- Introduction
- Things and their properties
- A generalisation
- Data and process
- Temperature
- Guns
- Properties again
- A childish digression
- Conclusions
- References
Introduction
We have talked of cabbages and kings already (see reference 1), of all the things that a brain might take an interest in, might have names or words for. Here we take a different tack, by way of preparing the ground for a discussion of the way in which LWS-N (see reference 2) codes up the contents of consciousness.
The underlying hypothesis is that the consciousness of adult humans usually depends on some understanding of that which they are conscious of, of the subject matter of consciousness. And that that understanding depends in some large part on the analysis of wholes into things, of things into named parts, properties and object relations. So we follow the path first trodden by database experts in the closing decades of the last century, a path characterised by a technique called data modelling.
Relational databases on computers have very explicit data models and they are capable of storing all sorts of information about all sorts of things. They power many of the services that those computers provide. We do not suggest in what follows that brains are organised in the same way, rather that brains have to some of the same things that databases do, that trying to build a bridge between the two might be instructive and that data models might be a good place to start.
 |
| Figure 1 |
First, the labelled box, denoting a number, possibly a large number, of things of the same sort. Things like people, cars or the paintings in a picture gallery. In a computer, each box, each kind of thing, would have its own kind of record. With a person record perhaps being something like Figure 1, with all the straightforward property values strung together in some set arrangement, in some set format. With first name and family name being separated to make it easier to sort the records into the right order. The less straightforward property values would likely need something more complicated.
 |
| Figure 2 |
Some purists like to insist that data is held just once, so that when something changes there is just one change to make. So we do not, for example, keep monthly totals in month records. An insistence, a rule, which can be tiresome in practise and which we think brains are unlikely to respect.
 |
| Figure 3 |
In our database, this relationship will often be modelled and will almost certainly be implemented with a link record, as shown second from the left. And, usually, one will end up with having data in that link record. It is there, after all, and computers, like nature, abhor a vacuum. In this case, for example, one might put a percentage there, perhaps computed from DNA analyses.
Or thinking back to our personnel database, we have records for people and posts – and link records which say when this person occupied that post. A record containing, at the very least, start and end dates. A link record which implements a many to many relationship between people and posts.
Or thinking of a parts database, the sort of thing a manufacturer might keep, any one part is made up of a number of other parts (members) and is part of a number of other parts (owners), giving us what is, in effect, a many to many relationship between parts and themselves, commonly shown as on the left.
 |
| Figure 4 |
But two words of warning. First, data models are prone to mission creep, are prone to try and do too much. We need to remember that we are not doing philosophy, we are trying to help along a forthcoming description of brain function. Second, complicated systems can be modelled in lots of different ways, and getting the way that is right or best might be a matter of nice judgement. Or it might not matter – it all depends on the context.
Things and their properties
From personnel systems, we now turn back to the world of brains.
 |
| Figure 5 |
We stick with the straightforward, with things like elephants, telephone boxes, Elton John, kings, queens and cabbages. Some of these things will be particular things, some of them will be more common, this being more or less the distinction between proper nouns and common nouns. The world, our world in consequence, is conveniently organised into things. But what do we know about these things?
We know that our things have properties. Taking our examples in order, properties like number of legs, colour, age, date of accession, date of first confinement and variety of cabbage. We will probably know the values of some of the properties of some of the things and we might – but do not – speculate about whether we can know of a thing without knowing the value of any of its properties.
In this context, an important property of a thing is its name, most often a single word, sometimes a small number of words. While our brains often have things without names, with a pre-literate child knowing about giraffes without knowing what they are called, being able to name things is a big help when it comes to comparing one experience with another or wanting to share those experiences with others. Names strike a balance between keeping the number of names reasonable, keeping them reasonably different, one from another, and keeping the number of things to a name reasonable, three simple desiderata which mask a huge amount of variation. Generally speaking, names are not unique. Some of the issues here were touched on at reference 1.
And then there are relationships between things which we have broken down into parts, children and associations. This leg is part of that person. This dashboard is part of that car. Then this person is one of the son’s of that person. Equivalently, this card is one of that person’s credit cards. And weakest of all, there is some connection, some association between this person in Shropshire and that person in Worcestershire. Notice how autonomy has increased here. A part is not autonomous at all; a part is in a close and fixed relationship with whatever it is that it is part of. A child is autonomous, a child need not be in close proximity with its father or even in touch, but the relationship is not one that can be broken. While an association can be broken; one might lose touch with the person who is far away in Worcestershire. Then there is kind: the parts tend to be one off, the children tend to be grouped into a number of kinds – so a household record might have both people and pets as child records - while the associations are more vague.
