Regulators in Australia, Denmark, the U.S., and the U.K., among others, are exploring potential age restrictions, which, if enacted, would also eat into the user counts of Meta’s apps.
I know it'll never happen and still eat into FB's profits. But if countries are going to be this insistent on age-gating the internet, they need to just be like Asia and tie these kinds of...
I know it'll never happen and still eat into FB's profits. But if countries are going to be this insistent on age-gating the internet, they need to just be like Asia and tie these kinds of accounts to some national ID verification. That's the only way you're properly going to enforce age restrictions.
AI is a term that doesn’t mean anything. This is the description Bloomberg had Given that it’s tabular data, they’ll probably use boosted trees or something. What is wasteful in that?
AI is a term that doesn’t mean anything. This is the description Bloomberg had
With a proprietary software tool it calls an “adult classifier,” Meta will categorize users into two age brackets — older or younger than 18 — based on the person’s own account data. The software can sift through a user’s profile, see their follower list and what content they interact with, and will even scan unsuspecting “happy birthday” posts made by friends to predict a user’s age.
Given that it’s tabular data, they’ll probably use boosted trees or something. What is wasteful in that?
The fun thing about this demographic is that a lot of them are naive and you can get them to out themselves. See this streamer for example who did a fake PS5 giveaway for anyone "born after <date...
The fun thing about this demographic is that a lot of them are naive and you can get them to out themselves.
Though possibly not the people who will inevitably be incorrectly flagged as kids, as Facebook has always done over the years. I'm all for anything that frustrates people away from these social...
Of course, kids will find a way around these as well.
Though possibly not the people who will inevitably be incorrectly flagged as kids, as Facebook has always done over the years. I'm all for anything that frustrates people away from these social networks, though, no matter their age.
Offtopic. I'm really really annoyed by the shift to replace "algorithm" with "AI" to drive clicks. I get it. AI is the cool hype thing. The problem is when even linear regressions get labelled AI...
Offtopic.
I'm really really annoyed by the shift to replace "algorithm" with "AI" to drive clicks. I get it. AI is the cool hype thing. The problem is when even linear regressions get labelled AI cause that makes the line go up.
To be fair, that was a blatant misuse of "algorithm" too. Lots of benign things are algorithms. Sorting posts by date, the thing everyone wants instead of an algorithm, requires a sorting algorithm.
To be fair, that was a blatant misuse of "algorithm" too. Lots of benign things are algorithms. Sorting posts by date, the thing everyone wants instead of an algorithm, requires a sorting algorithm.
I think AI is overused as a buzzword, but I think insisting people stop calling even other ML models "AI" is a silly battle. People have been referring to ML models, including those as simple as...
I think AI is overused as a buzzword, but I think insisting people stop calling even other ML models "AI" is a silly battle. People have been referring to ML models, including those as simple as linear regression, as AI for a long time. The term has existed to describe the academic discipline since the 50s and a ton of very important early AI research was using things that even I would just call "algorithms" nowadays. As for linear regression, machine learning has been part of AI as a discipline since before the term "machine learning" was invented. Insisting that machine learning techniques aren't real AI is a fundamentally ahistorical and very recent point of contention.
Granted, people do slap the word "AI" on literally anything they can even vaguely find a connection with because it's a marketing buzzword, so I get the annoyance. But it can be frustrating when people overcorrect and start complaining about things that we were pretty uncontroversially calling AI prior to ChatGPT coming out.
That’s not true. AI’s classical academic definition is the study of rational agents in environments. I would note that this is a description of “what”, not “how”. A Pac-Man bot using DFS to path...
That’s not true. AI’s classical academic definition is the study of rational agents in environments.
I would note that this is a description of “what”, not “how”. A Pac-Man bot using DFS to path to berries? That’s AI, because it involves an agent that is attempting to be rational in an environment.
A neural network classifying hotdogs or not? Not an AI, because there is no agent, no reward metric, or an environment.
A Pac-Man bot using a neural network? That is an AI, once again we have an agent, rewards, and an environment.
Machine learning is a term describing “how” - its class of parameterized posterior probability models fitted with data. AI is about “what”. Behind the hood can be anything. It’s apples and oranges.
