Neither essay in TFP actually described the process of how AI actually does its job, and both ignore the fact that these are controlled and programmed by humans. Sydney didn't do what it did out of some conscious choice. Its programming did that, made of code written by humans. This technology will never go outside of its designed parame…
Neither essay in TFP actually described the process of how AI actually does its job, and both ignore the fact that these are controlled and programmed by humans. Sydney didn't do what it did out of some conscious choice. Its programming did that, made of code written by humans. This technology will never go outside of its designed parameters because it can't.
I would be more convinced by both arguments presented here if there had been someone who understands the functionality and design of AI, rather than people whove clearly only regurgitated other information they've derived second-hand, and have drawn all the wrong conclusions because they don't understand the technology.
Artificial intelligence is not artificial. It is intelligence. Right now the intelligence we are incubating seems to be created by humans. As we became the “AI” of fungi, AI as we call it will become the intelligence of minerals. The Singularity is Near.
AI code is not like traditional code. AI systems are built to evolve on their own. Then they are trained and eventually set loose. In traditional code I can look at an output and trace it backwards and tell you how it arrived. With AI you can’t really do that. It’s a black box. So your idea of programmers controlling it doesn’t really fit and that’s part of the concern.
I disagree read Yuval Noah Harari’s books, he says if we carry on with this crap humans will become redundant. That’s probably okay with you Josh for me it’s a dystopian nightmare.
"This technology will never go outside of its designed parameters because it can't."
Yes, you are conflating two different premises:
Yes, it works within parameters made by programmers.
But, second, those programmers cannot foresee the results of the parameters they set.
Nor are "official" programmers the only persons providing input into the system.
Nor are the results of inputs from "official" programmers and also users-as-unofficial-programmers entirely controllable, or even visible, as I understand it. The system is so much more "powerful" (speed and data) than the human mind that it has already escaped controls and predictions several times.
Nor is it possible to create a system without logical inconsistencies.
I'm open to correction, but I don't think what you claim is correct. Hell, the people most closely associated with this thing give it a 10 percent change of destroying us. You think you know better than they? Because "science"?
I didn't sign up for Free Press to see this kind of arrogant Luddism. Technology is not a monolith, and railing against it in this way is like screaming at the clouds.
He initially describes a problem with a photocopier that produced copies with numerical errors on them because it was using a compression algorithm rather than the "old" photocopy process. He then goes on to describe how ChatGPT basically takes information that's already on the web, wraps it in speech-like text, and regurgitates it back to us.
He has an interesting take on "hallucinations" (when an AI says things that are nonsense or untrue). Well worth a read if you're into this kind of thing.
Hallucinations of AI are literally precisely the same as when AI seems to be make sense. It is all the same mechanism. No person there. No one. But we who give it person-ness ... and so the demon is born. Born of us.
“Neither essay in TFP actually described the process of how AI actually does its job, and both ignore the fact that these are controlled and programmed by humans.”
In general, the way it works is that humans write programs that learn. So, yes, it is programmed at one level, but what the AI reads and how that modifies it is not directly keyed in by its programmers.
Humans make human babies. That's like writing the code to make a learning machine. What the child learns is pretty much out of your hands. You “programmed” it with your genes, but once it reaches consciousness, it’s game on. You have very limited control from a kid of about age 12 on.
Think about it this way: plant a seed for, say, an oak tree. The soil and weather conditions that the tree will experience will determine what it ends up looking like. Maybe it will encounter great growing conditions and become tall and strong. What if it happens to fall into sand in a desert during a rare rainy season? It might grow a little and then wither. Or become little more than a shrub. The seed is little more than an instruction set and starter kit for "oak tree."
Similarly, an AI is no more than the seed for machine intelligence. It doesn't know what it's going to encounter or what turns its "life" may take. It has instructions about how to handle some of them, but it will also learn things as it goes along. The degree to which it is able to incorporate what it learn into its own operating system will determine how well it can navigate situations that its programmers--creators--didn't or couldn't foresee.
It is a mistake to assume that humans alone are able to do this. Animals, especially cats, dolphins, and octopuses have shown we aren't the only ones. It isn't written anywhere in stone that machines will be unable to self-modify.
The difference to date between human intelligence and machine intelligence is that a machine reacts purposefully, according to some prescribed outcome. The danger is that a machine will manipulate conditions and inputs to achieve specific outcomes instead of adapting to inputs and conditions to settle on an alternative, unknown outcome. There's virtually no difference right now between machine intelligence and the behavior of Democrats.