The top relationship, about kinds, is different again. This thing is a house. Equivalently, this thing is a member of the set of houses. We have used a double headed arrow here because the relationship between things and kinds of things is many to many. A houseboat, for example, is both a boat and a house. Relationships of this kind can be chained together in a more or less arbitrarily complicated fashion.
We often have something called property inheritance with kinds: if A is a kind of B and C is a property of B, then, in the absence of direction to the contrary, C is a property of A. Similarly, if C always takes the value D in B, then in the absence of direction to the contrary, C will take the value D in A. Or, if C mostly takes the value D in B, then in the absence of direction to the contrary, we will guess that C takes the value D in A. Our opening position, our opening gambit.
We gloss over the distinction between the subset relation and the membership relation, which can sometimes be important. Saying that this set of cows is a subset of that set of cows – ‘these cows are Holsteins’ – is a different kind of saying than saying that this cow is a member of that set of cows – ‘Daisy, who has strayed into my cabbages, is one of Farmer Giles’s cows’. A distinction which we might mark with a collection property: this thing is a collection, that thing is an individual. We resist the temptation to talk about things which are neither, which are quantities of something.
We include images, because we may well have (usually) visual images of our things, but we do not explore how those images might be annotated to make them more useful - beyond noting that images may carry history as well as facts. It might be fair to say that autobiographical (or episodic) memory is more dependant on images than semantic memory. It is also the case that we expect such annotation to be supported by the analysis of our things into the structured layer objects of LWS-N.
We might also have images of other kinds, perhaps carrying emotions and feelings about things. Emotions and feelings which might be useful, in the short term, in helping to decide what to do about things.
And we can start to get a bit involved when we consider that some properties and some property values are things in their own right. The brain seems to take this sort of thing in its stride, but designers of computer systems have to take care if they want a system which is going to run at reasonable speed.
A generalisation
 |
| Figure 6 |
And with the range of possibilities for phrase values being suggested on the right. Parsing the example is left as an exercise for the reader.
Note that we have also gained the possibility of labelling the thing relations suggested in the world of things of Figure 5. We can say in the label what sort of thing the pointer ought to point to.
A formulation which removes much of the more specific content of the first diagram, in favour of greater homogeneity, greater simplicity. And it is also suggestive of how all this might be expressed in facts on the ground, in the brain.
A formulation which, as it happens, is very like that of the XML which underpins much of the traffic over the Internet.
A version of the story already told at reference 3.
Data and process
It is sometimes said that brains do not separate data and process in the clean way of computers and that this makes comparisons difficult and less illuminating. One might make the same point of computers driven by neural networks and computers driven by more traditional algorithms.
It is, nevertheless, clear that both brains and neural networks do hold data, do store data. They can both, for example, remember about things. So how is a reasonable question.
It is also clear that the division between data and process is not as clear cut in traditional computer systems as at first sight might appear, although the examples that we know about reflect process creeping into data more than data creeping into process. Where by data we mean data about the process, data which describes the process, meta-data if you will, rather than the data which is the subject of the process. So a payroll system processes pay records but uses data about tax rates so to do. This last would be an example of this meta-data. It is usually good practise to hold meta-data as data, in data files where it can be looked after, rather than having it embedded in the process, where it is apt to get lost.
A rather different sort of example would be the dynamic allocation of bits of process – often called modules, routines or even dynamic link libraries (DLL’s) – with that allocation being controlled, in real time, by the state of the data. It is no longer true, as it was when computers and computing were starting out, that a computer program has to be built at the beginning, a process sometimes called compilation, a term we have borrowed for the frames of LWS-N, and then run to completion without change. Dynamic really means dynamic.
And speculating, there are ways that data can be extracted from process. So, for example, one might monitor the health of a disc drive by monitoring the time it took to do a random access. Or by counting the number of times that the water cooling clicked in. The sequences of random accesses or of clicks-in are data. Or, if one was a spy, one might modulate some public bit of information, like the time to respond to an apparently legitimate request for something or other from the outside world, with data being carried, covertly, in that modulation. Slow but effective.
So not really the clean separation we started out with.
Temperature
By way of example of data and its properties, we now take a look at the sort of information one might hold about the temperature of a motor car. Which sounds simple enough.
The first complication is that one might have a number of temperatures. Temperature in the first cylinder head, temperature in the second cylinder head, temperature in the oil sump, temperature in the passenger space, temperature in the luggage space. Which means that the property label might be quite long, not to say complicated.
The second complication is that one might, despite best efforts, have a variety of scales and a variety of degrees of precision. So this temperature might be Celsius to two significant decimal places, that temperature might be Fahrenheit to the nearest degree. One might deal with this by having several property types for temperature, with each property type having its own meta properties, a bit of jargon meaning a property of a property.
