Episode 131: Knowledge as a Concept
- Links to this episode: Spotify / Apple Podcasts
- This transcript was generated with AI using PodcastTranscriptor.
- Unofficial AI-generated transcripts. These may contain mistakes. Please check against the actual podcast.
- Speakers are denoted as color names.
Transcript
[00:00:00] Blue: Hello out there. This week on the Theory of Anything podcast, Bruce continues his exploration of the concept of concepts by looking at knowledge as a concept. And of course, along the way, pushes back against several critical rationalist dogmas. As always, I enjoyed listening to Bruce here, and I hope you also get something out of our shortest episode ever.
[00:00:33] Red: Welcome to the Theory of Anything podcast. Hey, Peter.
[00:00:36] Blue: Hello, Bruce. How are we doing today?
[00:00:38] Red: Doing good. Okay. Today’s going to be a short episode. Painfully short. Never say that.
[00:00:46] Blue: Painlessly short.
[00:00:48] Red: Yeah, because then you’re going to ask questions and we’re going to go on for three hours.
[00:00:53] Blue: We can’t predict that. We can’t predict.
[00:00:55] Red: I just know that my script is very short. That’s really what I mean.
[00:00:59] Blue: The PowerPoint is short. Okay. Yeah,
[00:01:01] Red: that’s right.
[00:01:02] Blue: Okay.
[00:01:03] Red: So we’re going to talk today. So what we’ve been talking about up to this point is in previous podcasts, we talked first about Popper’s critique of concepts and how he found theories to be a hundred times more valuable than concepts. But then I critiqued Popper’s critique and found room for some strong disagreement. Concepts are quite a bit more valuable than Popper thought if you consider the conjecture side of his epistemology. I also argued that there is no clear demarcation between concepts and theories, and that a concept is sometimes treated as a theory as if it can be falsified, and a theory is sometimes treated as a concept, i.e. flexible and extensible so that by definition it can’t be falsified. For example, the idea that sound is a wave like water is a wave was one that we used in a past podcast. If treated as a theory, it will prove false because sound waves are compression waves and water waves are transverse waves, so the analogy is wrong, and if you try to treat it as a falsifiable theory, it will prove to be a false theory. But if treated like a concept, this idea will start a move towards a very useful but abstract concept of wave that ultimately proves so fruitful that you almost have to have that concept to make certain kinds of progress in physics. It represents a conceptual shelling point that leads to fruitful conjectures. However, I agree with Popper that a concept like wave, when being treated as a concept, which is generally how we treat it, that it is not falsifiable. Indeed, concepts almost by definition have an amazing level of flexibility that make them, by definition, unfalsifiable.
[00:02:45] Red: But this unfalsifiable flexibility is exactly what makes them so valuable to the conjecture process. Because of this, I agree with Popper that theories which are falsifiable by definition are 100 times more valuable than concepts if we are only talking about the critical side of his epistemology. But if we take concepts versus theories in general, I can’t agree that concepts are 100 times less valuable than theories. Now, in the last episode, we talked about my criticisms of the pseudo -Deutch theory of knowledge and the related two sources hypothesis. In today’s episode, we’re going to pull these ideas together and show how they relate, okay? So to understand why the pseudo -Deutch theory of knowledge relates to our discussion of concepts and theories, let’s start with a good summary of David Deutch’s constructor theory of knowledge. So this is often, his theory is often summarized as once knowledge is embodied in a suitable environment, it tends to cause itself to remain so. That’s taken from the beginning of Infinity page 78. Often, I’ve seen this theory stated in summary like this. It was actually, I think, a survey that came up with this summary. It says knowledge is information that has three properties. It is capable of enabling its own preservation. It can be copied from one embodiment to another without changing its properties. It can enable transformation and retain the ability to cause them again. In and of itself, I see nothing wrong with Deutch’s constructor theory of knowledge. And so
[00:04:16] Blue: that was the back up a sec. That was a summary of the constructor theory of knowledge. Is that what constructor theory is? No. Or is this just a subset? This is a subset, yeah. A subset of what? So it’s not like constructor theory goes into physics and everything. It’s just that this is just one slice of constructor theory. That’s
[00:04:42] Red: correct. It’s taking constructor theory and applying it to the idea of knowledge.
[00:04:47] Blue: Okay.
[00:04:48] Red: Okay.
