Are There Limits to Human Learning?

[questionpost]

I mean I know that because there are physical limitations that there has to be some kind of limitation, but it seems like it could in a way go on forever, like you could just keep learning and generalizing previous information or learning and sort of putting it in your subconscious or instinctively to sort of "compress it" and make room for new learning...

 

[Pythagorean]

there's several different way to break up the world to understand it. Where you decide to divide it up and how you choose to interpret it is called an ontology.

It would be hard to hold several conflicting ontologies at once, so that may set a limit on learning. But as far as the lifetime of a human is concerned, you will never learn evertyhing in even one ontology. We probably, as a society, won't learn everything in our ontology either.

 

[questionpost]

there's several different way to break up the world to understand it. Where you decide to divide it up and how you choose to interpret it is called an ontology.

It would be hard to hold several conflicting ontologies at once, so that may set a limit on learning. But as far as the lifetime of a human is concerned, you will never learn evertyhing in even one ontology. We probably, as a society, won't learn everything in our ontology either.

Are there any sort of..."measurable" limits? Like the amount of information someone knowing being equal to some 2 terabytes or something? We won't learn everything, but it's predicted by the end of this century there will be machines that can exceed the capacity of the human race, but what exactly is that capacity?

 

[Ryan_m_b]

Are there any sort of..."measurable" limits? Like the amount of information someone knowing being equal to some 2 terabytes or something? We won't learn everything, but it's predicted by the end of this century there will be machines that can exceed the capacity of the human race, but what exactly is that capacity?

No; not until we have a better understanding of how memory is stored in the brain. There are people (popular [strike]science fiction[/strike] futurologist writers) who claim that it can be worked out in terms of bits and bytes but their methods are always deeply flawed. For example some people claim the human brain's memory capacity to correspond to one bit per synapse but that shows no understanding of the function of synapses and the process of memory storage in the brain.

 

[Pythagorean]

Are there any sort of..."measurable" limits? Like the amount of information someone knowing being equal to some 2 terabytes or something? We won't learn everything, but it's predicted by the end of this century there will be machines that can exceed the capacity of the human race, but what exactly is that capacity?

As Ryan said, the computer view of memory is not appropriate.

First of all, memory is kind of an umbrella term. There are several different types of memory (procedural, episodic, and semantic for instance) and the idea that we just kind of store memories indefinitely like a computer is flawed. Every time we recall information, some modeling has to take place to "fill in the gaps". This is why eye witnesses are infamous for their lack of reliability... every time they think about the past, some information will be lost, and some will be replaced with incorrect information. One of the brain's functional goals is to preserve the integrity of the organisms world model: in this way, minor factual details may be altered to keep the world model consistent.

So most memories are worked into a generalized model of the universe via the hippocampus, which (so says the current theory) takes a bunch of little memories and integrates them into one memory, then while you sleep, the hippocampus updates the rest of your neocortex with the generalizations (for instance, updating your semantic memory via hebbian learning by evoking activity in the temporal lobes... if you had a new experience with a dog the day before, your definition of a dog might be updated if new, novel information about dogs was discovered).

Christoph Koch does a lot of work with this. He finds "concept" neurons in the temporal lobes that only fire in response to a particular holistic stimulus (i.e. the "Beatles" neuron will fire only when the subject sees a picture of the Beatles or even the word 'Beatles' on the screen; so it's the concept, not the visual stimulus)

Of course, this isn't to say that that one neuron completely encodes the Beatles. The neuron is involved in a complicated network, so we are only looking at the tip of the iceberg.


Moran Cerf, Nikhil Thiruvengadam, Florian Mormann,Alexander Kraskov,Rodrigo Quian Quiroga, Christof Koch & Itzhak Fried, On-line, voluntary control of human temporal lobe neurons, Nature 467, 1104–1108 (28 October 2010) doi:10.1038/nature09510

 

[WhoWee]

As Ryan said, the computer view of memory is not appropriate.