The third complication is that one might record the time and place at which each temperature was taken – with the place being the position of the car at the time in question. So rather than a simple property of our car, we have raised a new thing, a temperature record, and our car has lots of child records for temperature. Not the sort of something that one can see when inspecting the car at all.
Incidentally, the same sort of device might be used to record the position of a moving object, rather than the temperature of an object which happens to be moving. A detail record might record a point position or it might record a short, straight line approximation to a short segment of a journey. Rather as in mathematical proofs one often breaks a continuous curve into such short segments. Noting in passing, that the journey of a real, macroscopic object has to be continuous – if tortuous and sometimes prone to abrupt changes of direction. All of which might be relevant in the data supporting a frame of (LWS-N) consciousness involving moving objects.
The fourth and last complication is that one might have supporting images. One cannot usually take a picture of temperature, but one might, for example, take pictures of the inside of cylinder heads when they get too hot. Pictures which would need to be labelled, possibly extensively labelled, to be useful.
All of which illustrates the point that one’s model of a car, one’s description of a car, might include a lot of stuff that the car itself does not include. What we will be interested in in what follows is what the brain does include, in itself, rather than what we might know about the brain, perhaps as a result of scanning it many times over some more or less protracted period of time.
Guns
 |
| Figure 7 |
So if our thing was the Victory and we wanted to model the guns, what would we do?
Given that we have lots of moveable things, that is to say guns, of one of a small number of kinds or calibres, child records are perhaps more appropriate than part records. If the sort of information which one wanted to keep differed for the different calibres, one might have one kind of child record for each. Alternatively one might have just the one kind of child gun record, and have calibre as a property held on that record.
Alternatively, thinking of the limited capacity of working memory, we might focus on the five decks carrying guns, the three gun decks proper (hence ‘three decker’) plus the two upper decks. So perhaps a child record for each deck, with the number of guns being a property of that deck, rather than attempting to enumerate each and every gun.
More complicated, one might do both. Child records for the decks, but more child records, of another kind, for special guns. Perhaps a child record for the No.13 starboard gun on the lower deck which needed a bit of maintenance attention. The sort of information which would be very much in the mind of that deck’s gun captain (assuming that ship such as the Victory had such a person).
 |
| Figure 8 |
‘Yamato's main battery consisted of nine 46 centimetre (18.1 inch) calibre Type 94 naval guns – the largest calibre of naval artillery ever fitted to a warship ... was capable of firing high-explosive or armour-piercing shells 42 kilometres (26 miles). Her secondary battery comprised twelve 155 millimetre (6.1 inch) guns mounted in four triple turrets (one forward, one aft, two midships), and twelve 127 millimetre (5.0 inch) guns in six twin mounts (three on each side amidships) [second hand] ... In addition, Yamato carried twenty four 25 millimetre (0.98 inch) anti-aircraft guns, primarily mounted amidships. When refitted in 1944 and 1945 for naval engagements in the South Pacific, the secondary battery configuration was changed to six 155 millimetre guns and twenty four 127 millimetre guns, and the number of 25 millimetre anti-aircraft guns was increased to 162 [perhaps there is a mistake here]’.
Here we do not have a nice neat arrangement with each sort of gun on its own deck, indeed, guns are not on decks in the same way at all. Perhaps we would need something more like the upper layers of the parts explosion which be used by the builder of anything large and complicated. A first split into primary and secondary battery. The primary battery carrying three child records, one for each of the three turrets. The secondary battery being further split into its three kinds of gun, with perhaps just a child record for each kind, with anything more complicated being too much for working memory. Or then again, if we were the weapons artificer responsible for the state of the well used barrels, we might want a more barrel-centric data model.
But we have done enough to demonstrate that there are, indeed, lots of ways of modelling the data. The answer mostly lies in what you want to do with it, the reflexive tendency of those of us with stamp collecting or train spotting tendencies to collect whatever information we can lay our hands on notwithstanding.
Also that computers are apt to be a bit pedantic. If they are going to have a record for a gun, they like to have a record for every gun. Or in the case of a bus garage, every bus, not just the ones which happen to have broken down, somewhere out on the Queen’s Highway. Preventative, not just restorative maintenance. Human beings are probably more flexible, which does save storage space, perhaps in contexts like working memory, where this probably matters, but which can also cause problems, with the information not always being to hand when it is wanted.
Properties again
Properties often have names themselves, just like things, and we have already used several such, number of legs, colour, age, date of accession, date of first confinement and variety of cabbage.
We know something about what things have what properties. Most of us know, for example, that most living things have age, that most animals have sex and most humans have marital condition.
That the age of humans in years is usually a non-negative number less than three score and ten. We might know a little more about the distribution of property values, that, for example, that there are roughly the same number of men as women.
We know something about the values of properties. Some properties take yes-no values, some properties take numeric values. Sometimes we know that something has such and such a property without actually knowing the particular value that it takes. While demographers distinguish ‘not stated’ from ‘not applicable’.
