All Substack authors received the latest email from "headquarters" announcing new AI tools will be introduced on Substack, and that we will be facilitated by it. This is my comment submitted there.
Normally I wouldn't even write about A.I. because this site is more about linguistics than technology. But this was an email from Substack, so I figured that everyone coming to my site, would have read it, all the Substack people that is.
I freely admit that AI is a force and we don't know how it will hit us. (Well, we do; it is already happening since Cambridge Analytica in 2016.) These are social media memes that people shout, but I don't think anyone really knows what it is. I was talking with a programmer yesterday, and he was explaining Algorithms in that you can ask data for specific facts, like "who wears orange hats", or some such nonsense.
Even he spoke of AI "approaching consciousness", which, who knows what that means???
Will it do or say something that has never been said or thought of before? On what basis? does it have "whims". I think a lot of people debate "what is consciousness", almost as unknowingly as speaking of AI. Consciousness is a big topic, in that some day there might be a post about it here, from me or others.
One thing about consciousness, it must be empty, until you fill it up, with all your thoughts, beliefs, memories and conjectures. After all that; IS THERE A WAY TO EMPTY IT? Or does it just get more and more full?
I am of the opinion that AI functions as an "objective", "impartial" something on which to offload responsibility for our decisions and decision-making as a whole. The attempts to ascribe consciousness or superlative progress to AI is a method of justifying this surrender of responsibility.
Not to say that there is no growth in the space, but there is always an ulterior motive to outside observers like ourselves. Ultimately, only those who directly work on these models will know the real score.
There is a big narrative around it. That already rings some very loud alarm bells.
What am I a skeptic? Technology is give and take, right. Or is it all for free? We have the internet and smart phones, and they have everything that we do or say for 25 years. Now you can line up and read in your thoughts directly into Chat GPT, kind of a self immolation. So far we think, well, it is just a story, how can it hurt me? Maybe your right.
But then why do we have all this free stuff which big tech spends 100's of $ Billions on, if our life story is worthless? You know, our WNT site is about creating a community, and part of that is protecting each and every one. We're not going to stop big tech, but sooner or later we will have to talk about practical and easy steps to remove some of the potential risk. I think it can be done without withdrawing from the grid.
This line sticks with me: "So we are left with A.I. as an 'intelligence' that can never get out of the past tense." Oh, yes. So, I'm concerned about the implications of that. We're constantly moving forward based on the past — but sometimes we have to break with the past, rebel against it, go somewhere new. I'm afraid we'll be reduced to a giant library of past impulses and mistakes and formulaic constructions of problems, that this will come to seem the norm. I think a year of ChatGPT has already shoved us in a direction tech was already pushing us in via the web and social media: reducing writing to information processing. I refuse to be limited in that way, but it doesn't mean I won't be in any case. Still, the refusing and resisting matters for me and my writing.
ML merely creates a response based on probabilities of the data in its training set, with some additional massaging to make the output come out in a way that we find acceptable for the specific purpose we plan to use it for (this massaging comes from what's called fine-tuning, or from additional filters added to try and filter out "unwelcome" behavior - most of which can be "jail-broken" since they're tacked on top without altering the objectionable data in the training set).
What ML can not come even close to is understanding the underlying concepts in the way a human does.
There are tons of examples of questions you can ask an ML system that illustrate that it doesn't understand anything, and is just regurgitating modified forms of it's training data. One great example is that when you ask an ML image creator to generate a traditional watch face (hour and minute hands) showing a specific time, you'll usually get back a watch showing the time of 10:10 instead. The reason for this is that almost all the watch faces in it's training data come from watch advertisements, which all show the time at 10:10 (since it's been identified as the most visually appealing time to show).
No, "AI" isn't conscious - that's just projection and wishful thinking. And no, ML won't ever become conscious. That's not to say someone won't eventually find a way to make a conscious AI, just that ML is not the path that will get there - it will take new technology (probably several of them, at least) to reach AGI.
However, ML is a *tool*, when when used *properly* can make us wetware intelligences more productive. At some tasks, it will outperform us every time. At others, it will never be able to match a human. The tasks it will outperform us on are the ones we're generally bad at anyway, like finding patterns in oodles of data. The tasks it will never match us at are the ones that require true originality.
Those of us that learn to use the tool for what it's good for - and learn to avoid trying to use it for what it's bad at - will benefit from it. And groups (orgs, companies, countries, etc.) that can tell the difference will outcompete the groups that can't.
Normally I wouldn't even write about A.I. because this site is more about linguistics than technology. But this was an email from Substack, so I figured that everyone coming to my site, would have read it, all the Substack people that is.
