Pedantry over the definition of "AI" can go deep. In the classical sense within computer science, AI is both broader and different than the modern vernacular definition. An AI is a rational agent...
Pedantry over the definition of "AI" can go deep.
In the classical sense within computer science, AI is both broader and different than the modern vernacular definition. An AI is a rational agent operating within an environment. Notably that describes "what", not "how" (see: slides from my alma mater). Do the algorithms that power Pacman ghosts, just Dijkstra's algorithm, count as AI? Classically, yes! They are absolutely AI. I assure you these are even "stupider" than anything the video presents as "not AI".
Does Stable Diffusion count as AI? No! No it doesn't. Does ChatGPT? Maybe! You'll have to define the reward metrics and the environment, though, for the particular situation.
Then there's the new fangled layman definition, which is not clearly defined, but broadly I generalize it as sufficiently complicated neural networks. Notably, this defines the WHAT not the HOW. These neural networks can be used for anything and fit in this definition.
Now, while I understand the concerns the video author has, trying to add an additional layman definition which seems to make it congruent with AGI, or artificial general intelligence, seems both pointless and only serving to further confusion. Language is natural, so while I resist the new definition in my own language, it is inevitable that people will use the new definition.
Yeah it's incredibly frustrating for me when people who usually aren't familiar with the history of AI get up in arms about using it for anything that isn't AGI. That's not how we've ever used...
Yeah it's incredibly frustrating for me when people who usually aren't familiar with the history of AI get up in arms about using it for anything that isn't AGI. That's not how we've ever used that word! And while I think people have very distorted views of what modern AI can do (especially generative LLMs like ChatGPT), insisting that nothing other than AGI "counts" as AI doesn't actually help dispel misconceptions there. More often than not ime it's used as some weird flex or gotcha when the term "AI" isn't the actual discussion topic.
Funny because when AI was being posted all over Reddit...I said the same shit she did yet was told that I'm full of shit and wrong...even though I work heavily in the field lol. Just a bunch of...
Funny because when AI was being posted all over Reddit...I said the same shit she did yet was told that I'm full of shit and wrong...even though I work heavily in the field lol. Just a bunch of keyboard warriors telling me that AI does exist. And how AGI is the new word for AI...the public doesn't use AGI...they use AI...glad someone with clout actually is calling this out.
The public thinks and still does think that AI = something that needs no human input and is sentient.
Five minutes in and this has a very human-centric viewpoint. The reason presented for why AI is not intelligence is that you can't tell it when it's wrong or have it explain it's thought process....
Five minutes in and this has a very human-centric viewpoint. The reason presented for why AI is not intelligence is that you can't tell it when it's wrong or have it explain it's thought process. These are both aspects of language, not intelligence: two things which are conflated in this video.
I can't explain to a fly how to avoid fly paper. Even if I went up to a fly and said "hey fly, that sticky paper isn't a fruit you can eat. it's a scented strip of adhesive material designed to trap you", it wouldn't understand. Maybe I could 'train' a new fly to avoid fly paper by breeding a few generations of flies selected to avoid fly paper, which would work really well until it also starts avoiding hanging strips of fruit.
This might not be a universally held belief, but I think flies have some (very small) level of intelligence. So if you want to argue that machine learning models aren't intelligent, I think it would be more illuminating to compare them to flies, rather than people. Not only because they are somewhat alien intelligences, but also because they are much closer in computation power to an insect than to a person.
But that flies in the face of how popular media is reporting on AI, how ai is being talked about by general enthusiasts, and how business grifters are trying to sell AI. Which is what this video...
So if you want to argue that machine learning models aren't intelligent, I think it would be more illuminating to compare them to flies, rather than people. Not only because they are somewhat alien intelligences, but also because they are much closer in computation power to an insect than to a person.
But that flies in the face of how popular media is reporting on AI, how ai is being talked about by general enthusiasts, and how business grifters are trying to sell AI. Which is what this video is about.
The video is very correct about the fact that as A.I. right now it doesn't not have the same level of reasoning and abstract thinking that is expected out of people.
The video just takes the opposite, similarly unnuanced stance. Obviously we don't have AI as intelligent as humans right now, but it's also a mistake to dismiss all AI as unintelligent. We need to...
