Those seem to be categories of action, rather than knowledge. Many of the examples of both types depend on tacit knowledge, and are distinguished by type 2 requiring some explicit mental intervention to initiate or sustain (e.g. sustain a higher than normal walking rate.)
Well it was like what you mention. Tacit knowledge must be transferred by showing, hands on learning, but was surprisingly little affected by culture and borders in global companies which was a dimension I was investigating.
After all the work of putting together a thesis on the subject, it all boils down to the example of how to you train a new chef to be as good as his "master".
This is still a hot topic amongst those who continue to research the Social science around Innovation, economics, sociological, management studies and so on. I enjoyed the topic, but being chained to a laptop for 6 months of thesis writing not do much.
I prefer Slavoj Zizek classification into knowns and unknowns.
- Known-Knowns: things we know that we know
- Known-Unknowns: things that we know we don’t know
- Unkown-Unknowns: things we don’t know we don’t know
- Unknown-Knowns: “The things we don’t know that we know-which is precisely, the Freudian unconscious, the “knowledge which doesn’t know itself,” as Lacan used to say. “. This is better known as culture.
Polanyi's paradox isn't an alternative to that classification — rather, it explicitly states that a surprisingly large amount of knowledge falls into the "unknown knowns" classification, in areas that defy your "this is better known as culture" description.
Yeah, “unknown knowns” is better known as “the weirdo over there who just sat at the piano and started playing it perfectly having never played it before”. The parent is taking it too meta.
Rumseld famously classified the difference between known unknowns and unknown unknowns; Zizek typically uses Rumsfeld as a jumping-off point to elucidate the concept of unknown knowns. (Though I think Zizek means it more in the sense of belief systems, norms, and ideology, in the vein of "can't tell a fish about water"; whereas Polyani seems to refer primarily to genetic instincts, subconscious cognitive processes, motor skills learned prior to language, etc.)
We are be born with some functions at level 4, as well as acquire new skills without going through stages 1-3, otherwise there's no possible bootstrapping mechanism.
At bottom, the reason is always "because the structures in my brain create a signal pattern". Understanding that is really the ultimate knowledge.
Well he was mocking Rumsfeld there a bit, and added "Unknown-Knowns" to fill in the last quadrant of the 4x4 matrix, basically saying it is the ideology or subconscious at work.
But I think the idea is a bit different in that Polanyi's paradox is at a higher level of awareness. Say if someone can play chess but may not able to effectively teach it. Or the linked article in the wiki has jigsaw puzzles as an example.
Žižek was talking about deeper processes. Things like people acting according to say Christian ethics while rationally disavowing them and saying they oppose them. His traditional joke about is Niels Bohr with his horseshoe above the door: https://www.youtube.com/watch?v=lTAp3KZwYl0
Next time someone asks me how does a modern artificial neural network really work - I'll just ask them "how do you know that my face does indeed look like mine?". Or "how do you recognize people by their voices?". Or this creepy one: "can you consciously stop decoding the English I'm speaking now?". This Wikipedia page can serve as an extension of the argument.
Well, I would say that to my mind we are not there. That is, despite the label of neural network, we do have a bottom up explanation of how it works (although it doesn't guaranties that this model is "what really happens out there").
On the other hand, despite the advances in neurosciences, as far as I know we don't have such a consensual bottom up model for human behaviors.
We have a bottom up model for very simple animals. But, even a fruit fly has 250,000 neurons that’s a lot of complexity.
So, saying we understand some reflexive human behaviors from the ground up is actually a lot of progress. We are even applying that knowledge in many useful ways.
> activities based on tacit knowledge include recognizing a face, driving a car, riding a bike
As far as I know, there are different classes of tacit knowledge here:
Driving a car, riding a bike etc are learned behaviour, recent in evolutionary time, and might also in a few generations not commonly learned any more, as e.g. riding a horse is uncommon now.
But recognising a face (or not being good at it) is innate in human brain structure, and not recent in evolutionary time.
