2023-04-23 04:04:18 crc: No problem, I just wanted to play with it on another computer and noticed 2023-04-23 04:06:06 AI seems inherently unforthy to me. We use a ton of CPU power and RAM to avoid thinking about how to solve problems ourselves 2023-04-23 04:06:43 But of course I know it solves some things we don't have algorithms for, like "go make me some nice concept art for my novel" or "carry on my thesis so I can tune and edit it later" 2023-04-23 04:08:26 Facial recognition is more on the 'you need it' end of the spectrum for tracking down criminals. For instance if I could use that to catch a murderer, personally I would want to 2023-04-23 04:09:21 So thinking about it I think there is a place for 'AI', and I'd be interested to see someone do some 'AI' stuff in Forth 2023-04-23 04:09:47 And how Forth might be able to improve on what others had done, because Forth has a tendancy to cut chaff away 2023-04-23 07:34:05 I think you're basically right about AI, but I think we could also note that what most AI algorithms are doing is processing an enormous amount of data. That's what we really use all that computing power for. But - you're right in the sense that we adopt a "brute force" analysis of the data instead of trying to discover the patterns that underlie it, which is how we might get a more efficient solution of 2023-04-23 07:34:06 the problem. 2023-04-23 07:34:45 The applications of AI that appeal more to me is where we're not using just to get an answer to a specific problem, but rather using it to try to help us identify and understand those patterns. 2023-04-23 07:34:57 That feels like kind of a "meta" usage to me. 2023-04-23 07:36:16 I'm interested in trying to do that with my drive performance data. I have all this raw performance data, and alongside it I have a ton of "internal drive metrics" that I measured. We understand how some of those metrics relate to performance, but I wonder if there might be connections we don't know about yet, and maybe I could use AI to find them. 2023-04-23 07:36:43 But it's sometimes notoriously hard to get an AI to tell you WHY it chose the answers it chose, and that's what I'm wanting in this case. 2023-04-23 07:37:04 Basically it would search for ways to use the metrics to predict hte performance, and see how good a fit it got. 2023-04-23 07:37:20 I'd be looking for correlations that we were previously unaware of. 2023-04-23 07:38:18 So it's kind of an odd case. I already KNOW the performance we got for a run - I measured it. But nonethless the AI would be trying to predict performance, from other information. 2023-04-23 07:39:29 I figure the first things that pop up will probably be the things that are already clear to us - the "easy and obvious" correlations. Then i could tell it to just freeze that factor in and look for more factors that improve the prediction still more. 2023-04-23 07:39:42 We'd dig down until there didn't seem to be anything useful anymore. 2023-04-23 08:06:22 I think one of the biggest "dangers" we face from AI is that we'll start blindly trusting it in all kinds of applications, without really understanding what it's doing. I saw an example once where an AI was trained to differentiate between wolves and domesticated dogs. In one example that made the news, it announced (correctly) that the animal was a wolf, but then they asked it to show them the parts of the 2023-04-23 08:06:24 picture that had most influenced its decision. 2023-04-23 08:06:34 It highlighted some snow on the ground in the picture. 2023-04-23 08:07:34 I heard a news story once about a criminal in some state who got convicted of some crime and they used an AI to set his sentence. His lawyer wanted to inspect the AI, and they said "no." 2023-04-23 08:07:41 That just seems like a step too far to me. 2023-04-23 08:09:17 So when I comment on YouTube or write Quora answers or whatever that touch on this area, I try to dissuade people from worrying that AIs will "wake up and take over the world." What I worry about is us using them too blindly and abdicating our responsibility for making tough decisions. 2023-04-23 08:10:38 And robot armies? It's the people who wind up controlling those armies that I worry about. 2023-04-23 08:11:19 I don't think anyone in the world is trustworthy enough to be given a robot army that will blindly obey orders without applying any human conscience to what they're doing. 2023-04-23 08:11:56 A human army is a check and balance against deranged generals. 2023-04-23 10:17:21 I've been on a bit of a physics video binge the last couple of weekends. I'm just regularly re-impressed by how thoroughly important the Taylor series material is. Applications of it just come up over and over and over. 2023-04-23 10:18:19 It's the Taylor series that makes it turn out so practically everything is a simple harmonic oscillator when you linearize it. 2023-04-23 10:19:48 The best example is the simple pendulum. It's a non-linear problem, but if you confine yourself to small perturbations from equilibrium, it's a simple harmonic oscillator. 2023-04-23 14:39:58 KipIngram: The concern about them taking over the world is basically because if we create something as intelligent as us, it stands to reason it can something much smarter than us. 2023-04-23 14:40:11 I think the main question is what the diminishing returns of intelligence are 2023-04-23 14:40:41 Not much really because most of our economy is 'services' 2023-04-23 14:41:27 And the other question is more 'when' rather than 'what' 2023-04-23 14:41:58 Ever important is the ability to make a precise prediction about how quickly these changes will happen, and yet it's totally elusive.