The video just takes the opposite, similarly unnuanced stance. Obviously we don't have AI as intelligent as humans right now, but it's also a mistake to dismiss all AI as unintelligent. We need to look more closely at the actual capabilities of these systems, rather than using philosophy to dismiss or accept them uncritically, and determine when we want to use them, how much supervision is required, etc. And this is a process that is continuous, not fixed in one point in time, as these systems improve.
Which is exactly what she's doing when she says it's a tool that needs expert supervision to check it's facts and efficacy. Not trying to be a dick, but did you finish the video?
We need to look more closely at the actual capabilities of these systems, ... determine when we want to use them
Which is exactly what she's doing when she says it's a tool that needs expert supervision to check it's facts and efficacy. Not trying to be a dick, but did you finish the video?
First of all, I'm mostly arguing against the point that AI systems aren't intelligent. I don't think there's a way to say for sure one way or the other (although I lean strongly towards them being...
First of all, I'm mostly arguing against the point that AI systems aren't intelligent. I don't think there's a way to say for sure one way or the other (although I lean strongly towards them being intelligent).
My interpretation of what she said was more pessimistic though: My interpretation was that she was saying that machine learning systems will always need expert supervision, and will always be worse than humans, because they are not truly intelligent. Where I disagree is I think there will be a point where AI systems surpass humans, and where human supervision is actually less reliable.
Fine, you can argue against that point that was never actually stated in the video. That's fine. But at the end of the day saying that the tech will one day be there doesn't justify any of the...
My interpretation of what she said was more pessimistic though: My interpretation was that she was saying that machine learning systems will always need expert supervision, and will always be worse than humans, because they are not truly intelligent. Where I disagree is I think there will be a point where AI systems surpass humans, and where human supervision is actually less reliable.
Fine, you can argue against that point that was never actually stated in the video. That's fine.
But at the end of the day saying that the tech will one day be there doesn't justify any of the current AI tech bubble.
AI isn't just chatgpt and this video isn't just talking about chat gpt. Chatgpt is just one of dozens of ai tools that exist or are being built, and there will no doubt be countless more in the...
AI isn't just chatgpt and this video isn't just talking about chat gpt. Chatgpt is just one of dozens of ai tools that exist or are being built, and there will no doubt be countless more in the future.
A.i. Definitely needs supervision for any use where the a.i. output may be used to make decision(medical diagnosis etc). You're so on the defensive about chat gpt you are missing the point entirely.
The video never seeks demonize a.i. it's just very realistic about the limitations and pushing back against this weird tech bubble nonsense pushing it as a magical solution.
I definitely agree. I work heavily in this field and one thing that I’ve grown to believe is that intelligence is not an emergent property of humans or even life itself, but an emergent property...
I definitely agree. I work heavily in this field and one thing that I’ve grown to believe is that intelligence is not an emergent property of humans or even life itself, but an emergent property of the universe. Highly proficient language models are often conflated with AGI because as people, language captures almost all of what we are able to experience or conceive.
Using neural networks as an example, they do learn novel, intelligent concepts that were not explicitly taught and that we do no understand. This has been the case for nearly a decade—For instance, image models have been shown to exhibit an understanding of 3D space, even though they have not been given 3D information. Recently, LLMs have been shown to exhibit spatial reasoning to an extent, and can lean on language to explain their conclusions and even arrive at better conclusions for doing so. They are also self reflective without being trained specifically to be reflective.
Disputing the use of the term AI in the video is to highlight a certain point: AI is not human-like intelligence as it is painted as in media and some people genuinely believe it is.
Disputing the use of the term AI in the video is to highlight a certain point: AI is not human-like intelligence as it is painted as in media and some people genuinely believe it is.
This is the main argument, the public hears AI and thinks sentient being. Academia has all kinds of nuances for AI which is just a generic term used. I don't know why people are trying to argue...
This is the main argument, the public hears AI and thinks sentient being. Academia has all kinds of nuances for AI which is just a generic term used. I don't know why people are trying to argue what academia uses vs the public. The general public thinks AI is the shit from syfi movies not ML or game AI that's dumb as a box of rocks.
Except that's not how AI fear-mongering is being peddled to the masses, is it? The way media outlets go on about it, you'd think SHODAN is just around the corner. No subtlety or nuance, no careful...
