brains need redoing.
say we have properties for each object: colour, hardness, size, temperature, etc.
norns can see size and colour, and feel hardness and temp.
this allows much greater generalisation- two things may not have to be exactly the same size for a norn to try applying knowledge about one to the other. perhaps it could even learn that objects in certain size ranges (and with other properties) behave a certain way: it classifies them itself, giving it far greater flexibility.
however, thios complicates vision. before, a norn was only capable of seeing 40 things, one for each category, and the engine chose which to show it. so it only needed 40 neurons.
now we have zillions of different objects, and the engine can't sort them neatly into groups. a lot more realistic, though :)
gah, people are so noisy...
anyway. how can we simplify vision?
in humans, our eyes do a lot of processing before they pass the information on to our brains... we don't normally see a bunch of different colours, we see objects, patterns.
one of the problems that immediately came to my mind is 'what if there are a hundred of the same object near the norn? how can its tiny brain cope?'
well, let's think about how humans do it. when you look at a crowd, you don't think "there are 374 people there", you think "there are lots of people there" or "there is a crowd there". we group things- we don't see a zillion blades of grass on the lawn, we see "a lawn", with patches of grass, some rocks, perhaps bare spots. if you try to think of your front lawn, do you remember where every blade of grass is? no. do you even remember where the patches of bare ground are? probably not, unless you've been doing a lot of gardening lately. you remember it's a lawn, that there is, say, a tree over in one corner, perhaps a few bare patches scattered about the place. you only remember the detailks that stand out, things that are unusual.
er, I think my train of thought has left without me again...
anyway. how would we turn this into something useful for norns? first of all, perhaps they should have the same 'one, two, three, many' counting system that we normally use (for everyday things, not for math, of course). but how? I'm not really sure...
we have a limited number of neurons. somehow we must use them to convey meaningful information about the environment- what objects the norn can see, an approximate location, and what colour and size they are. but there is no limit on the number of objects that could be in sight... hrm.
somehow we have to convey the idea of a group of similar objects.
maybe we should simplify this more, first, by taking out colour; replace vision with something more like radar, for now. 
hmm. but how do *we* see the world?
when I walk into a room, I usually glance around. I notice that there are windows, desks, computers and so on, but these quickly become background details I can easily ignore. I don't care that there is a pencil and ruler on the desk beside me, unless I am looking for a pencil or ruler, or I'm concentrating specifically on the contents of the desk. I will not notice the pencil again- until it is gone. then I will notice that something has changed, although I might not remember what. hmm, that's another thing about humans, they can remember what objects should be in a room even when they're not there. norns can't do that at all.
so... let's try thinking about the world in terms of what changes. a norn doesn't need to be constantly bombarded with the news that there are three big purple things moving around nearby; it should only notice them if they do something interesting. so let's say that we only tell it when something new comes along, or when something something leaves, or changes. but what if raindrops are constantly coming and going? it needs a way to become used to that and no longer notice it, the way we become used to the sounds of computer fans running in the background. then, it would only take notice when the rain stopped. 
but we've come up against another problem: if the norns only notice interestinbg things, how do they remember what objects are in the room?
well, while I'm staring at this laptop, I tend to forget a lot about where I am. but if I want something, I need only look up and see if it's here. and if there's something interesting in the room when I come in, i remember it... but let's say a norn comes into a room that has food in it, but at the time it's not hungrey so it goes off to play with the ball in the corner. after a while, it gets tired of the ball and becomes hungry. what now? does the norn think "I want to eat food", and then look at the room to see if there's food there?  does his growing hunger make the food more interesting, so that his 'eyes' (the engine, reallly) begin telling him again that there is food here, even though the food itself has not changed? does the norn have some kind of 'search' action, that allows it to find out if there is anything in the room close to the properties of food, without having to sift through all the information available on the room? or what?
personally I think a combination of 1 and 2 might be good.
I should mention that when I said "food" here, I was not referring to a specific food object, or group. I was referring to whatever the norn considered the best thing to eat at the time. perhaps "small, yellow, soft thing" would be more accurate, but 'food' is easier to type. :P
hrm. this brings us to another problem: language. with no classifiers, norns would not havbe an easy language for communication with humans.
imho, the whole 'language' thing is unrealistic anyway.  but it is important- the user needs a way to communicate with the norn, and using hte science kit feels like cheating.
