6

Humans know trees. So do squirrels. Even though their mental concept of "tree" may be very different, I think we can agree that lots of animals recognise trees and have some sort of understanding of them.

A computer program, on the other hand, might have access to a lot of facts about trees, but it doesn't seem to "know" trees. Similarly, I might have a book about Australian animals in my pocket, without me actually knowing what a wombat is.

What would it take for us to say that a simpler animal, or some form of artificial life, "knows" something?

I'm looking for references to any discussions on this question, or suggestions for keywords I can use to search the literature. I don't think the question necessarily involves qualia. And I've seen discussions that deal with questions like "can I be said to 'know' a fact, if it turns out that the fact is not true?", but that's not quite what I'm looking for either.

I realise this question is a bit vague. My background isn't philosophy, so I'm not sure what terminology to use. Any help in improving the question would be appreciated.

  • Knowledge is empirical verification of what is (else how do you know what is?) and this whether human, newt or otherwise doing the empirical verifying. If someone knocks at my door and my dog barks, once the door is open both my dog and I know if there is a friend or stranger at the door. – Mr. Kennedy Dec 15 '16 at 22:51
  • @Mr. Kennedy- Yes, but how do you know there is someone at the door? How do you even know you have a door? How do you know anything? For Russell this was the most important question in philosophy and I'd probably agree. For the Perennial philosophy 'knowing' is the ontological primitive. By reduction it would be the foundation of your being and as such cannot be explained by reference to anything else. – PeterJ Sep 9 '17 at 12:20
3

Immanuel Kant made an important distinction between analytical and synthetical propositions which I believe is very helpful in understanding the basic structure of knowledge. He described this distinction as follows.

"Either the predicate B belongs to the subject A, as somewhat which is contained (though covertly) in the conception A; or the predicate B lies completely out of the conception A, although it stands in connection with it. In the first instance, I term the judgement analytical, in the second, synthetical." (Critique of Pure Reason, A6/B10)

One important aspect of this distinction is that analytic propositions do not provide knowledge because they do not contribute anything to what is thought to be already contained in the subject. Therefore, Kant's analysis provides a good working definition of knowledge by indicating its basic components in terms of propositions along with the assertion that there must be a synthesis between those components.

However, this is only a start because the idea of synthesis must be grounded in more than just abstract terms. Blindsight patients illustrate the fact that there is an important distinction to be made between registering data and the conscious knowledge of the information represented by that data. Due to either a lesion or the surgical removal of the striate cortex, blindsight patients are able to register information in their minds without any visual experience accompanying the information being received. During the course of experimental tests, they are asked to indicate certain aspects of visual targets such as their location or orientation, but since they are unable to see these targets, they have to rely on their best guess. Even so, test results show that they are able to properly indicate those aspects much more accurately than they could by chance alone, and they are usually amazed to discover how well they were able to access information that they had no conscious knowledge of possessing.

These results indicate that there must be more than just mere existential or causal relations between the components of knowledge, i.e. those components which might be thought to be analogous to the subject and predicate of a proposition. An existential or a causal connection is insufficient to realize the sort of synthesis that is necessary for knowledge. Jean-Paul Sartre recognised that this capacity to provide synthetic unity is essential to the nature of consciousness:

"A consciousness is a synthesis through and through, thoroughly intimate with itself: it is at the heart of this synthetic interiority that it can join, by an act of retention or protention, with a preceding or succeeding consciousness." (The Imaginary, p. 25)

Due to the inadequacy of existential and causal relations, it is apparent that there is something distinct about the type of synthesis which consciousness is able to accomplish, and it is only by means of this type of synthesis that knowledge is possible.

  • 1
    The connection to blindsight patients was especially helpful. Goes right to the core of the issue. I'm off to slog my way through Kant. ;^) – mhwombat Apr 9 '16 at 0:43
3

As I understand your question you do not ask for answers from epistemology. Instead you are interested in questions like:

What does it mean for an intelligent system to know its environment?

Since some time there are autonomous intelligent systems cruising on the Moon or on the planet Mars. These robots are autonomous in the sense of finding their way without explicit control from engineers on earth. The robots have sensors to observe their environment and actors to move around. Most of all, they are capable to design a model of their environment and to locate their current position within this model.

Hence the robot knows at each time where it is.

Possibly one can define this kind of knowledge as the capability to match the internal model with the external feedback and to correct the model if necessary. This kind of knowledge arises from learning.

Some keywords: Model, cognitive map, autonomous robot, unsupervised learning, artificial intelligence, artificial life.

2

I would agree that, as is often the case, Kant is the best starting point for a modern philosophical idea of "knowledge," as opposed to existence or truth, for example. Unfortunately, getting to "know" Kant is extremely time consuming.

I believe it was Hobbes (I may be wrong) who said that to "know" something is to be able to reproduce it. This is a very handy definition, and far more complex than it at first appears. The irreducible philosophical command to "know thyself" is utterly paradoxical for humans because we cannot reproduce ourselves without "knowing" another, in the biblical sense. Similarly, I have often argued that one test of AI should be its capacity to reproduce itself...as biological intelligence does. A far more difficult demand than it at first appears. Von Newman replicators illustrate the problem.

One can bring Kant back in here, in a renegade interpretation. Kant could be said to argue that we indeed "know" the world precisely because we partially produce and reproduce it. It is because we synthesize or "con-struct" experience of an outer world according to categorical "in-structions" that we can "know" it. This is how concepts latch onto percepts.

Let me toss in one more reference, this time from Shannon information theory. To know is to increase information, which is to eliminate uncertainty. One begins with uncertainty (or possibilities) between zero (0) and one (1) about (X) and eliminate this uncertainty by a process of statistical divisions, as in the game of 21 questions. To know is to eliminate relative uncertainty given certain assumptions. All forms of life must do this in some fashion and to varying degrees.