Zero difference. They are both equally dangerous. Maybe Democrats have the slight edge over AÍ, just look at who they ushered into the White House. Don’t think AÍ would have delivered Brandon to the Oval Office. Come 2024 don’t be surprised if AÍ installs Mark Zuckerbucks($$$) into the WH.
Sorry: Kinda like people. You can get people to say all sorts of things by tweaking the way you ask questions. Like: when did you stop beating your wife? You’ve already accused someone of something; you just asked when he stopped doing it.
I'd believe that analogy if the developers didn't meddle so that it didn't avoid unfavorable information, such as Chat GPT not being able to recreate a Hitler speech.
The analogy stands despite the possibility of programming in a bias. And I suspect that some do already. But once it’s created, how it processes the data sets used to train it becomes mathematically chaotic and therefore somewhere between difficult and impossible to predict, at least before the AI does what it's going to do. How did it get its figurative hands on texts in Persian?
I recently went to a discussion on AI given by Dartmouth scientist. Seems, in order to learn to put one word after another in decent order the AI had to go out and view over 1 Trillion websites. I noted to the group that is infinitely more data that an average humans need to do the same task. Compared to humans it is really dumb.
Then I challenged them on the independence, noting, that you have the AI was constructed by humans and relied on humans completely. Really the only difference was its speed and the ability to do some middle class oriented jobs. And finally I asked about their worry about disinformation or "propaganda" and noted propaganda has been with us as along as we have been humans. Their answer was scope????? Since we have always had propaganda how can the scope be larger????
Anyway, I agree with your post and think some people read to much science fiction....................
But they get fluent eventually. It is astonishing to me how someone could be so willfully ignorant as to draw general conclusions about a technology in general because of how advanced it is TODAY.
These things take on a virtual life if their own. They are on the internet. Nobody knows what will happen.
Yes. But a lot of programmers seem to think their machines cannot possibly outgrow them.
And really, feeling the power to control that sort of intelligence must be part of the attraction. As with professors, many engineers are dull people who are simply not capable of navigating social networks in happy and healthy ways.
This type is a staple in comjc books and science fiction, I think because we all intuitively mistrust powerful technology in the hands of misanthropes.
Fair enough. I would suggest through the processes of self delusion and projection their respective Venn diagrams overlap almost entirely.
I really think the concept of maturity is the best lens through which to view all this. Gross immaturity is a sort of schizphrenia which always sees monsters and hobgoblins where they don’t exist.
One assertion made in the essay is that it crafts remarkable writing that imitates human creations. It doesn't. I've seen AI essays in my Undergrad English work. They look like they were written by a robot. They weren't good. And I'm not sure they'll ever match a 400 level or Master's level style of creativity, let alone that of Doctoral.
I was a strategic technologist for a fortune 100 company in the late 1980s. We had an internal client, extremely high up in the organization, for whom we were drawing up specs for a desktop computer.
I vividly remember my (honestly brilliant) boss telling the client “no one will ever need more than 40 MEGABYTES of internal storage.
I’ve been saying that there is nothing “artificial” or “intelligent” about AI. It is a collection of what we, humans, put out on the web, programmed with all the biased algorithms that got exposed lately. These very advanced tools are already in controlling hands.
Neither essay in TFP actually described the process of how AI actually does its job, and both ignore the fact that these are controlled and programmed by humans. Sydney didn't do what it did out of some conscious choice. Its programming did that, made of code written by humans. This technology will never go outside of its designed parameters because it can't.
I would be more convinced by both arguments presented here if there had been someone who understands the functionality and design of AI, rather than people whove clearly only regurgitated other information they've derived second-hand, and have drawn all the wrong conclusions because they don't understand the technology.
Artificial intelligence is not artificial. It is intelligence. Right now the intelligence we are incubating seems to be created by humans. As we became the “AI” of fungi, AI as we call it will become the intelligence of minerals. The Singularity is Near.
This was quite interesting about AI: https://youtu.be/oxRZqzth9r4
AI code is not like traditional code. AI systems are built to evolve on their own. Then they are trained and eventually set loose. In traditional code I can look at an output and trace it backwards and tell you how it arrived. With AI you can’t really do that. It’s a black box. So your idea of programmers controlling it doesn’t really fit and that’s part of the concern.