[00:04:49] Blue: I
[00:04:49] Red: was just saying, I see nothing wrong with Deutch’s constructor theory of knowledge. However, I do take issue with the fact that it is widely, and I’d say even dogmatically, understood to be equivalent to the two sources hypothesis, meaning that only biological evolution and human ideas are forms of information that have these three properties. This is for sure a false idea that can be easily falsified via counter example. Now, I discussed this problem at length in episodes 75 to 80, if you’re interested in that. Okay. But I can give two easy emblematic examples of how to refute this idea. The first is the one we just discussed. David Deutch’s own genetic algorithm that allows a robot to walk, just discussed in the previous episode. Deutch is very insistent that the genetic algorithm creates no knowledge and that the human created all the knowledge. Yet a genetic programming algorithm without a doubt, matches the three principles we just claimed summarizes Deutch’s own constructor theory of knowledge. This is, that is the genetic algorithm creates small snippets of code that are, one, capable of enabling their own preservation by being useful even. Two, can be copied from one embodiment to another without changing its properties. Three, it can enable transformations in this case, the walking robot, and retain the ability to cause them again. You might get some pushback on enables transformation, is walking a transformation. And this is something I bring up in episodes 75 to 80 and kind of dig into. Here we face a certain vagueness in the constructor theory of knowledge. It is unclear what does or doesn’t count as a transformation. Okay. No problem.
[00:06:27] Red: Let’s pick an example that is far more, far more clear cut refutation of the pseudo -Deutch theory of knowledge. Let me just argue that this example does fit all three of those. Okay. But I’m going to give you an even more clear cut example now. Consider the immune system. David Deutch explicitly claims in a podcast episode that the immune system creates no knowledge. So the crit -rat community will deny that the immune system creates knowledge. This inference is usually justified utilizing the pseudo -Deutch theory of knowledge, which is why I had to do an episode about it first. Okay. For example, it is easy to point out that genes are knowledge and there was a lot of genetic knowledge that went into creating the immune system. So by counting all the knowledge that genes brought to the table already, we infer utilizing the same sort of Deutch theory of knowledge white swan counting method discussed in the previous episode, that really the immune system must create no knowledge at all. On top of that, the defenders of this idea will usually point to the immune system and say, look, it’s perspiration, not inspiration, as if biological evolution was inspiration instead of perspiration. That doesn’t make sense to me. And because perspiration, perspiration and scare quotes supposedly doesn’t create knowledge, they conclude the immune system must not create knowledge. See episode 79, perspiration versus inspiration for a full discussion about this. Yet, as discussed in episode 77, the immune system exactly, exactly fits those three properties of knowledge that supposedly are the constructor theory of knowledge. When the immune system detects a pathogen, it intentionally starts a mini evolutionary process where it hyper mutates the DNA of the cells, searching for a recipe for an antibody.
[00:08:13] Red: If the recipe discovered by a cell receives feedback that it worked, these cells are allowed to replicate and the failed cells aren’t allowed to replicate. Thus, slowly changing the population of pathogen fighting cells to be the ones with the effective recipe for the antibody. Note that the knowledge for the recipes for the antibody is stored in DNA, just not in the germ lines DNA. Now, based on this example, I ask you to answer truthfully if the immune system fits the constructor theory of knowledge. Is it capable of enabling its own preservation? Yes, the successful antibody recipe causes itself to stay instantiated while the failed recipes vanish. Can it be copied from one embodiment to another without changing its properties? Yes, of course it can. That’s exactly the whole point. We want those cells to then replicate and be copied from one form to another until they can fight the entire pathogen. Can it enable transformations and retain the ability to cause them again? Yes, of course it can. Now we even mean a literal transformation in the sense of a recipe that creates an antibody. Okay, so apparently even under Deutsch’s own theory, Deutsch is incorrect about this. The immune system does in fact create knowledge under the constructor theory of knowledge. We only need one counter example to refute or falsify a theory. So I’m done, right? What I found was that the crit rat community vehemently denied this. No crit rat that I’ve talked to agrees with me that the immune system under Deutsch’s own constructor theory of knowledge qualifies as knowledge, even though it clearly fits all three properties required.
[00:09:55] Red: What crit rats did instead when I asked them about this was interesting that on the fly start to tack on new properties or criteria required to be quote considered knowledge. So suddenly there might be a new requirement tacked on that knowledge must persist longer than the length of a single animal, perhaps centuries, to be considered knowledge. At that point I might ask them if a human idea that lasts only for the length of a single animal is not knowledge. And of course they vehemently instead claim no if a human created it, it’s knowledge even if it doesn’t last generations. I might then challenge, so in other words they’ll tack on to for biology the requirement that it lasts for centuries, but then when you point out human ideas don’t last for centuries, they’ll say well that criteria only counts for biology not for humans is what they basically do. Now I might challenge them on the fact that that quote life of an animal might be very short indeed. Some mayflies have a whole lifespan in an hour. The immune system surely creates knowledge that persists longer than that and of course they’ll then say no for animals it has to last for generations and millions of years to be considered knowledge. In other words they gerrymander their theory of knowledge until it’s only includes human ideas and biological evolution via DNA that’s in the germline. Nor does it matter to them that their theory is if their theory is self -consistent or not. They have no problem at all with giving entirely different requirements to animals versus humans versus everything else. What matters was that nothing but the two sources biological evolution and human ideas was allowed to count as knowledge.
