SoylentNews

This raises a problem, though, as it implies that the central coordinator needs to know, i.e. have a link to, every object that _MIGHT_ be the foreground object, and that means all objects. This is readily doable by a program, which would just need to register the address of the foreground object, but neurons use a different connection protocol (i.e., physical proximity based connections), and appear to be limited to 10,000 or so connections. It’s hard to see how a network style connection would work, so it seems as if a hierarchical connection could be needed. This, however, would still raise problems with back connections. Also, connecting distant neurons to each other has no obvious mechanism quicker than growing an extension to the axon. Even that would require maintaining a particular chemical gradient for quite a long time, days or weeks.
This is a problem that repeatedly appears. Some known processes depend on a local chemical gradient to work properly, i.e. to properly localize action. Others seem to require coordination of distant connections between particular objects. Neither has an obvious analog in program logic, which implies that this is the wrong order of abstraction. Probably the correct physical analog is the neural column. Since a neural column is composed of thousands of neurons the arguments used above don’t apply. Unfortunately I don’t understand the properties of a neural column anywhere near as well, so it’s much less useful as a model.
Abandoning the analog, then, the active foreground object must register itself centrally so that it can be found by not-currently-linked objects that become active. The links established are two way, but stronger from the new object to the foreground object than in the reverse direction.
Now being the foreground object is a very transient condition.
… … ...
I note a bit of confusion here. The foreground object is not the same as the top object. The top object is the active object that contains all the other active objects. New objects need to be linked both to the top object and the foreground object. This enables “state specific memory” as it means that, e.g., being in a location makes it more likely that other things learned in that location will be brought to mind, but as the link is weak it does not itself suffice. Of course it could be reinforced in the location sufficiently enough that merely thinking of the location would re-activate the memory.

All this makes the “state” of state specific memory quite crucial, and I haven’t yet defined it. To say “it’s all those things that current memories get attached to” is true, but not very useful. I tend to conceive of “state”, in this sense, as the active world image, but this is also a bit vague. Experiments have shown that consciously being aware of something isn’t necessary for it to participate in state. It doesn’t seem useful to conceptualize it as an object, though it is the top container of active objects. Perhaps it’s pure epiphenomenon, and not a real thing at all, but in that case one needs to explain how the activity of the rest of the system could create the illusion that “state” exists, i.e. would provide the effects.
Still, state existing as phenomenon rather than as epiphenomenon seems to create numerous problems. E.g. it seems to exist in innumerable variations and seems to experience partial activation. The only problem it really seems to solve is limiting the necessity for centralized communication. So I need to address that.
Possibly an answer lies in the hierarchical embedding of objects. So, for example, kitchen remains kitchen whether or not the cat is currently being fed. There’s a time linked variation in the “current state of activation” of kitchen. In other words, objects need to allow for components that are not always active (or even present).
The result of this is that objects being linked into any nested subcomponent of the current foreground or top active object are linked into the entire chain, with the strongest link at the lowest level. Repeated stimulation will over time strengthen some links. Links that are not strengthened (unless above a threshold of strength) will decay. Perhaps there can also be degrees of strengthening, so that rubber will be strongly linked with tires, but linked to carts much more weakly.

The “separation” of neurons needs consideration. Clearly sensations that are physically close should be considered close, and likely have direct physical contacts at the neuron level, but other linkages are more difficult. On the other hand, these other linkages probably only happen at a higher level (i.e. at a more abstract level). A snarl combined with teeth being bared is at a relatively high level, white spot combining with white spot to for partial image that will be parsed as a tooth is at a much lower level. At what level is centralized communication needed, and on what basis should this be decided? Well, what’s the purpose of communication? One answer is to create links between objects, so perhaps only top level objects need to link … but this seems insufficient. Actually, the proposal seems roughly equivalent to “frames”, with lots of things left hanging, and that is known not to suffice.
I think that what is needed to solve this problem is the “state specific memory”. When a signal is already a part of the “state” then it doesn’t need the centralized communication, but can simply strengthen and expand the current one. Only when a new signal is being associated with the state does it need to communicate centrally to determine to what stat it is to be added. Since this will generate lots of false or “noisy” connections, it’s important that weak connections fade over time.

