I actually do think that, to a decent extent, I understand what AI is. And while this is a technicality, it really grinds my gears when a GPT model is compared to an autocomplete/predictive text. Yes, they both technically just predict text using statistical models, but it’s like comparing a modern jet to a paper airplane, because they both can fly.
[ChatGPT] has no idea what its words mean
Doesn’t it tho? It has an internal model of the world that it constructed by reading and processing tons of text. It knows that an apple is round-ish, comes in certain colors, can be eaten or grow into a tree. That knowledge is very limited due to the model’s inability to experience such things as shape or color; like a blind person knows the description of “red”, but doesn’t actually know what it is.
Of course, it’s debatable whether what a GPT model does can be considered “understanding”, but then again, we don’t really know what understanding IS, but I would argue it’s extremely close to what a human understanding is, albeit in a limited scope.
That being said, I think the discussion of how advanced (or not) our modern AI systems are, though interesting, is extraneous to the question at hand. The main question is “what is AI?”. From your comment, I can conclude that your definition of AI relies on the subject’s possession of sentience/consciousness. I think this is a flawed approach because a bee, while undoubtedly possessing rudimentary intelligence, in all likelihood, lacks consciousness. So consciousness should not be a qualifying criteria for an AI. Furthermore, I looked up several dictionaries for definitions of “AI”, and they all boil down to “man-made machines that perform human tasks”, here’s are some:
Cambridge Dictionary – “computer technology that allows something to be done in a way that is similar to the way a human would do it”
Merriam-Webster – “the capability of a machine to imitate intelligent human behavior”.
In conclusion, intelligence comes in all shapes and sizes; the only thing differentiating natural intelligence from artificial intelligence is the origin, i.e., if it was man-made, it’s artificial. By that definition, perhaps outdated and lacking insight, Data most definitely possesses AI. Not to mention the lack of full-fledged “sentience” as he can’t experience feelings.
This seems to have descended into a debate on “what is consciousness”, which as I originally said, is a question that isn’t easy to answer. My point was that modern AI inherrently isn’t aware of what it’s saying, not that it couldn’t be defined as an intelligence. As far as I know, there’s no solid evidence to prove that it can. To finish, I would like to apologise if my initial comment came across as condescending. I didn’t mean to come across as such.
This seems to have descended into a debate on “what is consciousness”
I disagree, while I did go on a tangent there with analyzing ChatGPT capabilities, my ultimate argument was that we shouldn’t even be discussing the consciousness topic at all. When deciding whether Data has AI or natural intelligence we only need to look at the source of his intelligence; it was man-made, therefore any painting Data produces is “AI art”, because Data only has AI, despite having capabilities on par or even exceeding those of a human.
To be honest, I did take it as being a little condescending, but it doesn’t really matter. All I wish is to have a discussion, and expand our knowledge in the process.
Thank you then! It seems like our debate stemmed from different definitions. Based on your definition of what constitutes AI, Data would absolutely count. By my definition, he is too advanced to be in the same category. But I get the impression that we would both agree that he is more advanced than any modern AI system. Once again, I’m sorry for coming across as condescending; I will have to choose my words more carefully in the future!
What are attention mechanisms of not being aware of what it has said so it can inform what it is about to say? Ultimately, I think people saying these generative models aren’t really “intelligent” boils down to deciding they don’t like the impact these things are having and are going to have on our society and characterizing them as a fancy statistical curve lets people short circuit that much harder conversation.
I actually do think that, to a decent extent, I understand what AI is. And while this is a technicality, it really grinds my gears when a GPT model is compared to an autocomplete/predictive text. Yes, they both technically just predict text using statistical models, but it’s like comparing a modern jet to a paper airplane, because they both can fly.
Doesn’t it tho? It has an internal model of the world that it constructed by reading and processing tons of text. It knows that an apple is round-ish, comes in certain colors, can be eaten or grow into a tree. That knowledge is very limited due to the model’s inability to experience such things as shape or color; like a blind person knows the description of “red”, but doesn’t actually know what it is.
Of course, it’s debatable whether what a GPT model does can be considered “understanding”, but then again, we don’t really know what understanding IS, but I would argue it’s extremely close to what a human understanding is, albeit in a limited scope.
That being said, I think the discussion of how advanced (or not) our modern AI systems are, though interesting, is extraneous to the question at hand. The main question is “what is AI?”. From your comment, I can conclude that your definition of AI relies on the subject’s possession of sentience/consciousness. I think this is a flawed approach because a bee, while undoubtedly possessing rudimentary intelligence, in all likelihood, lacks consciousness. So consciousness should not be a qualifying criteria for an AI. Furthermore, I looked up several dictionaries for definitions of “AI”, and they all boil down to “man-made machines that perform human tasks”, here’s are some:
In conclusion, intelligence comes in all shapes and sizes; the only thing differentiating natural intelligence from artificial intelligence is the origin, i.e., if it was man-made, it’s artificial. By that definition, perhaps outdated and lacking insight, Data most definitely possesses AI. Not to mention the lack of full-fledged “sentience” as he can’t experience feelings.
This seems to have descended into a debate on “what is consciousness”, which as I originally said, is a question that isn’t easy to answer. My point was that modern AI inherrently isn’t aware of what it’s saying, not that it couldn’t be defined as an intelligence. As far as I know, there’s no solid evidence to prove that it can. To finish, I would like to apologise if my initial comment came across as condescending. I didn’t mean to come across as such.
I disagree, while I did go on a tangent there with analyzing ChatGPT capabilities, my ultimate argument was that we shouldn’t even be discussing the consciousness topic at all. When deciding whether Data has AI or natural intelligence we only need to look at the source of his intelligence; it was man-made, therefore any painting Data produces is “AI art”, because Data only has AI, despite having capabilities on par or even exceeding those of a human.
To be honest, I did take it as being a little condescending, but it doesn’t really matter. All I wish is to have a discussion, and expand our knowledge in the process.
Thank you then! It seems like our debate stemmed from different definitions. Based on your definition of what constitutes AI, Data would absolutely count. By my definition, he is too advanced to be in the same category. But I get the impression that we would both agree that he is more advanced than any modern AI system. Once again, I’m sorry for coming across as condescending; I will have to choose my words more carefully in the future!
What are attention mechanisms of not being aware of what it has said so it can inform what it is about to say? Ultimately, I think people saying these generative models aren’t really “intelligent” boils down to deciding they don’t like the impact these things are having and are going to have on our society and characterizing them as a fancy statistical curve lets people short circuit that much harder conversation.