[00:11:38] Red: Here’s what I find interesting about this they never seem to notice that they were on the fly adjusting their theory. They always very sincerely thought and insisted that this had been what their theory of knowledge had said all along. Now let me use an anonymous real life example of this okay to one crit rat friend I pointed out you know there is nothing in your three criteria that you wrote down because we had written it down together. We are visually doing this very via text everything was written down and that we’ve already agreed upon this these three criteria that says knowledge has to last longer than the life of an animal much less centuries. So if instantiated if instantiated in DNA but that the length requirement is unnecessary for human ideas so may I write that down as a fourth criteria of the property of knowledge since it goes beyond the three that we’ve already agreed upon okay and I got an angry response no you can’t write that down and how and then he would tell me how it was just obvious that he that what he was saying and he couldn’t believe I couldn’t see how obvious it was based on such an argument he’d insist that therefore he hadn’t changed or exceeded the original three criteria apparently the idea that animal DNA knowledge must exist for centuries or millennia but human ideas don’t need to was in his mind already contained implicitly within the original three criteria that we’ve been talking about.
[00:13:06] Red: I did not sense this was being said in bad faith by the way I sincerely think he really and truly believed the original three criteria implicitly included this idea though I couldn’t personally see how it is harder to explain why he got angry over me wanting to make it explicit by writing it down though I have a guess even here but let’s first be clear his argument that it was quote it was just obvious really does make no rational sense because it is true if it is truly obvious to him but not to me which is what apparently was the case why not just write it down why not I just have an explicit list of the full criteria and rather than I have to ask his opinion all the time okay moreover this conversation was in text he had literally just barely written it down to me I
[00:13:57] Red: was only asking him to copy what he said verbatim and put it into the list of required properties to count as knowledge I even offered to let him keep a menu the list however he desired so long as I got to hold him accountable for whatever criteria he chose what I found is this person would not let me write it down he simply doubled down that it was obvious and that he couldn’t believe I couldn’t see how obvious it was but I still never got the sense that there was any conscious bad faith going on he seemed quite sincere to me I think all of this was just so obvious to him aka the truth aka the truth is manifest error that he was convinced if I needed to write it down I must be playing some sort of trick and he wasn’t going to allow me to do that I think this was how he was explaining it to himself why he was refusing to write down an additional property that of what counted as knowledge okay given that explanation that assumes that assumes I’m playing some sort of trick it becomes more understandable why he was getting angry over what really did seem a perfectly reasonable request when I inevitably um point out to a crit rat defender of their theory that their theory is immunized for repute reputation they will without fail declare it a philosophical theory so they perceive it as not being a theory requiring any sort of testability
[00:15:27] Red: so the lack of falsifiability when I point out to them that it can’t be falsified because of what he’s doing that doesn’t really bother him any okay this is what I call the crit rat loophole by the way so hopefully you’re catching my overall frustration here okay now um so what is going on such that this sincere individual can’t see the problem with their own argument here is my proposed theory of crit rat thinking about knowledge I think what’s going on is that they’re treating quote knowledge as a concept rather than as a theory okay recall that concepts often take the form of fuzzy categories there is no specific set of essential criteria that define that category now does that sound familiar it’s exactly what is going on with the crit rat pseudo -deutsch theory of knowledge that we’ve discussed you can search long and hard long and as hard as you want to find a set of essential defining properties of what makes something beautiful and you’ll never find such a set because it doesn’t exist because the word beauty actually has multiple meanings um connected by analogies instead as discussed in episodes 85 and 86 we actually define a fuzzy category via canonical examples rather than via a set of properties that defines the category the example I used back in episodes 85 and 86 taken from pinker was how bird is defined for most people by an image of a sleek black small black bird what we might call a canonical bird we know an ostrich is technically a bird too but we see it as a very edge case of the category of bird we see it as a bad example of a bird in fact this is hardwired into how humans think they can they can and they can corroborate that this theory through experiment by the way if you do a reading comprehension test where one sentence refers to an ostrich um versus say a robin and the next sentence refers to it as a bird it will literally take you longer to comprehend the word bird in the second sentence uh that that the word bird in the second sentence was referring to the ostrich compared to the robin precisely because we don’t really emotionally consider an ostrich to be a quote true bird because it doesn’t match the canonical um conceptual image of a bird very well what i’m proposing is the critrack community’s concept of knowledge particularly the pseudo -deutch theory of knowledge consists of two parts a specific theory the constructor theory of knowledge and an overall concept based on two canonical examples human ideas and biological evolution