First of all, memory is kind of an umbrella term. There are several different types of memory (procedural, episodic, and semantic for instance) and the idea that we just kind of store memories indefinitely like a computer is flawed. Every time we recall information, some modeling has to take place to "fill in the gaps". This is why eye witnesses are infamous for their lack of reliability... every time they think about the past, some information will be lost, and some will be replaced with incorrect information. One of the brain's functional goals is to preserve the integrity of the organisms world model: in this way, minor factual details may be altered to keep the world model consistent.

So most memories are worked into a generalized model of the universe via the hippocampus, which (so says the current theory) takes a bunch of little memories and integrates them into one memory, then while you sleep, the hippocampus updates the rest of your neocortex with the generalizations (for instance, updating your semantic memory via hebbian learning by evoking activity in the temporal lobes... if you had a new experience with a dog the day before, your definition of a dog might be updated if new, novel information about dogs was discovered).

Christoph Koch does a lot of work with this. He finds "concept" neurons in the temporal lobes that only fire in response to a particular holistic stimulus (i.e. the "Beatles" neuron will fire only when the subject sees a picture of the Beatles or even the word 'Beatles' on the screen; so it's the concept, not the visual stimulus)

Of course, this isn't to say that that one neuron completely encodes the Beatles. The neuron is involved in a complicated network, so we are only looking at the tip of the iceberg.


Moran Cerf, Nikhil Thiruvengadam, Florian Mormann,Alexander Kraskov,Rodrigo Quian Quiroga, Christof Koch & Itzhak Fried, On-line, voluntary control of human temporal lobe neurons, Nature 467, 1104–1108 (28 October 2010) doi:10.1038/nature09510

Very interesting. Would procedural type memories include problem solving skills - specifically use of formulas?

 

[Pythagorean]

I would think that procedural learning is fundamentally associated with problem solving, but there is probably also a lot of semantic memory involved, especially during "ah ha!" moments.

Once you have it down and you're just grinding away with the same old algorithm, it's probably almost entirely procedural.

 

[atyy]

Within a given network model, and with respect to a certain sort of task, there are estimates of memory capacity. Here's one estimate together with an experiemental test of some aspects of the model.

http://online.itp.ucsb.edu/online/brain_m11/compte/

 

[Pythagorean]

I think we have to be careful associating memory only with neurons and synaptic strength. For instance, astrocytes are known to program synaptic strengths, which we know is associated with memory, but to what extent do astrocytes store memory (or make predictions that look like memories)?

There's also the matter of Orb2, a prion-like protein that is correlated with the persistence of long-term memories in flies. There's no mention of synaptic strength in the article, indicating a role for the underlying molecular networks in memory management.

Inevitably, I think the problem is that memory is an emergent phenomena with several underlying systems. And it's also a bit conflating to consider the modeling perspective. I know there are several times where I simply remodel an answer to a question from first principles. So I have only remembered the first principles, not the information I can reproduce with the first principles. Though you would surely measure neural activity as I worked to make predictions from those first principles.

----------------------------------------------------------------------------
references:

De Pittà M , Volman V , Berry H , Ben-Jacob E , 2011 A Tale of Two Stories: Astrocyte Regulation of Synaptic Depression and Facilitation. PLoS Comput Biol 7(12): e1002293. doi:10.1371/journal.pcbi.1002293

http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1002293ritical Role of Amyloid-like Oligomers of Drosophila Orb2 in the Persistence of Memory

Majumdar, Amitabha; Cesario, Wanda*Colón; White-Grindley, Erica; Jiang, Huoqing; Ren, Fengzhen; Khan, Mohammed*Repon; Li, Liying; Choi, Edward*Man-Lik; Kannan, Kasthuri; Guo, Fengli; Unruh, Jay; Slaughter, Brian; Si, Kausik

Cell doi:10.1016/j.cell.2012.01.004 (volume 148 issue 3 pp.515 - 529)

http://www.cell.com/abstract/S0092-8674(12)00005-0

 

[WhoWee]

I would think that procedural learning is fundamentally associated with problem solving, but there is probably also a lot of semantic memory involved, especially during "ah ha!" moments.

Once you have it down and you're just grinding away with the same old algorithm, it's probably almost entirely procedural.

Unlike the eyewitness example - when problem solving - is the procedural memory typically recalled in whole with the gaps filled in with semantic memory (which erodes) per the situation?

