On Mon, Mar 14, 2022 at 4:47 PM Ben Goertzel <[email protected]> wrote:
> Whether and in what sense semantic primitives can be found depends > wholly on the definitions involved right? > > Crudely, define ps(p,e) as the number of primitives that is needed to > generate p% of human concepts within error e > That's one way you could define the problem. But what assumptions are you baking into that? Are you assuming that the set of human concepts is finite? Are you assuming that human concepts do not contradict? Are you assuming it is meaningful to measure an "error" of concept e? Are you assuming you can even count concepts? Doesn't counting incorporate an assumption of equivalence, and error? Is it really useful to make such abstract assertions about meanings, and then work backwards to decide what they have to be to fulfil those assertions? Let's look instead at definitions of meaning which have proven useful. Then we don't have to make abstract assertions about them. For a mathematical sense of meaning we have Goedel's proof that any sufficiently powerful system will be incomplete. That's maths. My history comes from a linguistic sense of meaning. In the linguistic domain I find evidence that primitives derived for linguistic meaning, highly successful ones, constructing phonemes etc, contradict. So that's two senses of "meaning", actual useful senses, where I say there is evidence that semantic primitives do not exist: mathematical, and (structural) linguistic. I think Coecke is broadly speaking coming at linguistic meaning from a Category Theoretic sense. So that might be seen as a marriage of the two. Though I personally don't think the mathematical perspective is the most useful one. The linguistic perspective is bottom up. And I think far more suggestive of practical solutions. I don't know what other evidence you want that semantic primitives can't be found. I would argue that, given eyes to see it, the insight that semantic primitives don't exist has already been the single largest assertion of philosophy, for centuries now. Stephen Hicks traces it to Kant: "In the history of philosophy, Kant marks a fundamental shift from objectivity as the standard to subjectivity as the standard." https://www.stephenhicks.org/2010/01/19/why-kant-is-the-turning-point-ep/ Though philosophers are kind of lost. They are full of it, Heidegger, Wittgenstein, Hegel(?)... Basically all philosophers really. Russell came across it while trying to find primitives for maths... But philosophers, because they are trapped within thinking itself, can't actually find a solution to this problem. So they tie themselves in knots (and society, while they're about it...) We in AI have the advantage. We can express our theories in physicality. We can show how meaning can emerge, even if it contradicts. We can offer them a solution. But we have to accept the problem before we can offer the solution! Actually the "problem" is a solution! It's a feature not a bug. It actually explains a lot. It gives us an explanation for things like consciousness (expansion: larger than itself), freewill (larger than what it starts with), creativity (getting larger again.) If we wanted semantic primitives to exist we'd have to give up all these things! We'd have to still be puzzled by consciousness, and freewill, and creativity... Oh wait. We are... It's not even that hard to find this expanding structure. They're actually falling out of the woodwork all around us. We get too much structure. GPT-3 1.75B parameters. And it contradicts. WTF?!! Potentially all we need to do is accept this profusion of contradictory structure our "learning" procedures give us is the expansion that it seems to be. We can go on finding meaningful structure the same way. Only not be puzzled it doesn't reduce to primitives anymore, which it never did! But, by all means. Barge on. Seek semantic primitives. Start with a mathematical assertion and work backwards to the way the world ought to be. Who needs a solution for freewill, consciousness, creativity, one-shot learning... If semantic primitives suit your mathematical conception of how the problem ought to be formulated, then I'm sure that's the way the world should be, even if it's not! Just quietly in the wings here, amused to see the topic of semantic primitives come up again. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T0f3dcf7070b3a18e-Md6f20b5b9dfba408d4d0e91c Delivery options: https://agi.topicbox.com/groups/agi/subscription