[00:18:03] Red: because it is the concept that matters to them not the theory the canonical examples are what matter here once you realize that critrats understand knowledge as a concept based on two canonical examples of biological evolution and human ideas rather than as a specific falsifiable theory defined by the three criteria that we mentioned um that were the summary of the constructor theory of knowledge a great deal of how critrats behave suddenly makes sense the immune system conceptually speaking clearly is less impressive than the two canonical examples yes the immune system fits perfectly the three properties of knowledge as does the walking robot but it just feels off compared to the canonical examples just as an ostrich just doesn’t feel like a true bird so when i’m having a conversation with critrats about knowledge um they are doing their best to impart to me the feeling intended by their concept of knowledge whereas i’m playing a different game namely i’m stress testing the theory itself to see if it can be falsified we’re simply playing two very different games okay we’re both right within those two games they’re anger with me over not seeing how obvious it is starts to make sense i’m sure to them since they can’t see i’m playing a different game than them i must seem rather obtuse that i can’t see the difference between the mundaneness of the kind of knowledge that the immune system creates compared to the rather cool examples they can canonically think of when it comes to the open -ended knowledge creation processes found in biological evolution and human ideas if i’m right in my conjecture then critrats primarily see knowledge as a concept which is meant to be unfalsifiable rather than as a theory which is meant to be tested and possibly falsified and in this case actually is falsified might this idea explain other critrat theories by reimagining them as unfalsifiable concepts rather than theories so here’s a question for you how many popular critrat theories are really less like theories and more like concepts where by concept i mean a vague and even flexible and extensible conceptual idea with no real content so there is no way to reasonably test or check the concept
[00:20:19] Red: that is my thought for this episode
[00:20:22] Blue: it’s been thought provoking as as always i wonder how many other podcasts there are out there that are just delving into like completely different strands of philosophy there this deeply i feel like there probably aren’t that many i mean maybe i just don’t know about them but i hope this says something about about how just thought provoking and far -reaching critical rationalism is maybe even our even our critics would would say so i mean i guess the alternative is that we’re just completely on the wrong path and we’re just crazy to think about this stuff so much i think probably i have friends and family who might agree with that assertion deep down so i don’t know
[00:21:14] Red: let me say something about this because you were talking about this from the philosophical point of view we’re really getting into the nuts and bolts of agi now i mean like it may not be obvious but that’s what we’re doing right
[00:21:26] Blue: yeah
[00:21:26] Red: we’re actually starting to give real thought to what’s a what’s a human idea what’s a concept and how does it differ from a theory how does it relate to conjecture how does it relate to um and i can even give you a kind of high level sketch at this point you’ve got something called concepts which are just human ideas whatever human ideas are i’m going to call them concepts so tautologically they’re concepts and i admit i don’t entirely know what they are um and you got these concepts and we know that they can be very flexible we know that they’re non falsifiable but we can turn them into theories we can we can kind of zoom out we can become fuzz out as i’ve as i’ve said it and we can think of them in a kind of conceptual way where they’re not falsifiable and then we can kind of dig in and we can make different conjectures about what the specifics are and then we can test those conjectures using criticism because now they’re very specific and they’re explicit and they’re falsifiable okay and i suspect that we’re starting to get a very rough picture of what it is that humans are doing which is directly related to the question of agi right this ability for us to fuzz out when we need to make conjectures and become explicit when we need to be critical and need to test our conjectures to see if they’re true or not i suspect i just gave one of the big keys to agi right there right
[00:22:46] Blue: okay so we’re not podcasters we’re agi researchers we are okay it was it was my interest in agi that has caused me to dig into this idea of concepts and what popper said about it and how it relates to his epistemology and things like that so okay not an accident
[00:23:04] Red: this has happened
[00:23:05] Blue: i love it um well thank you bruce you have a good day
[00:23:10] Red: all right you too
[00:23:19] Blue: hello again if you’ve made it this far please consider giving us a nice rating on whatever platform you use or even making a financial contribution through the link provided in the show notes as you probably know we are a podcast loosely tied together by the popper deutch theory of knowledge we believe david deutch’s four strands tie everything together so we discuss science knowledge computation politics art and especially the search for artificial general intelligence also please consider connecting with bruce on x at b neilson 01 also please consider joining the facebook group the many worlds of david deutch where bruce and i first started connecting thank you
Links to this episode: Spotify / Apple Podcasts
Generated with AI using PodcastTranscriptor. Unofficial AI-generated transcripts. These may contain mistakes; please verify against the actual podcast.