so... what do we do about language? we still have verbs, at least. but nouns are the complicated part...  do we go back to having the user teach the norn words for objects? if so, can the norn generalise about nouns? it would be a pain having to teach iut the word for every single darn object- there are already too many words in c3. but how would we teach a norn that words apply to a group of things? and how would this affect its learning? *can* a norn be designed so that it can learn that a set of different-looking objects can be grouped into a category? perhaps "things-that-are-edible"? it could be very useful for the norn, I suppose. but how would a norn understand the 'good-to-eat' property? something that reduces the hunger drive, I suppose. (I'd like to do away with these three separate hunger drives. I don't see how they help the norn, they're just extra drivesit needs to satisfy. I certainly can't tell whether *I* need carbs or protein or whatever.)
then again... I have no idea whatsoever how to design a brain that could cope with objects being in multiple categories. especially if these categories were chosen by the norn. (why have a "big and purple" category if you've never seen anything that's big and purple?)
soo... we have a bunch of objects. let's say norns can learn that small yellow objects are usually soft, and small yellow soft objects will reduce hunger when eaten. how do we do that? well, I can't remember precisely what the old way of doing it was. iirc either the dendrites were set up so that "if it is food and I am hungry" the norn would be encouraged to eat by a strong connection to the 'eat' neuron in whatever the lobe was called... or... ah, yes, that was it, I think.
so... now we have 'if it is small and yellow and soft, and I am hungry'. 
there's another thing I meant to mention a while ago. a norn can get general information about what's in the world from senses like smell and vision... but what about specific information on the object it's paying attention to? (I think I'll call the object _it_ from now on, since that';s whta it is in caos.) so. not only does the norn get info about interesting things happening in the background, it can also 'look' at _it_ to get specific information about the colour and size of it, and if it picks it up, it finds out how hard it is, and how hot it is.
(hrm. if a norn picks up a burning hot object, is it fair to cause pain if it had no idea it would be hot? either dangerously hot objects should be identifiable by other attributes (eg. large orange fireplace) or the norn shhould be able to sense heat simply by being near it, the same way we can put our hand near a hot stove to see if it's safe to touch.)
egads, coffee is actually beginning to smell good. I *really* need sleep.
anyway... yeah. norns can look at and pick up objects to find out what they're like. it'd be cool if they could learn stuff like 'small yellow objects are likely to be soft', too (and rather useful if it can learn 'big orange objects are very hot'). hmm. btw, do norns have an involuntary 'drop everything when in extreme pain' action? after all, humans have that, so it's only fair for norns to have it too. otherwise they might just stand there, holding the burning hot object and getting severely burnt...
anyway. back to the topic: can this actually be done? and can it be done without needing a supercomputer?
do we put in a new lobe for information about _it_? or do we have a separate lobe for each property? I don't know much about lobes.
how do we even deal with the large amounts of information the brain will now be getting? it's no longer a simple "it is food". now the norn has to deal with 'it is this colour and it is this size and it is this hardness..." I'm not really sure how you tell it that. do you have a neuron that sends a larger number the larger the object is? if that's possible, it might make things much easier- each property is simply a number between zero and one. the hotter/bigger/whatever, the larger the number. 
not sure how that would work for colours, but it can't be too bad- after all, these aren't *real* colours, these are just a property of the object that we choose to call colour (and will probably have some relation to the colour of the sprite, unless someone wants to confuse the user). 
anyway... now in theory the norn can learn that it's good to eat small yellow things, and bad to pick up large orange things. (hrm. this blurs the meaning of "pick up". should norns be able to to attempt to pick up norns and other heavy things? or should it have another verb for finding out the hardness n'stuff of non-pick-up-able objects?)
in theory, the norn will also be much better at generalising- it can attempt to eat small red things and big yellow things, until it learns a better way of dealing with these objects. perhaps it can even generalise so far as to try eating things that are close to being yellow, and not very small- if something has a size of .25, and something else has a size of .26, are they really all that different? 
humans can tell that a penny and a dime (canadian money) are different sizes when they're on top of each other, but if you have two objects the same colour and shape, one the size of a penny and one the size of a dime, and they're in different rooms, you won't be able to tell which is bigger without a ruler.  nor will you care, unless you find out that the bigger one does something drastically different.
hrm, but how would norns generalise about the properties? let's say it encounters a new object. this has a size of .3 and a colour of .7. in the past, the norn has learnt that objects with a size of .2 and a color of .1 are good to eat, but objects with a size of .9 and a color of .2 are painfully hot. what would be the most sensible thing for it to do? 
well, neither of the previous objects were anywhere near the colour of this one, but hte good-to-eat object was pretty close in size. so the most sensible course of action would be to try eating it.
but how would a norn decide that? it would have to compare the information it has about the cureent object with the memories it has (well, they're not really memories, but whatever). but this isn't a simple case of seeing if two values match or not- we want to know how *well* they match.
I think I'll leave that until I know more about how the neurons work.