A property value which is unique, or at least more or less unique, is often called an identifier. In the real world, tax reference numbers and national insurance numbers are such identifiers, while usually, names such as ‘Elton John’ are not. Humans, generally speaking, are not much good at identifiers.
Contrariwise, we might allow a property to have more than one value. Medical demographers find it helpful, for example, to allow people to have more than one cause of death. Ornithologists might find it helpful to allow birds to have more than one colour or more than one location.
In a computer, all this sort of thing might be captured in what Figure 5 calls meta-properties, otherwise properties of properties, reminding us that properties and property values can be thought of as things in their own right. So a meta-property of a property called ‘age at accession’ might be that it is an integer in the range zero to 99 inclusive, our computer being too mean to allow space for a third digit in this particular context.
Some properties will be visible, we will be able to deduce their value by inspection of the thing in question. Others will be audible or touchable. Other properties again might require closer examination or knowledge of the history. Not a property of the thing as it is at all.
Lastly, we mention a rather special property, albeit a rather speculative one. We might give things a property called ‘position’, the coordinates of the position in the brain which is activated whenever the thing in question comes up, which would light up on a sufficiently high resolution brain scan or on a cleverly placed electrode. With the thing being said to be current, or active, when its position is active. The first catch being that it is not clear whether such positions – which we have tentatively named ‘ensembles’ – exist, whether they are unique and whether they have exclusive occupation of their space. It is clear that there are neurons which fire for particular things, but that is not quite the same thing, not quite enough. See reference 4 for a fairly recent discussion. The second catch being knowing to which brain, in which coordinate system, the position refers to. A particular brain or some standard brain, quite possibly a statistical confection? See reference 5 for some of the issues.
In any event, a property which we are unlikely to have any conscious knowledge of, beyond its possible existence.
A childish digression
Man did not come into the world a million or so years ago with all this stuff ready made. Nor do children now. So we offer a few observations.
Children quite quickly learn about things, that there are things in the world, that much of the world is organised into things, and there are usually lots of different instances of any one thing. But that there is only one Mummy and only one Daddy. They learn to recognise particular instances of things and to remember about them. All this without the benefit of words.
At the age of around two, they start to learn words and to attach those words to things. Common nouns in the first instance. They are surprisingly good at saying that this soft toy or cartoon drawing, bearing some passing resemblance to a real giraffe, is a giraffe. The neural systems which do object recognition are kicking in and learning their stuff, helped along by supporting adults. We note in passing that deaf children herded together have been known to develop their own private sign language, so presumably other children would learn words without adults, but presumably a lot more slowly. See reference 6 for some light coverage or reference 7 for some heavy coverage.
Next up is parts of things, initially derived from human – or perhaps animal – body parts. Legs, arms and noses.
Then properties, which seem to take a bit longer. Things like colour, hot, cold and wet. Property names a bit longer still.
Along the way we have already had some particular objects like Mummy, Daddy and George the giraffe. While some time later, we get relations between objects. This child belongs to that mummy.
So having modelled the business, with increasing success, with neural networks, we can see how a brain might learn to apply words to things. See, for example, reference 10. But it is much less clear how it learns to do the other stuff. Or, to use yet another concept from computing, how is the boot-strapping done?
Conclusions
By way of setting the data scene, we have introduced some of the jargon and issues around data modelling, focussing on things, the properties of things and the relationships between things, with just some passing thoughts about the fact that these things might be moving about.
Adult brains know about all this stuff, possibly implicitly, rather than explicitly and consciously, so we suggest that these same issues are addressed, in one way or another, by said adult brains. Which issues we will come back to in a post to come.
References
Reference 1: https://psmv3.blogspot.com/2018/08/of-cabbages-and-kings.html.
Reference 2: http://psmv3.blogspot.co.uk/2018/05/an-update-on-seeing-red-rectangles.html.
Reference 3: http://psmv3.blogspot.com/2017/01/expressions-and-their-orders.html.
Reference 4: Gnostic cells in the 21st century - Rodrigo Quian Quiroga – 2013.
Reference 5: https://psmv4.blogspot.com/2018/11/places-in-brains.html.
Reference 6: The Deaf in Ottoman Syria, 16th - 18th Centuries - Sara Scalenghe – 2005.
Reference 7: Language in the Light of Evolution - James Hurford - 2007 and 2012.
Reference 8: https://psmv3.blogspot.com/2017/04/a-ship-of-line.html.
Reference 9: The 100-gun ship ‘Victory’ – John McKay – 1987.
Reference 10: ImageNet Large Scale Visual Recognition Challenge - Olga Russakovsky and others – 2014.
Group search key: srd.
No comments:
Post a Comment