The idea that a classifier is less of a rational agent in an environment than a Pac-Man bit is just does not jive with how I've remotely seen the terminology used within academic environments. To...
The idea that a classifier is less of a rational agent in an environment than a Pac-Man bit is just does not jive with how I've remotely seen the terminology used within academic environments. To be clear, I've seen people refer to the Pac-Man bot type thing as AI even in the modern day, but I don't think your definitions of "rational agent" and "environment" are coherent if they exclude classification tasks.
I have never heard of a classifier being considered a context where there’s a rational agents and an environment. What would be the environment? The utility function?
I have never heard of a classifier being considered a context where there’s a rational agents and an environment. What would be the environment? The utility function?
I'd presume it would be related to the domain of the classifier, which of course can vary wildly. For an example, my master's thesis involved classification of entities in images -- classifying...
I'd presume it would be related to the domain of the classifier, which of course can vary wildly. For an example, my master's thesis involved classification of entities in images -- classifying whether the entity in this bounding box is a dog or a boy or a chair or whatever. I don't really understand how one could define "rational agent" in a way that excludes this but includes a Pac-Man pathfinding algorithm. As for the environment, surely in this case that's the input images (and accompanying bounding boxes and other pieces of data), just as the environment for the Pac-Man bot would be the maze itself?
There's some key differences there. In Pacman, environment evolves with time and action. If you move forward next to a wall at time t0, moving forward is no longer an option at time t1. It's also...
There's some key differences there. In Pacman, environment evolves with time and action. If you move forward next to a wall at time t0, moving forward is no longer an option at time t1. It's also state in that way.
The input images to a bounding box detector does not evolve with time. It's a simple pure function mapping a state onto a vector; it doesn't take in a state and evolve it.
I'm not sure there's really a meaningful difference between Pac-Man's location relative to a wall and an image's having been placed in a category or remaining uncategorized. I don't think it makes...
I'm not sure there's really a meaningful difference between Pac-Man's location relative to a wall and an image's having been placed in a category or remaining uncategorized. I don't think it makes sense to claim that the former is having a greater effect on the environment, and I don't think there's a coherent definition of environment that excludes the latter, at least not in the context of how AI is discussed by those who work on it academically. I'm even more certain that there's no coherent definition of a "rational agent" that excludes a classifier but includes a Pac-Man pathfinding algorithm. In general, I don't think your definition of AI as you describe it gels with what I'm familiar with from the field as a whole. It seems to be rather idiosyncratic.
Er, what? Do you mean like the... earth's environment? The difference is twofold; one is that an environment has time as a dimension, and you can model it either as mutating state or as pure state...
I don't think it makes sense to claim that the former is having a greater effect on the environment
Er, what? Do you mean like the... earth's environment?
The difference is twofold; one is that an environment has time as a dimension, and you can model it either as mutating state or as pure state that is recursively observed by the agent and then evolved by a state-evolution function.
Second, is that the agent's actions at time tx affect the state of the environment at time t(x+1).
I think we'll have to agree to disagree. This is just going to be circular from here on out. That is the definition when I was an undergrad.
I think the rational agent idea is part of the definition of AI, but my impression is that the academic definition tends to be broader than that because the definition of "intelligence" is not...
I think the rational agent idea is part of the definition of AI, but my impression is that the academic definition tends to be broader than that because the definition of "intelligence" is not really a settled matter.
From Artificial Intelligence by Russell and Norvig:
Historically, researchers have pursued several different versions of AI. Some have defined intelligence in terms of fidelity to human performance, while others prefer an abstract, formal definition of intelligence called rationality—loosely speaking, doing the “right thing.” The subject matter itself also varies: some consider intelligence to be a property of internal thought processes and reasoning, while others focus on intelligent behavior, an external characterization. From these two dimensions—human vs. rational and thought vs. behavior—there are four possible combinations ...
Acting humanly: The Turing test approach
Thinking humanly: The cognitive modeling approach
Thinking rationally: The “laws of thought” approach
Acting rationally: The rational agent approach
I think your definition would be the fourth category here, while a classifier for whether or not someone is 18+ would be the third category.