 

[Pythagorean]

Unlike the eyewitness example - when problem solving - is the procedural memory typically recalled in whole with the gaps filled in with semantic memory (which erodes) per the situation?

I wouldn't think semantic memory becomes very important once people learn something and commit it to procedural memory. For instance, when we get in our car and suddenly we're at our destination, and we can't recall having made any driving decisions at all, it's generally because we've handed the task off to our procedural memory and no new novel information was presented (i.e. a kid wasn't wondering in the middle of the road) that would have caused the procedural memory to suddenly hand the task back to our cognition asking "what do we do? what do we do? what do we do?".

Procedural memory is considered "implicit memory", which means it's not a cognitive process, while semantic memory is "explicit memory" (and thus cognitive). Now that you know the english language, you don't have to sit and think about ordering your verbs and nouns and adjectives, you just have a general idea of what you want to say and your procedural memory does all the work for you (of course, you spent many years in your youth learning the material, at which point it involved a lot more semantic memory).

You may get bored and think about your procedural tasks and alter your semantic memory about them (by sensitizing yourself to them and kind of "imagining" they are something new) or you may make a mistake that turns out to be lucrative and develop a new procedure, but most of the time, people are desensitized to their procedural tasks and just do them without thinking about them.

 

[Pythagorean]

Hrm, actually... I'm sure procedural memory almost always draws on semantic memory with language and probably also many procedural tasks that require object identification (such as driving). I was thinking more about programming of semantic memory in the post above.

Physical tasks (like crocheting, say) may draw more on somatic memory.

 

[Bobbywhy]

If a person spent every waking moment of every day learning new things it seems to me the only limit on her learning would be dying.

 

[DubaUe]

(procedural, episodic, and semantic for instance)

what are these 3 types of memory and how do they work?

 

[Pythagorean]

what are these 3 types of memory and how do they work?

procedural is generally completely behavioral and utilizes the basal ganglia. It has a lot to do with muscle control and order of operations and is influenced heavily by reward/fear conditioning.

episodic memory is remembering events as occurrences in time. Often, several episodic memories will lay the groundwork for a semantic memory (so all of your encounters with apples were episodic, but your concept/definition of what an apple is, is semantic, but probably draws from those episodic memries). Episodic memory is managed in the hippocampus, post binding (so several sensory inputs merge together to tell a contextual story).

Semantic memory is how we identify objects and concepts. The episodic memories throughout your day are processed by your hippocampus. While you sleep, your hippocampus "writes to" the neocortex, particularly the temporal lobes, which has a strong association with semantic memory. So the theory is that your hippocampus takes your days interactions and organizes them by subject into your semantic memory, updating your definitions of things.

A lot of this has been studied experimentally by people like Mark Gluck, looking at how Parkinsons patients vs. Amnesia patients learn differently (Parkinson's patients have lesions/damage in the basal ganglia, amnesiacs have lesions/damage in the hippocampus). He has a good presentation on his findings:

 

[Hobin]

If a person spent every waking moment of every day learning new things it seems to me the only limit on her learning would be dying.

Surely you don't think spending every waking moment learning is the most efficient way to learn? (Yes, yes, I know you didn't say so literally.)

 

[Bobbywhy]

I mean I know that because there are physical limitations that there has to be some kind of limitation, but it seems like it could in a way go on forever, like you could just keep learning and generalizing previous information or learning and sort of putting it in your subconscious or instinctively to sort of "compress it" and make room for new learning...

The OP asked about physical limitations to learning. You are right that it would not be reasonable to "spend every waking moment of every day learning" because that is not the way we humans operate. My intention was to say that I do not see any physical limitation to how much we can learn.

When a person dressed in a funny cap and gown receives a rolled-up sheepskin that says they have achieved a certain level of academic competency does that mean their learning capacity is completely full? No. To me, learning is a process that continues long after formal schooling. For a curious observer each level of new understanding in a subject area opens up more questions. I have never experienced any limitation of ability to learn more. Yes, distractions, physical fatique, and thousands of other human needs intervene to limit our learning new things. But I repeat: the only definite limitation is death.