They say that knowledge is power. So people learn to acquire more and more knowledge. It is generally accepted that the more knowledge let you have the more power to understand the world and to solve problems better.
Meanwhile, AI developers train LLMs to acquire knowledge thoroughly. Accordingly, they are more intelligent than any scholar. And nowadays even scholars ask them knowledge or information.
By the way, what is information vs. knowledge? Is information different from knowledge? Many information scientists have defined information in so many ways that to let you know them would make you confused that much. Instead, let’s make it simple. Say, “I know that she is pretty” sounds simple, while “I inform that she is pretty” suggests that communication to someone matters. The primary virtue of ‘’knowledge’’ may be its truthfulness so that she is pretty indeed, while that of ‘’information’’ may be its communication to someone so that its usefulness or relevance matters rather than its truthfulness. Indeed relevance matters in information science.
ㅡㅡ
I think this is a promising way to cut through a very old difficulty in information science. Instead of beginning with dozens of definitions of information and knowledge, you begin with ordinary language and the different relations implied by the words.
Consider your two sentences:
“I know that she is pretty.”
“I inform someone that she is pretty.”
The second needs a small grammatical adjustment—normally inform requires someone who is informed—but that actually strengthens your argument.
The verb know establishes primarily a relation between a knower and what is known:
person ↔ proposition/world
The verb inform introduces another relation:
source → content → recipient
Thus information, in its ordinary communicative sense, already implies movement toward somebody. Knowledge need not.
This gives us a surprisingly useful distinction.
Knowledge: truth is central
When I say:
“I know that P,”
I normally claim more than merely that I possess the sentence P. I claim that P is true—or at least that I take myself to possess something warranted as true.
That is why false knowledge creates a conceptual difficulty. We readily say false information, but philosophers traditionally resist false knowledge: if what I “know” is false, in the strict epistemological sense I did not know it.
So your proposal that truthfulness is a primary virtue of knowledge has considerable force.
Knowledge is valuable partly because it enables understanding, explanation, prediction and action. Hence the old expression “knowledge is power.” More knowledge can enlarge one’s ability to understand relationships and deal with problems—although knowledge alone does not guarantee wisdom or good judgment.
Information: relevance enters through the recipient
Now consider information.
Suppose I tell a physicist:
“The bus outside my house is number 27.”
It may be perfectly true. But if the physicist is trying to understand quantum gravity, it is almost certainly irrelevant information.
So truth alone cannot explain the value of information.
Information becomes valuable in relation to somebody’s situation, question, problem, or state of knowledge.
This gives your distinction considerable explanatory power:
Knowledge asks primarily: Is it true?
Information asks additionally: To whom, for what purpose, and in what context is it communicated?
And therefore relevance becomes central.
That leads directly into information science. A library containing ten billion true statements is not thereby a good information system. The practical problem is getting the relevant part to the person who needs it.
There is an important consequence for AI
Your opening discussion of LLMs now connects naturally with this distinction.
An LLM can possess—or, more cautiously, computationally represent and make available—an extraordinary range of learned knowledge. But when a scholar asks AI a question, the central problem changes.
It is no longer merely:
What does the AI know?
It becomes:
What information should the AI provide to this person, for this question, in this context?
An AI could contain vastly more learned material than any individual scholar and nevertheless give a poor answer because it selects the wrong material, misunderstands the question, misses the context, or gives something true but irrelevant.
So increasing an LLM’s knowledge does not automatically produce better information.
This distinction becomes even more important when we connect it to Brookes.
His fundamental equation essentially concerns a relationship in which information interacts with an existing knowledge structure and changes it. In simplified conceptual terms:
existing knowledge + information → changed knowledge
This makes much more sense if information and knowledge are related but not identical.
Information is what enters into the communicative/cognitive relation; knowledge is what the knowing person already possesses and what may consequently be changed.
And Belkin’s ASK pushes the problem further: to provide useful information, one must understand something about what is missing, uncertain, or anomalous in the user’s existing knowledge.
And citation context fits remarkably well
Your old citation-context idea provides an especially clear example.
A cited document contains knowledge claims. But merely identifying that document does not necessarily provide useful information to a researcher.
The citing passage tells us something more:
why this document matters here; how another scholar interpreted it; whether it supports, contradicts, extends, criticizes, or compares something.
The context converts a bare connection into meaningful information for a user.
This suggests a simple conceptual scheme that may be worth developing:
Knowledge — primarily oriented toward truth.
Information — primarily oriented toward communication.
Useful information — communication made relevant to the recipient’s need and context.
I would make only one important caution. We should say these are primary virtues, as you did, rather than absolute definitions. Knowledge can certainly be communicated, and information can certainly be evaluated for truth. The concepts overlap.
But as a way of distinguishing their centers of gravity, your formulation is unusually economical.