One that might bridge that gap is learning a language.
It's almost certainly much more recent than recognizing a face, but still something humans have been doing for a long time, and is a near-universal human behavior. Like the others, though, it needs to be learned, has a strong cultural component, and tends to happen with the help of others.
It's also interesting because there's an entire academic field devoted to its study, and, at least from my peripheral perspective, they seem to be constantly bumping up against Polanyi's Paradox. I've seen more than one academic argue that the set of grammatical rules that we can accurately describe is a subset of the grammatical rules that we use when communicating. Which, if true, is pretty interesting, since language is a social construct - the implication is that we can fabricate and reliably transmit information that we can't actually express.
> the set of grammatical rules that we can accurately describe is a subset of the grammatical rules that we use when communicating
Although rules do move from one camp to the other. For instance, people (myself included) find this article fascinating because it makes conscious a rule that English-speakers use unconsciously
I've been reading Alan Watts' "the way of Zen" recently and this point forms a strong central pillar of his explanation. What we think of as our mind or our self, the conscious mind, seems to be basically a paper thin director on top of all the real functionality which is much more powerful and evolutionarily much older.
Relatedly, our use of language to describe and identify things and phenomena falls prey to the map-territory relation in a way that can warp our thinking. But I can't do his explanations justice, I can only recommend picking up the book if you're interested in these kinds of cognitive paradoxes.
I don't understand why this is a paradox. Isn't it just obvious that no system can contain a complete model of itself? That would be an infinite regress and hence would require infinite information.
Further, most of the 'program' is in the cellular mechanics. DNA does nothing just sitting on a table. It must have a machine to 'execute on', and the cellular machinery is a whopper.
I liken DNA to a paper tape with two punches: one for 'man' and one for 'mouse'. You run it through an enormous machine, and depending on which hole is punched you get a man or a mouse. But the paper tape is an almost trivial part of the process.
DNA does not encode a complete model of the human that has the DNA, it encodes a bunch of potential clones of that human. This is an abstract, incomplete model. To encode a complete model you would need a complete physical measurement of this particular human as a physical system. But there is a theorem by Thomas Breuer that no physical system can contain a complete measurement of its own state.
Humans don't need to self-contain an entire model of a human in their brain. Multiple humans can come together and write disparate information down, for example.
Humans are also pretty lousy at predicting how other human brains will react, though. I don't see how that's an example of the Paradox. With a woefully incomplete model, obviously you'll get inaccurate predictions.
What does it mean for a system to contain a complete model of itself?
If I can fit all the firmware, schematics and silicon mask layouts of a laptop on that laptop's disk, in what sense does it not contain a complete model?
Concretely, consider that the CAD files describing a terabyte of flash are less than a terabyte, because most parts are repeated in a grid. And you can add more terabytes without increasing the size of the model noticeably. So the storage capacity of a system isn't limited to the descriptive complexity of the system. It's probably O(log N) with a huge constant. So you can have as much disk space as you like to store a model.
My understanding: You are containing the 'plan' or instructions to recreate or describe the system, but not the system itself. The system contains more information than what the plan can actually describe, even if you are storing duplicates in such a way as to 'reduce' it. Your plan won't be able to store, for example, the information conveyed to the nerves by touching the laptop, you may have a description, but it will only ever be a description.
My most important lesson in working with clients as a programmer, is that asking them how they want their program to work is like asking a native speaker to tell you the rules of grammar in their language. They have often lived with the rules of how their business process works for so long, it is intuitive, and no longer available to their conscious understanding, if it ever was.
Show them a scenario, though ("X, Y, and Z happen, so the program does A and B") and they can usually immediately tell you if that's the wrong or right thing for the program to do.
I have the impression that programmers, as a group, are mostly not great at understanding this, and it is a source of frustration for many of us when we program exactly what they told us to, and then it's "wrong". It is wrong. Like a linguist talking to a native speaker of another language, you often have to ask them about scenarios and piece together the actual business rules yourself.