Except that's not how AI fear-mongering is being peddled to the masses, is it? The way media outlets go on about it, you'd think SHODAN is just around the corner. No subtlety or nuance, no careful definition of terms, just the use of the term "AI" and tacit allusion to sci-fi horror stories.
Let's please not compare how people who talk about games use "AI" and how the media and tech bubble enthusiasts have been talking about AI. The majority of people who play video games understand...
Probably the most common use of AI outside academia is for games.
Let's please not compare how people who talk about games use "AI" and how the media and tech bubble enthusiasts have been talking about AI. The majority of people who play video games understand that enemy AI is not particularly advanced or anything that closely resembles human intelligence .
Any videogame enemy with behaviour with the slightest sophistication above "blindly charge the player" is touted as groundbreaking AI, and that bar hasn't substantially raised in 20 years.
Any videogame enemy with behaviour with the slightest sophistication above "blindly charge the player" is touted as groundbreaking AI, and that bar hasn't substantially raised in 20 years.
Didn't F.E.A.R. have some ground breaking "AI" for its time, even though it was just basically a script of if statements lol? I don't think anything has changed from that era. It's, if attacked,...
Didn't F.E.A.R. have some ground breaking "AI" for its time, even though it was just basically a script of if statements lol? I don't think anything has changed from that era. It's, if attacked, find cover and try to kill human player, flank if you have more than one CPU with you, throw grenade, but nothing seriously amazing.
Yea F.E.A.R. is that game that I believe everyone at the time was talking about with the AI, the story was good but not ground breaking and neither was the gameplay, it was the AI that had all the...
Yea F.E.A.R. is that game that I believe everyone at the time was talking about with the AI, the story was good but not ground breaking and neither was the gameplay, it was the AI that had all the articles about it.
As someone learning statistics, the more I learn about machine learning techniques and AI the more I am convinced it’s just fancy statistical modelling with some rebranding. I’m curious if anyone...
As someone learning statistics, the more I learn about machine learning techniques and AI the more I am convinced it’s just fancy statistical modelling with some rebranding.
I’m curious if anyone knows of any techniques that are very dissimilar to standard statistics. Where do you draw the line. Maybe classification is still just stats, maybe a neural network is actually an AI technique and not “just stats”.
As far as I can tell there’s a lot of analogues for example image classification I.e cat vs not cat seems very similar to a logistic regression or at least a logistic activation function and labelling images as cat vs not cat and doing a logit on image parameters might get you a model that can discern images successfully. I can only assume the techniques for AI which I do not know / haven’t gotten to learning yet, do the same kind of thing in a more sophisticated way.
I like that observation. I think that a lot of it is a matter of categorization and the level of abstraction that you chose. I have some passing knowledge of neuroscience (just the very basics,...
I like that observation. I think that a lot of it is a matter of categorization and the level of abstraction that you chose. I have some passing knowledge of neuroscience (just the very basics, really) and at the right level of abstraction, the human brain also looks like just a bunch of statistical functions.
To be clear, I am not saying the human brain is just that. But I am hoping that new understandings of the human brain will emerge as the by-product of AI research.
It is all just stats. All of the most advanced methods I've heard of amount to playing logistic models off each other in various ways. (I have roughly undergraduate-level experience in both...
It is all just stats. All of the most advanced methods I've heard of amount to playing logistic models off each other in various ways. (I have roughly undergraduate-level experience in both computer science and linguistics)
Most famously you have oppositional setups (incl. GANs) where multiple relatively simple models are designed to disagree or outperform each other in the hopes that the overall accuracy goes up. There are also some newer ideas including the "mixture of experts" where you make a bunch of small specialised models and combine their outputs in some way. These techniques do get better results but from where I'm standing the cutting edge of the field looks a lot like people fiddling with knobs on a synth to me.
I mostly agree with her viewpoints, being quite the pessimist on AI, although I have to say i'm slightly more positive than she is. I do think these tools will/are being used for major advantages...
I mostly agree with her viewpoints, being quite the pessimist on AI, although I have to say i'm slightly more positive than she is. I do think these tools will/are being used for major advantages in the right context, it's just that yes the vast majority is just mashing together nonsense or being overconfident about wrong answers.