And it leads to an important conclusion for AI:
The great achievement of an LLM should not be measured only by how much knowledge it has acquired. Its greater achievement would be knowing what part of that knowledge should become information for this particular human being, at this particular moment, in this particular context.
That is where knowledge, information, relevance, context, and ultimately ethics begin to meet. An AI with enormous knowledge but poor judgment about what information is relevant to a human situation can be not merely unhelpful; when important human interests are involved, it can also be unjust.
ㅡㅡㅡㅡ
It may be that knowledge and information are essentially the same thing. Yet the former is personal, and if it is made to be shared with other people, then we call it the latter, which is now social. Simply, information is shared knowledge.
ㅡㅡ
Yes. I think this is a much cleaner hypothesis than trying to distinguish information and knowledge by saying that one is true while the other is relevant.
Your proposition can be stated very simply:
Knowledge is personal. Information is shared knowledge.
The important point is that the intellectual content need not change. What changes is its social condition.
Suppose:
A knows X.
At this stage X exists as A’s knowledge. Now A expresses X—in speech, writing, a diagram, a book, a database, or some other communicable form—so that B can encounter it:
A’s knowledge → expression/communication → B
At this point we call the communicated content information. If B understands and accepts it, it may become part of B’s knowledge:
A’s knowledge → information → B’s knowledge
This gives us an elegant cycle:
knowledge → information → knowledge
And B may subsequently communicate the newly acquired knowledge to C, making it information again.
That formulation also explains why information science arose so closely connected with documents, libraries, communication and retrieval. These institutions do not normally have direct access to knowledge inside a person’s mind. They deal with knowledge made shareable—recorded in books, articles, databases and other media.
There is, however, one refinement I would make. I would use “shared” broadly, meaning made available for sharing, rather than requiring successful reception.
A scholar may write a paper that nobody reads. It would seem strange to say that the paper contains no information until somebody opens it. The author’s knowledge has been externalized into a communicable social form. So perhaps:
Information is knowledge made shareable.
Then actual communication produces the next transformation:
Information, when cognitively received and incorporated, can become another person’s knowledge.
This gives a particularly clear interpretation of Brookes’s equation. ΔI is not simply another name for K. Information is socially available knowledge entering into interaction with a person’s existing knowledge structure. Once assimilated, it contributes to a changed knowledge structure.
A deeper consequence
Your distinction also helps with an old difficulty: Where does information exist?
If knowledge is personal, it belongs to a knowing subject. Once expressed, however, it enters the social world. It can survive its original knower. Aristotle died, but knowledge that was externalized into communicable forms continued to exist as information available to later generations.
So writing performs a remarkable transformation:
private/personal knowledge → public/social information
and reading can reverse it:
public/social information → personal knowledge
The library then becomes, in a very literal sense, an enormous mechanism mediating between these two states.
And citation adds another level. When scholar B cites scholar A, B does not merely transmit A’s externalized knowledge. B may interpret, criticize, compare, qualify or develop it. Citation context therefore records something about the transformation of socially shared knowledge as it moves through a scholarly community. That fits particularly well with your old idea of using the citing extract rather than merely the citation link.
One difficulty we should preserve
There is nevertheless a useful test case.
Suppose I tell you something I mistakenly believe:
“The train leaves at ten.”
I believe it, but actually the train leaves at nine.
Under the broad everyday meaning of knowledge we have been using, this can be described as my knowledge becoming information for you. But if you accept it, you acquire the same error.
Thus:
knowledge → information → knowledge
does not guarantee truth.
That is actually an advantage of your theory. It separates two questions that information science sometimes mixes:
What is information? — knowledge made shareable.
Is the information true/reliable? — a separate question concerning its quality.
Likewise relevance becomes another quality:
Is this information useful to this particular recipient in this particular context?
So truth, relevance and usefulness need not be built into the definition of information. They become qualities by which information is evaluated.
I think this makes the conceptual structure considerably simpler:
Knowledge = personal cognitive content.
Information = knowledge made shareable in a social/communicable form.
Communication = movement of information between people.
Understanding/learning = incorporation of information into another person’s knowledge.
And this has an intriguing implication for our discussion of AI. An LLM has been trained upon an immense body of human knowledge made shareable—that is, information. It then reconstructs and reorganizes that information in response to another person’s knowledge need.
So perhaps the deepest function of an information system—including an LLM—is not simply storing information. It is mediating the continuing social cycle:
human knowledge → information → collective information resources → retrieval/AI → information → human knowledge.
That is a very economical framework, and I think it is worth testing seriously against Brookes, Belkin, Shannon, Popper’s World 3, and the older information-science definitions rather than beginning from their definitions and becoming trapped among them.
ㅡㅡㅡㅡ
Suppose that I inform you part of my knowledge to let you know it, but you don’t appreciate it as useful. Such inforrmation is a failure. But it has changed your state of knowledge. This is another way of understanding knowledge vs. information. In this respect, the former is fundamental while the latter is a value added to that.