I freely admit that AI is a force and we don't know how it will hit us. (Well, we do; it is already happening since Cambridge Analytica in 2016.) These are social media memes that people shout, but I don't think anyone really knows what it is. I was talking with a programmer yesterday, and he was explaining Algorithms in that you can ask data for specific facts, like "who wears orange hats", or some such nonsense.
Even he spoke of AI "approaching consciousness", which, who knows what that means???
Will it do or say something that has never been said or thought of before? On what basis? does it have "whims". I think a lot of people debate "what is consciousness", almost as unknowingly as speaking of AI. Consciousness is a big topic, in that some day there might be a post about it here, from me or others.
One thing about consciousness, it must be empty, until you fill it up, with all your thoughts, beliefs, memories and conjectures. After all that; IS THERE A WAY TO EMPTY IT? Or does it just get more and more full?
.
I am of the opinion that AI functions as an "objective", "impartial" something on which to offload responsibility for our decisions and decision-making as a whole. The attempts to ascribe consciousness or superlative progress to AI is a method of justifying this surrender of responsibility.
Not to say that there is no growth in the space, but there is always an ulterior motive to outside observers like ourselves. Ultimately, only those who directly work on these models will know the real score.
There is a big narrative around it. That already rings some very loud alarm bells.
What am I a skeptic? Technology is give and take, right. Or is it all for free? We have the internet and smart phones, and they have everything that we do or say for 25 years. Now you can line up and read in your thoughts directly into Chat GPT, kind of a self immolation. So far we think, well, it is just a story, how can it hurt me? Maybe your right.
But then why do we have all this free stuff which big tech spends 100's of $ Billions on, if our life story is worthless? You know, our WNT site is about creating a community, and part of that is protecting each and every one. We're not going to stop big tech, but sooner or later we will have to talk about practical and easy steps to remove some of the potential risk. I think it can be done without withdrawing from the grid.
.
This line sticks with me: "So we are left with A.I. as an 'intelligence' that can never get out of the past tense." Oh, yes. So, I'm concerned about the implications of that. We're constantly moving forward based on the past — but sometimes we have to break with the past, rebel against it, go somewhere new. I'm afraid we'll be reduced to a giant library of past impulses and mistakes and formulaic constructions of problems, that this will come to seem the norm. I think a year of ChatGPT has already shoved us in a direction tech was already pushing us in via the web and social media: reducing writing to information processing. I refuse to be limited in that way, but it doesn't mean I won't be in any case. Still, the refusing and resisting matters for me and my writing.
"AI" as the term is currently used is really just "ML" (Machine Learning). And by definition and design, it's derivative, not original. (For more on ML limitations, see https://bunkum.substack.com/p/some-benefits-limitations-and-dangers).
ML merely creates a response based on probabilities of the data in its training set, with some additional massaging to make the output come out in a way that we find acceptable for the specific purpose we plan to use it for (this massaging comes from what's called fine-tuning, or from additional filters added to try and filter out "unwelcome" behavior - most of which can be "jail-broken" since they're tacked on top without altering the objectionable data in the training set).
What ML can not come even close to is understanding the underlying concepts in the way a human does.
There are tons of examples of questions you can ask an ML system that illustrate that it doesn't understand anything, and is just regurgitating modified forms of it's training data. One great example is that when you ask an ML image creator to generate a traditional watch face (hour and minute hands) showing a specific time, you'll usually get back a watch showing the time of 10:10 instead. The reason for this is that almost all the watch faces in it's training data come from watch advertisements, which all show the time at 10:10 (since it's been identified as the most visually appealing time to show).
No, "AI" isn't conscious - that's just projection and wishful thinking. And no, ML won't ever become conscious. That's not to say someone won't eventually find a way to make a conscious AI, just that ML is not the path that will get there - it will take new technology (probably several of them, at least) to reach AGI.
However, ML is a *tool*, when when used *properly* can make us wetware intelligences more productive. At some tasks, it will outperform us every time. At others, it will never be able to match a human. The tasks it will outperform us on are the ones we're generally bad at anyway, like finding patterns in oodles of data. The tasks it will never match us at are the ones that require true originality.
Those of us that learn to use the tool for what it's good for - and learn to avoid trying to use it for what it's bad at - will benefit from it. And groups (orgs, companies, countries, etc.) that can tell the difference will outcompete the groups that can't.
Chief, thanks for this post.
ANYONE WHO HAS MADE IT DOWN TO HERE: be sure to follow Chief Bunkum's link up above. He will tell you what you want to know about A.I..
.