Interesting summation, Allen. Thanks. I subscribe to a couple of interesting critics of AI who point out often that the tech is way overhyped. When it comes to writing code, for example, sure, AI can write code, but it makes errors, and then a coder has to go in and find the errors in code he never wrote, which leads to less efficiency rather than more. There will no doubt be a period of adjustment as companies figure out what works and what doesn't. I think we'll need coders for the foreseeable future. What AI is good for in terms of coding is quick modelling by non-professionals to sell new apps and give coders a strong sense of what a client is asking for. There are plenty of other examples of AI being useful, but not quite as useful or reliable as advertised. I would also challenge the notion that robots don't get tired. There's a lot of sleight of hand where AI is concerned. For instance, it mimics intelligence without understanding, and many are fooled into believing that AI is self-reflexive or has an idea what the world is.
Thank you, as always, for writing, Asa. I anticipated that saying anything about AI was a risk, since I see articles every day about this or that side of it, its being over- or under-valued, and so on. I think your conclusion is right. AI knows what AI knows, meaning that it is, as you say, self-reflexive and able to reproduce only what it has produced. This is easily forgotten, since it is so, well, big. I am always amused when AI writes answers to m email for me. It never sounds remotely like me (whom is it imitating? not me) and often reverses or fragments the point I want to make.
Funny. We're using "self-reflexive" differently. I'm saying it's not self-aware, and doesn't actually know what it's doing. It does not, for instance, understand semantics, i.e. the meaning or import of words. Instead, what it does is mimic, like a mockingbird or a parrot--though admittedly with more sophistication. I broke Claude by subjecting it to what I call, the "Eunoia Test." Eunoia was a godawful book of "poetry" written by isolating words with only one vowel to the exclusion of all other vowels: so, let's say, words that only use "a" as a vowel, where no other vowels are present. One of the more obvious, sophisticated words would be "catamaran." Claude was able (but with difficulty) to make a list of such words. Then I asked it to write 5 short poems (that did not have to rhyme) using those words only, but they had to make sense. It couldn't perform the task mainly because it doesn't know how to "make sense." My guess is that (a) it couldn't handle relating to words as both data and semantic units; and (b) it doesn't deal in semantic units at all since it operates using a token system--which I recommend you look into. In short, by asking it to formulate sensible phrases from a list of words, I exposed the smoke and mirrors. The bottom line is that AI comes up against the limits of our own understanding: it doesn't understand gravity (because we don't); and we still haven't cracked the "hard problem"--how it is that we have an experience of the world. There's no consciousness there and there is no route by which it can achieve such a thing.
Fascinating read Allen. So much we don't know about how all of this is going to play out. Quite a bit of this was new to me, so thanks for that. Waymo brought a chuckle.
Interesting summation, Allen. Thanks. I subscribe to a couple of interesting critics of AI who point out often that the tech is way overhyped. When it comes to writing code, for example, sure, AI can write code, but it makes errors, and then a coder has to go in and find the errors in code he never wrote, which leads to less efficiency rather than more. There will no doubt be a period of adjustment as companies figure out what works and what doesn't. I think we'll need coders for the foreseeable future. What AI is good for in terms of coding is quick modelling by non-professionals to sell new apps and give coders a strong sense of what a client is asking for. There are plenty of other examples of AI being useful, but not quite as useful or reliable as advertised. I would also challenge the notion that robots don't get tired. There's a lot of sleight of hand where AI is concerned. For instance, it mimics intelligence without understanding, and many are fooled into believing that AI is self-reflexive or has an idea what the world is.
Thank you, as always, for writing, Asa. I anticipated that saying anything about AI was a risk, since I see articles every day about this or that side of it, its being over- or under-valued, and so on. I think your conclusion is right. AI knows what AI knows, meaning that it is, as you say, self-reflexive and able to reproduce only what it has produced. This is easily forgotten, since it is so, well, big. I am always amused when AI writes answers to m email for me. It never sounds remotely like me (whom is it imitating? not me) and often reverses or fragments the point I want to make.
Funny. We're using "self-reflexive" differently. I'm saying it's not self-aware, and doesn't actually know what it's doing. It does not, for instance, understand semantics, i.e. the meaning or import of words. Instead, what it does is mimic, like a mockingbird or a parrot--though admittedly with more sophistication. I broke Claude by subjecting it to what I call, the "Eunoia Test." Eunoia was a godawful book of "poetry" written by isolating words with only one vowel to the exclusion of all other vowels: so, let's say, words that only use "a" as a vowel, where no other vowels are present. One of the more obvious, sophisticated words would be "catamaran." Claude was able (but with difficulty) to make a list of such words. Then I asked it to write 5 short poems (that did not have to rhyme) using those words only, but they had to make sense. It couldn't perform the task mainly because it doesn't know how to "make sense." My guess is that (a) it couldn't handle relating to words as both data and semantic units; and (b) it doesn't deal in semantic units at all since it operates using a token system--which I recommend you look into. In short, by asking it to formulate sensible phrases from a list of words, I exposed the smoke and mirrors. The bottom line is that AI comes up against the limits of our own understanding: it doesn't understand gravity (because we don't); and we still haven't cracked the "hard problem"--how it is that we have an experience of the world. There's no consciousness there and there is no route by which it can achieve such a thing.
Fascinating read Allen. So much we don't know about how all of this is going to play out. Quite a bit of this was new to me, so thanks for that. Waymo brought a chuckle.