The youtube video she references is dumb, but cute, and it's interesting that a creator can throw something like that together now when it would just be WILDLY out of reach for almost anyone. Clearly she wasn't a fan, but I think it's one of the better examples of how AI can quasi lower barriers to entry. It WILL lead to a flood of this kind of stuff, and the novelty is going to wear off fast, but as a proof of concept of what these tools allow, I think it's one of the best examples yet.
Also clearly not enough of a discussion of her TNG Muppet breakdown. Sure she's got a good line on casting, but you can't have a Muppet movie without musical number's and there's a complete lack of brainstorming there.
indeed -- at one hour running time that's enough time for 8000 spoken words. here's the video description AI tools are helpful and cool as long as you know their limitations. AI doesn’t...
indeed -- at one hour running time that's enough time for 8000 spoken words.
here's the video description
AI tools are helpful and cool as long as you know their limitations. AI doesn’t exist.There is no fidelity in AI. AI is build on biased data sets and will give biased results. AI should not be used to make decisions.
That's a good summary of the first half or so of the video, the "AI does not exist" part. There is also the second half, the "but it will ruin everything anyway" part, and in my opinion it's the...
That's a good summary of the first half or so of the video, the "AI does not exist" part. There is also the second half, the "but it will ruin everything anyway" part, and in my opinion it's the more interesting one.
Generally warning about how even though AI doesn't exist (read: AI tools do not do what laypeople think they do), people think AI exists and will make decisions based on it anyway. We're going to see people generating bad AI art and trying to sell it. We're going to see scams built around AI, the same way scams are built around crypto and other "fashionable" tech. We're going to see companies firing employees in favor of AI tools, only to realize later the tool doesn't work as advertised.
A more complete description is:
"AI tools are helpful as long as you know their limitations. AI should not be used to make decisions, but it will be used to make decisions anyway. Try not to be part of the problem, and help educate others not to be. You're going to need to deal with the consequences of people misunderstanding AI.
Pedantry over the definition of "AI" can go deep.
In the classical sense within computer science, AI is both broader and different than the modern vernacular definition. An AI is a rational agent operating within an environment. Notably that describes "what", not "how" (see: slides from my alma mater). Do the algorithms that power Pacman ghosts, just Dijkstra's algorithm, count as AI? Classically, yes! They are absolutely AI. I assure you these are even "stupider" than anything the video presents as "not AI".
Does Stable Diffusion count as AI? No! No it doesn't. Does ChatGPT? Maybe! You'll have to define the reward metrics and the environment, though, for the particular situation.
Then there's the new fangled layman definition, which is not clearly defined, but broadly I generalize it as sufficiently complicated neural networks. Notably, this defines the WHAT not the HOW. These neural networks can be used for anything and fit in this definition.
Now, while I understand the concerns the video author has, trying to add an additional layman definition which seems to make it congruent with AGI, or artificial general intelligence, seems both pointless and only serving to further confusion. Language is natural, so while I resist the new definition in my own language, it is inevitable that people will use the new definition.
Yeah it's incredibly frustrating for me when people who usually aren't familiar with the history of AI get up in arms about using it for anything that isn't AGI. That's not how we've ever used that word! And while I think people have very distorted views of what modern AI can do (especially generative LLMs like ChatGPT), insisting that nothing other than AGI "counts" as AI doesn't actually help dispel misconceptions there. More often than not ime it's used as some weird flex or gotcha when the term "AI" isn't the actual discussion topic.
Funny because when AI was being posted all over Reddit...I said the same shit she did yet was told that I'm full of shit and wrong...even though I work heavily in the field lol. Just a bunch of keyboard warriors telling me that AI does exist. And how AGI is the new word for AI...the public doesn't use AGI...they use AI...glad someone with clout actually is calling this out.
The public thinks and still does think that AI = something that needs no human input and is sentient.
Well that's just anonymous internet. Everybody's an expert and they are never wrong.
Love acollierastro. Great points regarding data bias and the problem it causes. Not a new problem that the tech world has faced in recent years.
Five minutes in and this has a very human-centric viewpoint. The reason presented for why AI is not intelligence is that you can't tell it when it's wrong or have it explain it's thought process. These are both aspects of language, not intelligence: two things which are conflated in this video.
