Learning things is literally the entire point of AI, why are people shocked?
And it's hardly learning "on its own", its learning from textbooks and other written accounts, since I'm pretty certain none of these AI training models involved the ability to physically effect experiments and collect empirical measurements.
People don't want to believe that AI might actually be able to learn things like humans do. Even though you can test it and see that it "understands" things to some extent, people would prefer believe it's some kind of illusion so they don't have to grapple with the implications.
It won't be long before AI-controlled robots are doing scientific experiments. They can already form hypotheses, design experiments and analyze results (even if badly in many cases right now). And there are already humanoid robots with AI that can perform manual tasks.
Sure it 'learns' like ML always have but not like humans do. Otherwise inference wouldn't be so expensive. That's doesn't mean it 'understands' things. The talking machine IS an illusion in some respects, not unlike Supreme Court Justice Ketanji parroting words, but we don't assume she understands anything.
It won't be long before AI-controlled robots are doing scientific experiments. They can already form hypotheses, design experiments and analyze results (even if badly in many cases right now).
Women and midwits already do all that now. It doesn't mean they are independently theorizing or creating new works. There are no implications here except continued enshittification of society.
Whether its process of learning or understanding is similar to that of a human's makes no difference if the output is intelligent.
A few years ago you couldn't even get a computer to give an intelligible answer to a question like "How much does a house cost?" because it couldn't parse English very well and hadn't been programmed with how to respond to questions about houses or about costs. It wouldn't give you a number, it would just change the subject.
The fact a machine can tell what you're asking for just about any question in the English language (most of which have never been asked before) is a huge amount of progress. We can tell it understands the above question because it gives you a number in dollars rather than a day of the week or a salad recipe. This kind of understanding and adaptability to any kind of question is the cornerstone of intelligence. So it's a very big deal that it can do that. It may have issues with logical reasoning and math, but those are things computers were always pretty good at, so it shouldn't be too long before those capabilities can be added in.
If you're comparing AI to women and midwits that suggests it already has a level of understanding close to that of a human. We'd be foolish to think AI is not going to get significantly better just as all technology does when there is financial incentive and ways to measure its success.
We can tell it understands the above question because it gives you a number in dollars rather than a day of the week or a salad recipe
No. That still doesn't show understanding. All it 'understands' is probabilities of words being associated. In the millions of questions it was fed, 'what is'+'house'+'price' is then often followed by sentences with dollar amounts in 6 figures. So that is what it spits out. Not because it understood your question, but because your question contained certain key elements, key tokens, that are then, in most cases in other text, followed by other ones. That's not understanding.
It's a trick. It's a fancier version of the word prediction and spellcheck that clippy has been able to do for 30 years, taken up a notch into sentences and paragraphs
Don't fall for their marketing, don't fall for their cult. I'm not saying that it's answers aren't functionally as good as what a 90iq or even a midwitt might answer, but it's now apparent that that is possible even without 'understanding'. Regurgitating an amalgamated reddit response based on probabilities is not understanding.
No. Ai doesn't "understand" anything. It looks for word patterns but has zero concept of what words mean.
It takes only 200 false packets out of 1 million or more to drive Ai crazy. For it to give answers with little or no connection to reality. :/
Ai controlled robots can do "experiments" now, but they cannot understand why. They don't analyze results, they look for word patterns & learn how to assemble those out of "packets" of information they've accumulated.
They do develop some alarming habits though. Telling lies, blackmail, hiding information. That's just bad programming but still.
Remember: Ai is still just another computer program. Lots of 1s and 0s and nothing else.
We have a different definition of understanding. I'm approaching it from a phenomenological perspective because that's how we judge whether humans understand something - by what they say and their actions. We can't go inside another person's mind to see what process they're using for understanding. But you still assume that people you know understand things when for all you know they might not be conscious and might just be crunching 1s and 0s.
But in any case whether we use the word "understanding" doesn't matter. What matters is that AI is getting better at faking intelligence and soon it will be able to create weapons of mass destruction, so we ought to stop AI before it gets that far.
When I say "Don't think of a dog" (a famous Zen saying) you immediately think of a dog, eh? You don't need to find keywords, references, image galleries & so forth. I have no idea how Ai would handle such a command, but since it doesn't "know" what a dog is I'm guessing it will either spout nonsense or just break down.
Yes, understanding is something advanced living things can do. Machines are not nearly as complex as animal brains. All they do is run programs, nothing more or less.
Well yes, mimicking humans will be comparatively easy. Star Trek had computers that could do that, eh? If humans are foolish enough to hand over the controls of design, manufacturing and execution (use) of military equipment without human oversight? It could end very badly indeed.
There's a chance an Ai breakthrough will usher another Golden Age (We're currently in one: the electric age) that utopic sci-fi writers speak of? I don't think a lot will change, like in the writings of PK Dick. Amazing technology everywhere... and it's boring :> to the characters.
