how they do their skill. They give us a set of rules which comprise the database defining their skill (maybe 4,000 rules in some skills, it’s been said). A “Why?” is a request for a model. The giving of an explanation is the delivery of a model into a mind/database.
It’s perfectly possible, but cheating a bit, to explain all the observations known so far about a topic, using a hugely complicated model. A “perfect” explanation of the weather might contain entries for each day, saying, “for February 6th 2005: rain from 2pm to 3pm and sunny the rest of the day” – and some similar kind of comment for each day we’ve already had. Hindsight is easy; prediction is useful. Simplicity is preferable, though note – the scientific theory of the creation of the universe is hugely more complicated than genesis, but we still prefer it. If we’re lucky enough to have many theories all of which predict perfectly, then yes, we select the simplest. But if all our possible theories are imperfect, the simplest usually isn’t the best, even though simplicity can help us choose and trim theories.
We prefer theories that make successful predictions, particularly if they’re unexpected. That means the new theory is better, at least for those surprising predictions, than the old theory. Every expectation implies a theory. You can make a model of the mind that consists largely of predictions. (I did!)
Each stage in the inevitable progression of models (genes… minds… science) doesn’t just inherit the ability to model, it inherits the need… as well as enhancing the ability.
Applied cognitive science, especially robotics, requires us to make models of the world. The imperative for an animal to survive has the same effect and follows the same rules needed for a cognitive robot to work efficiently, and also for a scientific theory to be of high quality.
If these rules do not amount to the process of science, then we should drop the process of science and adopt these rules instead.
But we don’t need to, since good science is always whatever is the current best practice in knowledge engineering.
I’ve looked at the scientific behaviour of many scientists over the last fifteen years, and compiled a summary of advice suitable for the vast majority of common scientific theoretical problem areas: the “Sciencepolice-14” rules. They’re made up entirely of a mosaic of corrections to the deviations from good science that I’ve observed.
You could tile a floor with the common errors made by scientists.
Put on a page, the mosaic draws handy guidelines for good science:
the “Sciencepolice-14” rules.
By a happy quirk of fate, the symbol has remained largely unused, at least by the West, in science. One reason for this is that it is the symbol from the Cyrillic alphabet used to indicate ladies lavatories in Russia. Naturally, Russian intellectuals weren’t tempted to use it for anything else, though in fact it was originally the biological symbol for female in Russia, since that letter starts the Russian word for woman (zhena, I think).
In Sciencepolice talks I intend to offer an example of the violation of one such rule, and illustrate the damage it causes to science, though in this presentation I also have to introduce the whole general idea as well.
We’ve now seen why science is as it is; and we’ve noted some handy guidelines for avoiding pitfalls.
One pitfall on its own causes more problems in palaeontology than almost any other, and workers in many sciences fall into it:
In science, current beliefs can rest on unsound science…
(The following quote taken from the preface to: “Reconstructing The Past: Parsimony, Evolution & Inference” by Elliott Sober. MIT 1988. Second printing 1991 ISBN 0-262-19273-X (hb) 0-262-69144-2 (pb)
…concepts which have proved useful for ordering things easily assume so great an authority over us, that we forget their terrestrial origin and accept them as unalterable facts. They then become labeled as “conceptual necessities”, “a priori situations”, etc. The road of scientific progress is frequently blocked for long periods by such errors. It is therefore not just an idle game to exercise our ability to analyse familiar concepts, and to demonstrate the conditions on which their justification and usefulness depend, and the way in which these developed, little by little, from the data of experience. In this way they are deprived of their excessive authority.
…and no science is more vulnerable to this than palaeontology.
At this point, I invite members of the audience to indicate who amongst them thinks not only that the wolf is the ancestor, but that they can explain not only why it is, but why it’s obvious it is?
Doesn’t work for horses. The Mongolian wild horse is different from the tarpan, considered to be the wild ancestor, and which survived in Europe until the 19th century. The Mongolian wild horse has a different number of chromosomes from domestic horses, making it less likely to be the ancestor.
The rule doesn’t work for cattle. Cattle were domesticated twice, once in Europe and once in south west Asia. The two domesticates can interbreed, as presumably could the two ancestors, but both ancestors are now exinct.