I can't explain to a fly how to avoid fly paper. Even if I went up to a fly and said "hey fly, that sticky paper isn't a fruit you can eat. it's a scented strip of adhesive material designed to trap you", it wouldn't understand. Maybe I could 'train' a new fly to avoid fly paper by breeding a few generations of flies selected to avoid fly paper, which would work really well until it also starts avoiding hanging strips of fruit.
This might not be a universally held belief, but I think flies have some (very small) level of intelligence. So if you want to argue that machine learning models aren't intelligent, I think it would be more illuminating to compare them to flies, rather than people. Not only because they are somewhat alien intelligences, but also because they are much closer in computation power to an insect than to a person.
But that flies in the face of how popular media is reporting on AI, how ai is being talked about by general enthusiasts, and how business grifters are trying to sell AI. Which is what this video is about.
The video is very correct about the fact that as A.I. right now it doesn't not have the same level of reasoning and abstract thinking that is expected out of people.
The video just takes the opposite, similarly unnuanced stance. Obviously we don't have AI as intelligent as humans right now, but it's also a mistake to dismiss all AI as unintelligent. We need to look more closely at the actual capabilities of these systems, rather than using philosophy to dismiss or accept them uncritically, and determine when we want to use them, how much supervision is required, etc. And this is a process that is continuous, not fixed in one point in time, as these systems improve.
Which is exactly what she's doing when she says it's a tool that needs expert supervision to check it's facts and efficacy. Not trying to be a dick, but did you finish the video?
First of all, I'm mostly arguing against the point that AI systems aren't intelligent. I don't think there's a way to say for sure one way or the other (although I lean strongly towards them being intelligent).
My interpretation of what she said was more pessimistic though: My interpretation was that she was saying that machine learning systems will always need expert supervision, and will always be worse than humans, because they are not truly intelligent. Where I disagree is I think there will be a point where AI systems surpass humans, and where human supervision is actually less reliable.
Fine, you can argue against that point that was never actually stated in the video. That's fine.
But at the end of the day saying that the tech will one day be there doesn't justify any of the current AI tech bubble.
AI isn't just chatgpt and this video isn't just talking about chat gpt. Chatgpt is just one of dozens of ai tools that exist or are being built, and there will no doubt be countless more in the future.
A.i. Definitely needs supervision for any use where the a.i. output may be used to make decision(medical diagnosis etc). You're so on the defensive about chat gpt you are missing the point entirely.
The video never seeks demonize a.i. it's just very realistic about the limitations and pushing back against this weird tech bubble nonsense pushing it as a magical solution.
I definitely agree. I work heavily in this field and one thing that I’ve grown to believe is that intelligence is not an emergent property of humans or even life itself, but an emergent property of the universe. Highly proficient language models are often conflated with AGI because as people, language captures almost all of what we are able to experience or conceive.
Using neural networks as an example, they do learn novel, intelligent concepts that were not explicitly taught and that we do no understand. This has been the case for nearly a decade—For instance, image models have been shown to exhibit an understanding of 3D space, even though they have not been given 3D information. Recently, LLMs have been shown to exhibit spatial reasoning to an extent, and can lean on language to explain their conclusions and even arrive at better conclusions for doing so. They are also self reflective without being trained specifically to be reflective.
Disputing the use of the term AI in the video is to highlight a certain point: AI is not human-like intelligence as it is painted as in media and some people genuinely believe it is.
This is the main argument, the public hears AI and thinks sentient being. Academia has all kinds of nuances for AI which is just a generic term used. I don't know why people are trying to argue what academia uses vs the public. The general public thinks AI is the shit from syfi movies not ML or game AI that's dumb as a box of rocks.
Except that's not how AI fear-mongering is being peddled to the masses, is it? The way media outlets go on about it, you'd think SHODAN is just around the corner. No subtlety or nuance, no careful definition of terms, just the use of the term "AI" and tacit allusion to sci-fi horror stories.
Let's please not compare how people who talk about games use "AI" and how the media and tech bubble enthusiasts have been talking about AI. The majority of people who play video games understand that enemy AI is not particularly advanced or anything that closely resembles human intelligence .