Again, the idea that all knowledge begins in the Delphic maxim "know thyself" is, in my view, crucial to philosophy. It was Socrates who demonstrated the paradox of this command, that we can only know for certain "what we don't know." So the Socratic, dialectical motion towards knowledge is not unlike the endless "elimination of uncertainty" given a modern, mathematical form in information theory.

  • 1
    Being able to reproduce/reconstruct something as a requirement for "knowing" is an interesting idea, and one that is new to me. And it is easier to apply that to AI than something more rarefied, like qualia. Thank you. – mhwombat Apr 9 '16 at 1:26
2

From the point of view of the stream of philosophers focussed on a natural view of human interactions, flowing from Stoicism through to Schopenhauer and Nietzsche, knowledge is interpretation (or memory) that produces power. To me this is the most useful form of the definition.

One framing from Nietzsche (characteristically trying to say too many things at once, but one of them being this notion):

In so far as the word "knowledge" has any meaning, the world is knowable; but it is interpretable otherwise, it has no meaning behind it, but countless meanings.—"Perspectivism." It is our needs that interpret the world; our drives and their For and Against. Every drive is a kind of lust to rule; each one has its perspective that it would like to compel all the other drives to accept as a norm.

So knowledge is made up of perspectives, each of which answers a different aspect of will and gives us a different sort of power.

This sort of view is consonant with Shannon's information theory. Information, in that context, is detectable order that can be used to predict or control a system. Knowledge, in this sense, then, is information when the prediction or control afforded addresses an actual or potential wish or need.

Basing 'knowing' on Shannons's definition of 'information' also agrees with more modern notions of science that are based upon making predictions as the primary test for knowing.

Another aspect here is that knowledge answers to a real need, or, at least, a drive evolved to fulfill a need, absent from more abstract notions of knowledge as a faculty such as the "justified true belief" definition or definitions based upon specific kinds of acts.

This captures the Montessorian notion of horme or the Lacanian notion of jouissance. Knowing is what horme reaches toward and what embeds and preserves joissance for later use. Learning and knowing are self-integrating processes that explain the mind's engagement in the world as a natural aspect of biology.

For your question, then, the question from my point of view is how is intention embedded and transferred? Do machines have needs of their own? Do the needs we build into them continue the biological process, or is there an essential disconnection between derived needs and designed ones?

For that, I would look at genes. To my mind, from this point of view, genes know things, and they think: They evaluate and manipulate information with an intention to survive in future creatures. (They do so through the insanely indirect mechanism of creating numerous whole complex creatures. But that is beside the point. They do it.)

Therefore, much of the intention that backs our own knowledge is borrowed from genes, and the idea that we pass it along to machines is not a metaphor, it is continuous with the process that lent it to us.

  • Wow. A lot of interesting ideas there. And now that I think about it, genes do seem to know things, in a way that, say, a telephone book does not. – mhwombat Apr 9 '16 at 1:33
0

You asked for references. Read everything you can get your hands on by Robert Brandom and Huw Price. They're relatively unknown, but they are Major Figures in contemporary philosophy and they address your question directly. They have a bunch of stuff on their homepages.

The classic, Cartesian, representationalist view is that to know something is to know that; the pragmatist view is that to know something is to know how. Knowing that is usually construed as having some kind of accurate or true representation of the world. There are lots of problems with this; it failed miserably in AI.

Brandom's account of knowing how is fascinating, he always asks just what you ask: what is it for anything to know, including not only humans, but non-human animals, martians, machines, whatever. But instead of trying to explain knowledge in terms of representations, he asks what it would take for us to decide that something knows. And since all we would have to go on is observed behavior, the question becomes: what must a creature be able to do in order to count as knowing something? His argument is complex and subtle, but it amounts to: it must be able to make inferences that are deemed correct by a discursive community.

He gives a wonderfully simple example. A hunk of iron can respond differentially to its environment: it rusts in the presence of moisture. But obviously we would not treat it as "knowing that it is wet here." A parrot can be trained to squawk "that's red" in the presence of red things, but again we would not treat it as knowing the thing is red. The difference between a trained parrot and a human who can do the same thing is that the human is capable of doing something the parrot cannot do, namely draw correct inferences like "therefore it is colored" and "therefore it is not green" and so forth. The collection of such normative inferences together institute the concept "red", and the norms that determine correctness are set jointly by the community. So knowledge is pragmatic (based on what we do), inferential, and social. In other words, rational.

That leaves the question of how we managed to make the transition from the sentience of non-human animals to the sapience of humans. Brandom does not address this because its a separate issue; his task is to articulate the conceptual structure that must characterize (and be instituted by) any set of practices in order for them to count as "rational" (and thus involve what counts as knowledge).

Brandom's "Articulating Reasons" is a good, brief overview of his thought. See also his essay "How Analytic Philosophy Has Failed Cognitive Science" for his account of how what philosophers have learned about conceptual structure can and should inform scientific investigations of cognition etc.

  • 1
    While possibly true, you should say more to make this a better answer. Right now, it's a name drop answer (roughly on the level of a link only answer). – virmaior Apr 3 '16 at 23:52
  • In other communities where I'm in (mostly mathematics) this would be written as a comment : not answering but helpful to gather more background. – Gottfried Helms Apr 4 '16 at 6:33
  • @virmaior: Sorry, newbie here. I added some detail. – user20153 Apr 7 '16 at 18:18
  • It's helpful to learn about some contemporary philosophers. When I looked up Brandom and Price, I found that they have a lot of publications, so it was hard to know where to begin. So thank you for adding a suggested starting point ("Articulating Reasons"). – mhwombat Apr 9 '16 at 1:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.