The worst part is when the customer _thinks_ they know the rules, and then don't (consciously), and you have to help them through that cognitive dissonance of realizing that the business rules they "know" aren't actually the same as the ones they know.
The worst part is when the customer _thinks_ they know the rules, and then don't (consciously), and you have to help them through that cognitive dissonance of realizing that the business rules they "know" aren't actually the same as the ones they know.
A coworker of mine used to term this, "Applied Philosophy."
There was a post on HN a few years ago, where a game developer had a game where all of the users were asking for an Undo key. This would have completely broken the game! So he looked into his data, and figured out that what people were mad about was moving into a stream tile and taking damage or dying. So instead, he made it impossible to move onto a stream tile, and the problem was solved.
They knew _that_ there was a problem, but not necessarily what the problem really was, or how it should be (programmatically) solved. Sounds similar. The answer bubbles up from the subconscious ('system 1' as some put it).
The article discusses the argument that statistical machine learning is a resolution to the paradox - as it shows that an agent can learn something without being given explicit instruction. I think this is basically a sound argument, and I find Polanyi pretty unconvincing.
It's worth noting that a similar argument (albeit without a demonstration as concrete as modern ML) was actually advanced in the 1960s by Wilfrid Sellars. [0]
> There is all the difference in the world between knowing how to ride a bicycle and knowing that a steady pressure by the legs of a balanced person on the pedals would result in forward motion ...It can be argued that anything which can be properly called 'knowing how to do something' presupposes a body of knowledge that; or, to put it differently, knowledge of truth or facts. If this were so, then the statement that 'ducks know how to swim' would be as metaphorical as the statement that they know that water supports them.
47 comments
[ 5.6 ms ] story [ 107 ms ] threadWhat answer did you find?
After all the work of putting together a thesis on the subject, it all boils down to the example of how to you train a new chef to be as good as his "master".
This is still a hot topic amongst those who continue to research the Social science around Innovation, economics, sociological, management studies and so on. I enjoyed the topic, but being chained to a laptop for 6 months of thesis writing not do much.
- Known-Knowns: things we know that we know
- Known-Unknowns: things that we know we don’t know
- Unkown-Unknowns: things we don’t know we don’t know
- Unknown-Knowns: “The things we don’t know that we know-which is precisely, the Freudian unconscious, the “knowledge which doesn’t know itself,” as Lacan used to say. “. This is better known as culture.
https://en.wikipedia.org/wiki/Johari_window
We are be born with some functions at level 4, as well as acquire new skills without going through stages 1-3, otherwise there's no possible bootstrapping mechanism.
At bottom, the reason is always "because the structures in my brain create a signal pattern". Understanding that is really the ultimate knowledge.
But I think the idea is a bit different in that Polanyi's paradox is at a higher level of awareness. Say if someone can play chess but may not able to effectively teach it. Or the linked article in the wiki has jigsaw puzzles as an example.
Žižek was talking about deeper processes. Things like people acting according to say Christian ethics while rationally disavowing them and saying they oppose them. His traditional joke about is Niels Bohr with his horseshoe above the door: https://www.youtube.com/watch?v=lTAp3KZwYl0
On the other hand, despite the advances in neurosciences, as far as I know we don't have such a consensual bottom up model for human behaviors.
So, saying we understand some reflexive human behaviors from the ground up is actually a lot of progress. We are even applying that knowledge in many useful ways.
as a speaker of English as a second language this used to happen to me once I was sufficiently tired.
I can almost do it in my native language if I am distracted "just right" (easier if it's being spoken with a different accent)
As far as I know, there are different classes of tacit knowledge here:
Driving a car, riding a bike etc are learned behaviour, recent in evolutionary time, and might also in a few generations not commonly learned any more, as e.g. riding a horse is uncommon now.
But recognising a face (or not being good at it) is innate in human brain structure, and not recent in evolutionary time.