The rule works for the two-humped camel, the bactrian, since some wild populations still survive, but not for the dromedary which exists in the wild only as a feral.
The rule can’t be used for working out the ancestor of either humans or chimps either, since they couldn’t both be each other’s ancestor, and anyway the ancestor was different again.
There is a widespread assumption that the wolf is the ancestor, backed up by eye-rolling and a total inability to remember why the wolf is obviously the ancestor… but also by a terrible ability to remember that a mob of big-mouthed yobos will howl at you if you suggest the Gray Wolf isn’t the ancestor.
Here is a dog family tree, and although it’s based on DNA, so should be pretty reliable, we should remember that domestic dogs have massively interbred with wolves, and we don’t know whether the DNA selected for this tree was some of that gifted by wolves (or indeed vice versa). A later image shows molecular evidence suggesting dogs are not the wolf’s closest relative.
From Lindblad-Toh, K. et al. (2005):
Perhaps the earliest type to split from all the rest was the tree-climbing Gray fox, of California. Probably no harm in imagining the ancestor of all the living canids (including foxes) as being like this:
This dog tree, from Tedford et al. 2009, is more recent, and shows many extinct types. Unfortunately it is based not just on bone morphology, probably less reliable than DNA, but actually just on skulls. The dissimilarity between this and the previous tree is however a good example of how dog ancestry is the complete opposite of certain.
There is a fossil canid from East China: Canis lupus variabilis. Koler-Matznick reports that it was found at Choukoudian, China, in layers dating 200,000 YBP [years before present] – 500,000 YBP. Some of its remains were associated with early human artefacts, and some from the layer predating human evidence.
Koler-Matznick suggests that although variabilis is a possible ancestor for the dog, a somewhat similar type living further south, but currently unknown from fossils, might be a better candidate.
Despite the Certainty often claimed for dog ancestry, the relationships of for example jackals to the wolf are uncertain (unless you completely disregard the skull-based Tedford et al. 2009), and the date of the wolf/dog split is completely uncertain – even after repeated molecular analyses. NB – there will still be a split date for dog and wolf, whether the grey wolf is the ancestor or not.
Remember: “The cladogram can’t show the true ancestor if it isn’t in the cladogram.”
Here are some jackals and a dhole. The golden jackal has now infiltrated Europe, largely unrecognised, and is probably slipping into Italy about now.
Zhang et al 2013 suggest: 32,000 years ago.
Vilà et al. (1997) suggested a separation date between dogs and wolves of 76,000 to 135,000 years (based on mitochondrial DNA)
Some recent papers casually talk of a 13,000 year split.
Below, is some molecular evidence suggesting dogs had an original plot concentrated far over to the left, but subsequently smeared over towards the European wolves with which dogs have repeatedly interbred. Note how this suggests that on this metric, coyotes are more closely related to wolves than dogs are… and also that coyotes and all the wolves show signs of being smeared over towards dogs, presumably also through interbreeding.
Below is the dingo, and its range. Just because it’s feral in Australia, it isn’t necessarily feral in the South East Asian peninsula. If that is part of its original wild range, then there’s your dog ancestor… and it even satisfies the simplistic rule of “just take the genetically closest wild relative” which so many use as the obvious reason for the gray wolf being the ancestor!
Here is Janice Koler-Matznick, along with one of the dogs that she now breeds (note: she is not just a dog breeder!) She has done more than anyone else to draw attention to the unjustified dogma which passes for the science of dog ancestry. She’s holding a skull of a New Guinea Singing Dog.
For those people who think recent papers even in the last two or three years on dog ancestry have been any more than a confused and pretentious mess, then I’m sorry, but you haven’t understood them. Science is largely about selecting the best theory; and the throwing away good of theories before you start the selection is prejudice, not science… which is one fence at which all recent dog papers fall.
Wolves are too wild for circuses.
Might eat the baby if you turn your back.
Tame dingos near Australian hunter-gatherer groups have poor condition c.f. wild dingos.
When hunting kangaroos, dingoes chase them away, so dingoes left at home.
Feral dogs revert not to apparent wolves but to the appearance of the dingo.
Domestic dogs have the hind dew claw that all wild dogs and dingos lack.