Any videogame enemy with behaviour with the slightest sophistication above "blindly charge the player" is touted as groundbreaking AI, and that bar hasn't substantially raised in 20 years.
Didn't F.E.A.R. have some ground breaking "AI" for its time, even though it was just basically a script of if statements lol? I don't think anything has changed from that era. It's, if attacked, find cover and try to kill human player, flank if you have more than one CPU with you, throw grenade, but nothing seriously amazing.
F.E.A.R. is exactly the example I had in mind when I wrote that, would you believe?
Yea F.E.A.R. is that game that I believe everyone at the time was talking about with the AI, the story was good but not ground breaking and neither was the gameplay, it was the AI that had all the articles about it.
As someone learning statistics, the more I learn about machine learning techniques and AI the more I am convinced it’s just fancy statistical modelling with some rebranding.
I’m curious if anyone knows of any techniques that are very dissimilar to standard statistics. Where do you draw the line. Maybe classification is still just stats, maybe a neural network is actually an AI technique and not “just stats”.
As far as I can tell there’s a lot of analogues for example image classification I.e cat vs not cat seems very similar to a logistic regression or at least a logistic activation function and labelling images as cat vs not cat and doing a logit on image parameters might get you a model that can discern images successfully. I can only assume the techniques for AI which I do not know / haven’t gotten to learning yet, do the same kind of thing in a more sophisticated way.
I like that observation. I think that a lot of it is a matter of categorization and the level of abstraction that you chose. I have some passing knowledge of neuroscience (just the very basics, really) and at the right level of abstraction, the human brain also looks like just a bunch of statistical functions.
To be clear, I am not saying the human brain is just that. But I am hoping that new understandings of the human brain will emerge as the by-product of AI research.
It is all just stats. All of the most advanced methods I've heard of amount to playing logistic models off each other in various ways. (I have roughly undergraduate-level experience in both computer science and linguistics)
Most famously you have oppositional setups (incl. GANs) where multiple relatively simple models are designed to disagree or outperform each other in the hopes that the overall accuracy goes up. There are also some newer ideas including the "mixture of experts" where you make a bunch of small specialised models and combine their outputs in some way. These techniques do get better results but from where I'm standing the cutting edge of the field looks a lot like people fiddling with knobs on a synth to me.
I mostly agree with her viewpoints, being quite the pessimist on AI, although I have to say i'm slightly more positive than she is. I do think these tools will/are being used for major advantages in the right context, it's just that yes the vast majority is just mashing together nonsense or being overconfident about wrong answers.
The youtube video she references is dumb, but cute, and it's interesting that a creator can throw something like that together now when it would just be WILDLY out of reach for almost anyone. Clearly she wasn't a fan, but I think it's one of the better examples of how AI can quasi lower barriers to entry. It WILL lead to a flood of this kind of stuff, and the novelty is going to wear off fast, but as a proof of concept of what these tools allow, I think it's one of the best examples yet.
Also clearly not enough of a discussion of her TNG Muppet breakdown. Sure she's got a good line on casting, but you can't have a Muppet movie without musical number's and there's a complete lack of brainstorming there.
Some kind of summary of why I should bother watching this video would be helpful.
indeed -- at one hour running time that's enough time for 8000 spoken words.
here's the video description
AI tools are helpful and cool as long as you know their limitations. AI doesn’t exist.There is no fidelity in AI. AI is build on biased data sets and will give biased results. AI should not be used to make decisions.
That's a good summary of the first half or so of the video, the "AI does not exist" part. There is also the second half, the "but it will ruin everything anyway" part, and in my opinion it's the more interesting one.
Generally warning about how even though AI doesn't exist (read: AI tools do not do what laypeople think they do), people think AI exists and will make decisions based on it anyway. We're going to see people generating bad AI art and trying to sell it. We're going to see scams built around AI, the same way scams are built around crypto and other "fashionable" tech. We're going to see companies firing employees in favor of AI tools, only to realize later the tool doesn't work as advertised.
A more complete description is:
"AI tools are helpful as long as you know their limitations. AI should not be used to make decisions, but it will be used to make decisions anyway. Try not to be part of the problem, and help educate others not to be. You're going to need to deal with the consequences of people misunderstanding AI.