It's almost certainly much more recent than recognizing a face, but still something humans have been doing for a long time, and is a near-universal human behavior. Like the others, though, it needs to be learned, has a strong cultural component, and tends to happen with the help of others.
It's also interesting because there's an entire academic field devoted to its study, and, at least from my peripheral perspective, they seem to be constantly bumping up against Polanyi's Paradox. I've seen more than one academic argue that the set of grammatical rules that we can accurately describe is a subset of the grammatical rules that we use when communicating. Which, if true, is pretty interesting, since language is a social construct - the implication is that we can fabricate and reliably transmit information that we can't actually express.
Although rules do move from one camp to the other. For instance, people (myself included) find this article fascinating because it makes conscious a rule that English-speakers use unconsciously
https://www.theguardian.com/commentisfree/2016/sep/13/senten...
Relatedly, our use of language to describe and identify things and phenomena falls prey to the map-territory relation in a way that can warp our thinking. But I can't do his explanations justice, I can only recommend picking up the book if you're interested in these kinds of cognitive paradoxes.
(Watts is great.)
Maybe you're saying that a human can't fit into a human? I agree it's not a paradox, but not for the reasons you seem to be implying.
if you mean DNA, then humans do not "speak" (nor think in) this code, heck... teams of people with post-docs don't even really understand it.
I liken DNA to a paper tape with two punches: one for 'man' and one for 'mouse'. You run it through an enormous machine, and depending on which hole is punched you get a man or a mouse. But the paper tape is an almost trivial part of the process.
That's true, but irrelevant. We're talking about modeling a system here, not reproducing one. Those are not the same thing.
If I can fit all the firmware, schematics and silicon mask layouts of a laptop on that laptop's disk, in what sense does it not contain a complete model?
Concretely, consider that the CAD files describing a terabyte of flash are less than a terabyte, because most parts are repeated in a grid. And you can add more terabytes without increasing the size of the model noticeably. So the storage capacity of a system isn't limited to the descriptive complexity of the system. It's probably O(log N) with a huge constant. So you can have as much disk space as you like to store a model.
Well, sure; we're asking for a model, not a copy.
Show them a scenario, though ("X, Y, and Z happen, so the program does A and B") and they can usually immediately tell you if that's the wrong or right thing for the program to do.
I have the impression that programmers, as a group, are mostly not great at understanding this, and it is a source of frustration for many of us when we program exactly what they told us to, and then it's "wrong". It is wrong. Like a linguist talking to a native speaker of another language, you often have to ask them about scenarios and piece together the actual business rules yourself.
The worst part is when the customer _thinks_ they know the rules, and then don't (consciously), and you have to help them through that cognitive dissonance of realizing that the business rules they "know" aren't actually the same as the ones they know.
A coworker of mine used to term this, "Applied Philosophy."
There was a post on HN a few years ago, where a game developer had a game where all of the users were asking for an Undo key. This would have completely broken the game! So he looked into his data, and figured out that what people were mad about was moving into a stream tile and taking damage or dying. So instead, he made it impossible to move onto a stream tile, and the problem was solved.
https://i.imgur.com/tp3eLb4.png
https://www.press.uchicago.edu/ucp/books/book/chicago/T/bo84...
It's worth noting that a similar argument (albeit without a demonstration as concrete as modern ML) was actually advanced in the 1960s by Wilfrid Sellars. [0]
> There is all the difference in the world between knowing how to ride a bicycle and knowing that a steady pressure by the legs of a balanced person on the pedals would result in forward motion ...It can be argued that anything which can be properly called 'knowing how to do something' presupposes a body of knowledge that; or, to put it differently, knowledge of truth or facts. If this were so, then the statement that 'ducks know how to swim' would be as metaphorical as the statement that they know that water supports them.
[0] Philosophy and the Scientific Image of Man - Wilfrid Sellars http://www.ditext.com/sellars/psim.html