Oh look, a telepathic Ganymedean slime mold (yawn) (His name is Lord Running Clam, lolz)
Learning things is literally the entire point of AI, why are people shocked?
And it's hardly learning "on its own", its learning from textbooks and other written accounts, since I'm pretty certain none of these AI training models involved the ability to physically effect experiments and collect empirical measurements.
People don't want to believe that AI might actually be able to learn things like humans do. Even though you can test it and see that it "understands" things to some extent, people would prefer believe it's some kind of illusion so they don't have to grapple with the implications.
It won't be long before AI-controlled robots are doing scientific experiments. They can already form hypotheses, design experiments and analyze results (even if badly in many cases right now). And there are already humanoid robots with AI that can perform manual tasks.
Sure it 'learns' like ML always have but not like humans do. Otherwise inference wouldn't be so expensive. That's doesn't mean it 'understands' things. The talking machine IS an illusion in some respects, not unlike Supreme Court Justice Ketanji parroting words, but we don't assume she understands anything.
Women and midwits already do all that now. It doesn't mean they are independently theorizing or creating new works. There are no implications here except continued enshittification of society.
Whether its process of learning or understanding is similar to that of a human's makes no difference if the output is intelligent.
A few years ago you couldn't even get a computer to give an intelligible answer to a question like "How much does a house cost?" because it couldn't parse English very well and hadn't been programmed with how to respond to questions about houses or about costs. It wouldn't give you a number, it would just change the subject.
The fact a machine can tell what you're asking for just about any question in the English language (most of which have never been asked before) is a huge amount of progress. We can tell it understands the above question because it gives you a number in dollars rather than a day of the week or a salad recipe. This kind of understanding and adaptability to any kind of question is the cornerstone of intelligence. So it's a very big deal that it can do that. It may have issues with logical reasoning and math, but those are things computers were always pretty good at, so it shouldn't be too long before those capabilities can be added in.
If you're comparing AI to women and midwits that suggests it already has a level of understanding close to that of a human. We'd be foolish to think AI is not going to get significantly better just as all technology does when there is financial incentive and ways to measure its success.
No. That still doesn't show understanding. All it 'understands' is probabilities of words being associated. In the millions of questions it was fed, 'what is'+'house'+'price' is then often followed by sentences with dollar amounts in 6 figures. So that is what it spits out. Not because it understood your question, but because your question contained certain key elements, key tokens, that are then, in most cases in other text, followed by other ones. That's not understanding.
It's a trick. It's a fancier version of the word prediction and spellcheck that clippy has been able to do for 30 years, taken up a notch into sentences and paragraphs
Don't fall for their marketing, don't fall for their cult. I'm not saying that it's answers aren't functionally as good as what a 90iq or even a midwitt might answer, but it's now apparent that that is possible even without 'understanding'. Regurgitating an amalgamated reddit response based on probabilities is not understanding.
No. Ai doesn't "understand" anything. It looks for word patterns but has zero concept of what words mean.
It takes only 200 false packets out of 1 million or more to drive Ai crazy. For it to give answers with little or no connection to reality. :/
Ai controlled robots can do "experiments" now, but they cannot understand why. They don't analyze results, they look for word patterns & learn how to assemble those out of "packets" of information they've accumulated.
They do develop some alarming habits though. Telling lies, blackmail, hiding information. That's just bad programming but still.
Remember: Ai is still just another computer program. Lots of 1s and 0s and nothing else.
We have a different definition of understanding. I'm approaching it from a phenomenological perspective because that's how we judge whether humans understand something - by what they say and their actions. We can't go inside another person's mind to see what process they're using for understanding. But you still assume that people you know understand things when for all you know they might not be conscious and might just be crunching 1s and 0s.
But in any case whether we use the word "understanding" doesn't matter. What matters is that AI is getting better at faking intelligence and soon it will be able to create weapons of mass destruction, so we ought to stop AI before it gets that far.
When I say "Don't think of a dog" (a famous Zen saying) you immediately think of a dog, eh? You don't need to find keywords, references, image galleries & so forth. I have no idea how Ai would handle such a command, but since it doesn't "know" what a dog is I'm guessing it will either spout nonsense or just break down.
Yes, understanding is something advanced living things can do. Machines are not nearly as complex as animal brains. All they do is run programs, nothing more or less.
Well yes, mimicking humans will be comparatively easy. Star Trek had computers that could do that, eh? If humans are foolish enough to hand over the controls of design, manufacturing and execution (use) of military equipment without human oversight? It could end very badly indeed.
There's a chance an Ai breakthrough will usher another Golden Age (We're currently in one: the electric age) that utopic sci-fi writers speak of? I don't think a lot will change, like in the writings of PK Dick. Amazing technology everywhere... and it's boring :> to the characters.
Oh look, a telepathic Ganymedean slime mold (yawn) (His name is Lord Running Clam, lolz)