The co-occurrence of objects even in description is sufficient to create the perception of a connection. Consider how this is used in the “Grandfather’s Clock” song. It is not without reason that Crowley said that the basic rule of ritual magic is “invoke often”.
These seem to all be things that are implemented via Hebbs Law1, but the mechanism is obscure. When there is a synaptic connection, then the mechanism is reasonably clear, but when there’s no connection except synchronicity it’s harder to explain. It does take many repetitions, so even a weak connection would be reinforced, rather like math tables … in fact probably exactly like math tables … but that doesn’t explain the mechanism. We know that physiologically it’s connected somehow to the hippocampus, so some specialized mechanism is quite appropriate. It has to be done via “passive monitoring”, i.e. via receiving signals from the active neurons … but probably only at a rather high level. And we believe that unusual wiring in this area is behind synethesia.
So … I am assuming that when a cluster of sensations above threshold of strength is activated that a signal is sent to a central function that receives the signals sent during a small interval of time2 and establishes or reinforces a connection between them. This appears as if it might strengthen the perception of boundaries between different clusters of sensation. It would also seem to foster the creation of composite objects. Perhaps it also enables the invention of new composite objects from known pieces.

The persistence of objects means not only that they continue to exist when you can’t observe them, but also, and more primitively, that when you are watching them they remain the same object1. This will probably be inherent in what it means to be an object, but such a concept cannot predate the concept of object.

The distinction between within and without is not easy. Even many adults haven’t really managed it, as denoted in phrases such as “You made me love you.” or “You made me so angry”, where internal actions are attributed to external causes, even though others would react to the same stimuli in different ways. This is probably because episodic events tend to be externally attributed, though of course denial of responsibility is another reason. But originally denial isn’t a reason as the mere existence of a separation between “me” and “not me” isn’t yet given, much less the bounds.

When things are too unusual they just aren’t “noticed”, and when things are too familiar they also aren’t noticed. In the first case the sensations are just perceived as “noise” and are filtered out. In the second case they aren’t interesting enough to notice, and are filtered out. You can override these filters, but it takes careful attention.
This presents a problem in “how to get started”. The resolution is the existence of a few built-in “forms”. One demonstrable one is a view of a smile, I’m sure there are other demonstrable ones. Just how many built-in forms exist is a good question, and some, probably most, will be quite difficult to detect.
Given a few forms to start with they can act as “seeds” for other “objects” to “crystallize” around. But the “how” of this crystallization needs explication.
Well, if one this is recognized, then other sensations can be linked to it. Thus if a smile is recognized, then a forehead can be linked as “occur above smile”, and this will cause it to be “expected” when the smile is seen, and hypothesized when not seen, perhaps because of being obscured by something else. This uses state specific memory to enable weak signals to become established as a definite object. After being stabilized as an object, it can then be recognized in other contexts. Please note that while this example is visual, most of the early linkage is kinesthetic or goniometric, and much will be auditory or cross-modality.

Sunday 2018/10/21
Sensory stimuli preferentially occur in certain built-in forms. Visual examples are lines, arcs of circles, and areas of light or dark.
Is an object initially required to have a particular distribution of features? The face recognizer has been shown to initially require two dark spots above a centered dark spot above an arc, but that is a highly specialized feature that is, if not species specific, probably primate specific. But what about a dresser or refrigerator? A Chair? Does later speed of recognition derive from the initial process? Or is it because since we usually see chairs upright, the upright chair is more readily accessible?
How are objects indexed for access? Seeing the same object from slightly different angles would alter the positioning of features, but many, or even all, of the features would still be present (vertical lines, areas, etc.). Others would be rotated, changing, e.g., the angle at which diagonal lines were seen. (I’m assuming translation rather than rotation of the view with respect to the object. Sometimes some of the features will be obscured, or partially obscured. So what seems to be going on is that the features visible are activated, causing the other features “linked” to them to be activated.
The preceding doesn’t sound sufficiently specific, but this is handled by “state specific memory”, i.e. the entire context is linked into each memory, this includes things like “Where am I?”, as in “what room”, and also emotional state, what recent thoughts have occurred, what other objects have been seen recently (i.e. are still partially active), etc.

Friday 2018/10/20
Object Persistence depends on recognition.
Object: A particular collection of “sensory” impressions.
N.B.: An object won’t be all of the “sensory” impressions from an area of focus, but only those selected as “foreground”.
To form an object, we must l rely on a loose interpretation of “Hebb’s law”, paraphrased as “neurons that fire together, wire together”, so when particular neurons are simultaneously stimulated repeatedly they begin to stimulate each other. Thus perceiving (or imagining) a piece of an object will render the entire memory active.