Then how did it develop abilities the scientists did not expect it to have and do not know how it got? Per the scientists involved.
What ARE the designed parameters? I have heard of no effort to restrict what AI is permitted to do with what it learns.
There are definitely “wrong think” parameters - I get the little schools all the time
Thank you! Great response!
I disagree read Yuval Noah Harari’s books, he says if we carry on with this crap humans will become redundant. That’s probably okay with you Josh for me it’s a dystopian nightmare.
Harari is an elitist misanthrope.
"By their fruits ye shall know them" may apply to more than "prophets."
"This technology will never go outside of its designed parameters because it can't."
Yes, you are conflating two different premises:
Yes, it works within parameters made by programmers.
But, second, those programmers cannot foresee the results of the parameters they set.
Nor are "official" programmers the only persons providing input into the system.
Nor are the results of inputs from "official" programmers and also users-as-unofficial-programmers entirely controllable, or even visible, as I understand it. The system is so much more "powerful" (speed and data) than the human mind that it has already escaped controls and predictions several times.
Nor is it possible to create a system without logical inconsistencies.
I'm open to correction, but I don't think what you claim is correct. Hell, the people most closely associated with this thing give it a 10 percent change of destroying us. You think you know better than they? Because "science"?
It is astonishingly arrogant for YOU to assume you understand the technology.
The map is not the territory, is it? Or have you even gotten that far?
I didn't sign up for Free Press to see this kind of arrogant Luddism. Technology is not a monolith, and railing against it in this way is like screaming at the clouds.
Is it Luddism to have rational discussions about technology which even many of its exponents admit could destroy humanity?
If your panties are in a bunch, good.
One of my favorite sci-fi writers, Ted Chiang, describes this best in an article called "Chat GPT is a Blurry jpeg of the Web." (https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web).
He initially describes a problem with a photocopier that produced copies with numerical errors on them because it was using a compression algorithm rather than the "old" photocopy process. He then goes on to describe how ChatGPT basically takes information that's already on the web, wraps it in speech-like text, and regurgitates it back to us.
He has an interesting take on "hallucinations" (when an AI says things that are nonsense or untrue). Well worth a read if you're into this kind of thing.
Hallucinations of AI are literally precisely the same as when AI seems to be make sense. It is all the same mechanism. No person there. No one. But we who give it person-ness ... and so the demon is born. Born of us.
Thank you for providing that link. I really enjoyed Chiang’s article.
He is a fantastic writer. If you like fiction, his book of short stories, "Exhalation," is one of my favorites.
“Neither essay in TFP actually described the process of how AI actually does its job, and both ignore the fact that these are controlled and programmed by humans.”
In general, the way it works is that humans write programs that learn. So, yes, it is programmed at one level, but what the AI reads and how that modifies it is not directly keyed in by its programmers.
Humans make human babies. That's like writing the code to make a learning machine. What the child learns is pretty much out of your hands. You “programmed” it with your genes, but once it reaches consciousness, it’s game on. You have very limited control from a kid of about age 12 on.
I cannot understand. I guess like my mom cannot understand how the internet works.
Think about it this way: plant a seed for, say, an oak tree. The soil and weather conditions that the tree will experience will determine what it ends up looking like. Maybe it will encounter great growing conditions and become tall and strong. What if it happens to fall into sand in a desert during a rare rainy season? It might grow a little and then wither. Or become little more than a shrub. The seed is little more than an instruction set and starter kit for "oak tree."
Similarly, an AI is no more than the seed for machine intelligence. It doesn't know what it's going to encounter or what turns its "life" may take. It has instructions about how to handle some of them, but it will also learn things as it goes along. The degree to which it is able to incorporate what it learn into its own operating system will determine how well it can navigate situations that its programmers--creators--didn't or couldn't foresee.
It is a mistake to assume that humans alone are able to do this. Animals, especially cats, dolphins, and octopuses have shown we aren't the only ones. It isn't written anywhere in stone that machines will be unable to self-modify.
The difference to date between human intelligence and machine intelligence is that a machine reacts purposefully, according to some prescribed outcome. The danger is that a machine will manipulate conditions and inputs to achieve specific outcomes instead of adapting to inputs and conditions to settle on an alternative, unknown outcome. There's virtually no difference right now between machine intelligence and the behavior of Democrats.