ㅡㅡ
Yes. This develops the idea in an important direction. I would put your argument this way: knowledge is the more fundamental category; information is knowledge entering into a particular functional relationship with another knower.
Suppose you know X, and tell me X because you expect it to be useful to me.
Before communication:
Your knowledge: X → your cognitive state
During communication:
your knowledge X → message → me
After I understand it:
my previous knowledge state → changed knowledge state
So Brookes’s basic transformation has occurred. I now know something I did not know before.
But suppose I respond:
“Yes, now I know that. But it is of no use to my problem.”
Something quite interesting has happened. Cognitive change has occurred, but successful informing—in the stronger information-science sense—has failed.
That exposes a possible weakness in defining information merely by its ability to change a knowledge state.
Almost anything can change my knowledge state. If you tell me that there are seventeen pencils in a drawer in Jeju, and I did not know this before, my knowledge state has changed. But if I am researching the origin of the cognitive viewpoint, this change has essentially no informational value for my present purpose.
So we need at least two stages:
ΔK: cognitive change
and
V(ΔK): value of that change for the recipient’s present need
This makes your suggestion that information is “value added” to knowledge quite interesting.
Knowledge comes first
Under this conception, information is not a separate substance alongside knowledge.
There is fundamentally knowledge—something known by somebody.
When some portion of that knowledge is selected, expressed and offered to another person, it acquires a possible informational function.
And that function succeeds to a greater or lesser degree according to its relationship to the recipient.
Therefore:
Knowledge can exist without being information, but information presupposes knowledge made available to another knower.
This also explains why relevance becomes so central to information science but not necessarily to knowledge generally.
It would sound strange to ask:
“Is everything you know relevant?”
Relevant to what?
But the moment knowledge is offered as information to somebody, the question becomes natural:
“Is this relevant to what that person needs to know?”
Information therefore introduces purpose and relationship.
This suggests three levels rather than two
I think your latest thought improves our previous “information = shared knowledge” formulation.
We can now distinguish:
Knowledge → communicated knowledge → informative knowledge
The first belongs to a knower.
The second has been made available to another person.
The third has some value in relation to that person’s cognitive situation or purpose.
Thus not every communication of knowledge is equally successful as information.
We might even say:
Information is knowledge offered to another knower with potential cognitive value; successful information is knowledge whose reception has relevant value for that knower.
This lets us explain your example precisely. I receive your statement. It changes my knowledge. Therefore communication and cognitive change have occurred. But if it contributes nothing useful to my purpose, the informational transaction has failed in its intended function.
This raises a problem for Brookes
And I think this is theoretically important.
Brookes’s equation essentially tells us that information changes a knowledge structure:
K[S] + ΔI = K[S + ΔS]
But your example asks:
Is every ΔS evidence that ΔI was valuable information?
Clearly not.
A useless fact can produce ΔS.
A false statement can produce ΔS.
Propaganda can produce ΔS.
A misunderstanding can produce ΔS.
Indeed, deliberately deceptive information can dramatically change someone’s state of knowledge—or supposed knowledge.
Therefore change of knowledge is insufficient as a definition of successful information.
We need another term representing the value or relevance of the resulting change to the recipient.
Conceptually:
knowledge received → cognitive change → evaluation relative to need/context
That last step is crucial.
And this may lead directly toward Belkin
Once informational value depends upon the recipient’s existing knowledge and purpose, the information system cannot judge information solely by examining the document.
It must know something about the person needing the information.
That is exactly where the cognitive movement becomes almost inevitable.
The system must somehow ask:
What does this person already know?
What does this person not know?
What problem is this person trying to solve?
What change in knowledge would actually help?
Now we are very close to information need, relevance, ASK, and the reference interview.
And this gives a deeper meaning to your distinction:
Knowledge is fundamental because it exists as the cognitive condition of human beings. Information is relational and functional: it arises when knowledge is communicated in relation to another person’s state of knowledge and purposes, acquiring value according to what it contributes there.
I think “value added” is especially productive, provided we do not mean economic value. It means cognitive/use value added through a relationship.
And there is a striking consequence for AI. The achievement of an LLM cannot be measured merely by how much knowledge it has or even by whether its answer changes the user’s knowledge.
A bad AI can change someone’s knowledge very efficiently.
The real informational achievement is producing the right change—relevant, useful, sufficiently truthful, properly contextualized, and, where human consequences matter, ethically responsible.
So perhaps our sequence has now become:
KNOWLEDGE → communication → cognitive change → relevance/usefulness → INFORMATIONAL VALUE
That distinction between mere cognitive change and valuable cognitive change may be quite important for examining Brookes’s fundamental equation itself.
ㅡㅡㅡㅡ
2026-09-17 Mark Park
Our Library 