On the broader topic (not aimed at you), I think it's interesting how so many people seem to have fairly strong convictions on what is and is not AI. The discussion on whether something is or is not AI seems to come up on half the posts on the internet that mention AI. I personally don't have strong convictions on the subject because I don't think a truly well-accepted definition exists, but I enjoy following along with the discussions to see how opinions are trending. I wouldn't tell anyone that they're wrong about any definition they are proposing, just that academic definitions can be pretty broad.
This seems to be such a recent phenomenon to me too, which is what baffles me most. I don't think I saw this type of argumentation often at all before ChatGPT took off. People in academia used it...
I think it's interesting how so many people seem to have fairly strong convictions on what is and is not AI. The discussion on whether something is or is not AI seems to come up on half the posts on the internet that mention AI. I personally don't have strong convictions on the subject because I don't think a truly well-accepted definition exists, but I enjoy following along with the discussions to see how opinions are trending. I wouldn't tell anyone that they're wrong about any definition they are proposing, just that academic definitions can be pretty broad.
This seems to be such a recent phenomenon to me too, which is what baffles me most. I don't think I saw this type of argumentation often at all before ChatGPT took off. People in academia used it broadly, as you describe (and that's more or less what I attempted to argue in my initial comment, just that it's used quite broadly in academia and has been for some time), and people in specific fields would use it to describe relevant bits and pieces that were relevant to them (for instance, how the term "AI" gets used in gaming and game design). I find how quickly and suddenly people started arguing about what counts as "really" AI so bizarre.
I don't know why you don't like 5 to 9D cinemas. Seriously though - If it doesn't have AI (written on it), it doesn't exist. At least that is what marketing people think. For me it's not useful...
I don't know why you don't like 5 to 9D cinemas. Seriously though - If it doesn't have AI (written on it), it doesn't exist. At least that is what marketing people think. For me it's not useful pointer because those people made the term useless by slapping it on everything. How many dimensions can you get in a cinema? Where is the end to it?
Is that really something that needs to be done?
Unfortunately, yes. Meta as corporation should do 'something' to 'protect' kids.
Yes.
I know it'll never happen and still eat into FB's profits. But if countries are going to be this insistent on age-gating the internet, they need to just be like Asia and tie these kinds of accounts to some national ID verification. That's the only way you're properly going to enforce age restrictions.
Apologies, I missed a word, is this necessary to do with ai? It's such a wasteful tool.
AI is a term that doesn’t mean anything. This is the description Bloomberg had
Given that it’s tabular data, they’ll probably use boosted trees or something. What is wasteful in that?
The fun thing about this demographic is that a lot of them are naive and you can get them to out themselves.
See this streamer for example who did a fake PS5 giveaway for anyone "born after <date 18 years ago>" (sorry for Twitter link) https://x.com/rosedoodles/status/1832836638788964587
A tale as old as time. It was pretty easy during the 2000s and 2010s if you had a squeaky voice, anyways.
Though possibly not the people who will inevitably be incorrectly flagged as kids, as Facebook has always done over the years. I'm all for anything that frustrates people away from these social networks, though, no matter their age.
Offtopic.
I'm really really annoyed by the shift to replace "algorithm" with "AI" to drive clicks. I get it. AI is the cool hype thing. The problem is when even linear regressions get labelled AI cause that makes the line go up.
To be fair, that was a blatant misuse of "algorithm" too. Lots of benign things are algorithms. Sorting posts by date, the thing everyone wants instead of an algorithm, requires a sorting algorithm.
I think AI is overused as a buzzword, but I think insisting people stop calling even other ML models "AI" is a silly battle. People have been referring to ML models, including those as simple as linear regression, as AI for a long time. The term has existed to describe the academic discipline since the 50s and a ton of very important early AI research was using things that even I would just call "algorithms" nowadays. As for linear regression, machine learning has been part of AI as a discipline since before the term "machine learning" was invented. Insisting that machine learning techniques aren't real AI is a fundamentally ahistorical and very recent point of contention.
Granted, people do slap the word "AI" on literally anything they can even vaguely find a connection with because it's a marketing buzzword, so I get the annoyance. But it can be frustrating when people overcorrect and start complaining about things that we were pretty uncontroversially calling AI prior to ChatGPT coming out.
That’s not true. AI’s classical academic definition is the study of rational agents in environments.