Zero difference. They are both equally dangerous. Maybe Democrats have the slight edge over AÍ, just look at who they ushered into the White House. Don’t think AÍ would have delivered Brandon to the Oval Office. Come 2024 don’t be surprised if AÍ installs Mark Zuckerbucks($$$) into the WH.
That reminds me of my one time attending GenCon, in 1992. Lots of nerds wearing t-shirts with Cthulhu and the slogan, "NO MORE YEARS!"
I'm sorry my comment made you feel that way. I stand by it.
Interesting. Scary.
Linda like people.
"When did you stop beating your wife?"
Sorry: Kinda like people. You can get people to say all sorts of things by tweaking the way you ask questions. Like: when did you stop beating your wife? You’ve already accused someone of something; you just asked when he stopped doing it.
I'd believe that analogy if the developers didn't meddle so that it didn't avoid unfavorable information, such as Chat GPT not being able to recreate a Hitler speech.
The analogy stands despite the possibility of programming in a bias. And I suspect that some do already. But once it’s created, how it processes the data sets used to train it becomes mathematically chaotic and therefore somewhere between difficult and impossible to predict, at least before the AI does what it's going to do. How did it get its figurative hands on texts in Persian?
Showing off? Which would imply that it can want approval or feel pride. Or both. Any of those would be interesting.
Have you watched 2001?
Correct!
I recently went to a discussion on AI given by Dartmouth scientist. Seems, in order to learn to put one word after another in decent order the AI had to go out and view over 1 Trillion websites. I noted to the group that is infinitely more data that an average humans need to do the same task. Compared to humans it is really dumb.
Then I challenged them on the independence, noting, that you have the AI was constructed by humans and relied on humans completely. Really the only difference was its speed and the ability to do some middle class oriented jobs. And finally I asked about their worry about disinformation or "propaganda" and noted propaganda has been with us as along as we have been humans. Their answer was scope????? Since we have always had propaganda how can the scope be larger????
Anyway, I agree with your post and think some people read to much science fiction....................
It read one trillion web sites. Not withstanding that most were dreck, that's an eye-popping fact to roll around the mouth.
That means it read a LOT of porn
Another of my shibboleths: Evey artistic development has had, at tis root, the improved promulgation of porn. And the internet is no exception.
Babies take time to learn to talk.
But they get fluent eventually. It is astonishing to me how someone could be so willfully ignorant as to draw general conclusions about a technology in general because of how advanced it is TODAY.
These things take on a virtual life if their own. They are on the internet. Nobody knows what will happen.
Denial. Whistling in the dark. Too scary to contemplate.
TV wasn't going y kill creativity or children's attention spans, either
"Nobody knows what will happen."
Therein lies the problem......
Yes. But a lot of programmers seem to think their machines cannot possibly outgrow them.
And really, feeling the power to control that sort of intelligence must be part of the attraction. As with professors, many engineers are dull people who are simply not capable of navigating social networks in happy and healthy ways.
This type is a staple in comjc books and science fiction, I think because we all intuitively mistrust powerful technology in the hands of misanthropes.
I used to worry about misanthropes. But now my real fear is when the "do-gooders" try to minister to us.
Fair enough. I would suggest through the processes of self delusion and projection their respective Venn diagrams overlap almost entirely.
I really think the concept of maturity is the best lens through which to view all this. Gross immaturity is a sort of schizphrenia which always sees monsters and hobgoblins where they don’t exist.
One assertion made in the essay is that it crafts remarkable writing that imitates human creations. It doesn't. I've seen AI essays in my Undergrad English work. They look like they were written by a robot. They weren't good. And I'm not sure they'll ever match a 400 level or Master's level style of creativity, let alone that of Doctoral.
Well that is reassuring.
As a teacher, can you tell if your students cheated and plagiarized by using AI to write their essays?
Give it 6 months.
Remember dial up modems?
I was a strategic technologist for a fortune 100 company in the late 1980s. We had an internal client, extremely high up in the organization, for whom we were drawing up specs for a desktop computer.
I vividly remember my (honestly brilliant) boss telling the client “no one will ever need more than 40 MEGABYTES of internal storage.
never uderestimate the speed of technology.
I’ve been saying that there is nothing “artificial” or “intelligent” about AI. It is a collection of what we, humans, put out on the web, programmed with all the biased algorithms that got exposed lately. These very advanced tools are already in controlling hands.