I would note that this is a description of “what”, not “how”. A Pac-Man bot using DFS to path to berries? That’s AI, because it involves an agent that is attempting to be rational in an environment.
A neural network classifying hotdogs or not? Not an AI, because there is no agent, no reward metric, or an environment.
A Pac-Man bot using a neural network? That is an AI, once again we have an agent, rewards, and an environment.
Machine learning is a term describing “how” - its class of parameterized posterior probability models fitted with data. AI is about “what”. Behind the hood can be anything. It’s apples and oranges.
That is, classically. Now it just has no meaning.
The idea that a classifier is less of a rational agent in an environment than a Pac-Man bit is just does not jive with how I've remotely seen the terminology used within academic environments. To be clear, I've seen people refer to the Pac-Man bot type thing as AI even in the modern day, but I don't think your definitions of "rational agent" and "environment" are coherent if they exclude classification tasks.
I have never heard of a classifier being considered a context where there’s a rational agents and an environment. What would be the environment? The utility function?
I'd presume it would be related to the domain of the classifier, which of course can vary wildly. For an example, my master's thesis involved classification of entities in images -- classifying whether the entity in this bounding box is a dog or a boy or a chair or whatever. I don't really understand how one could define "rational agent" in a way that excludes this but includes a Pac-Man pathfinding algorithm. As for the environment, surely in this case that's the input images (and accompanying bounding boxes and other pieces of data), just as the environment for the Pac-Man bot would be the maze itself?
There's some key differences there. In Pacman, environment evolves with time and action. If you move forward next to a wall at time t0, moving forward is no longer an option at time t1. It's also state in that way.
The input images to a bounding box detector does not evolve with time. It's a simple pure function mapping a state onto a vector; it doesn't take in a state and evolve it.
I'm not sure there's really a meaningful difference between Pac-Man's location relative to a wall and an image's having been placed in a category or remaining uncategorized. I don't think it makes sense to claim that the former is having a greater effect on the environment, and I don't think there's a coherent definition of environment that excludes the latter, at least not in the context of how AI is discussed by those who work on it academically. I'm even more certain that there's no coherent definition of a "rational agent" that excludes a classifier but includes a Pac-Man pathfinding algorithm. In general, I don't think your definition of AI as you describe it gels with what I'm familiar with from the field as a whole. It seems to be rather idiosyncratic.
Er, what? Do you mean like the... earth's environment?
The difference is twofold; one is that an environment has time as a dimension, and you can model it either as mutating state or as pure state that is recursively observed by the agent and then evolved by a state-evolution function.
Second, is that the agent's actions at time tx affect the state of the environment at time t(x+1).
I think we'll have to agree to disagree. This is just going to be circular from here on out. That is the definition when I was an undergrad.
I think the rational agent idea is part of the definition of AI, but my impression is that the academic definition tends to be broader than that because the definition of "intelligence" is not really a settled matter.
From Artificial Intelligence by Russell and Norvig:
I think your definition would be the fourth category here, while a classifier for whether or not someone is 18+ would be the third category.
On the broader topic (not aimed at you), I think it's interesting how so many people seem to have fairly strong convictions on what is and is not AI. The discussion on whether something is or is not AI seems to come up on half the posts on the internet that mention AI. I personally don't have strong convictions on the subject because I don't think a truly well-accepted definition exists, but I enjoy following along with the discussions to see how opinions are trending. I wouldn't tell anyone that they're wrong about any definition they are proposing, just that academic definitions can be pretty broad.
This seems to be such a recent phenomenon to me too, which is what baffles me most. I don't think I saw this type of argumentation often at all before ChatGPT took off. People in academia used it broadly, as you describe (and that's more or less what I attempted to argue in my initial comment, just that it's used quite broadly in academia and has been for some time), and people in specific fields would use it to describe relevant bits and pieces that were relevant to them (for instance, how the term "AI" gets used in gaming and game design). I find how quickly and suddenly people started arguing about what counts as "really" AI so bizarre.
I don't know why you don't like 5 to 9D cinemas. Seriously though - If it doesn't have AI (written on it), it doesn't exist. At least that is what marketing people think. For me it's not useful pointer because those people made the term useless by slapping it on everything. How many dimensions can you get in a cinema? Where is the end to it?