Philosophy Stack Exchange is a question and answer site for those interested in the study of the fundamental nature of knowledge, reality, and existence. It only takes a minute to sign up.
Sign up to join this community
I really want to say rejecting a line of reasoning because ChatGPT created it would be an ad machina argument.
(Note, I'm interested in the case where the rejection is made without any consideration of the validity, or lack their of, of the rejected material.)
See genetic fallacy.
In brief:
This fallacy avoids the argument by shifting focus onto something's or someone's origins. It's similar to an ad hominem fallacy in that it leverages existing negative perceptions to make someone's argument look bad, without actually presenting a case for why the argument itself lacks merit.
It is fallacious to make the formal argument that a conclusion is false because of its provenance.
It not fallacious to dismiss an argument out of hand because it comes from a source which is well known to almost always make specious, dishonest, or nonsensical arguments. It's not even an argument, but it is a good idea.
It is not fallacious to make the argument: "Because this argument comes from a source which is well known to almost always make specious, dishonest, or nonsensical arguments, it is probably specious, dishonest, or nonsensical."
It is not fallacious to make the argument: "I cannot always tell when an argument is specious, dishonest, or nonsensical. This argument comes from a source which is well known to almost always make specious, dishonest, or nonsensical arguments. Therefore even though I cannot tell that it is specious, dishonest, or nonsensical, there is a significant probability that it might be."
It is not fallacious to dislike and discourage the making of specious, dishonest, or nonsensical arguments, whether or not they're plagiarized. It's not even an argument, but it is a good idea.
It is not fallacious to dislike and discourage plagiarism. It's not even an argument, but it is a good idea.
Are we really dealing with an argument if the text is generated using a method that does not involve any kind of reasoning or understanding about the subject?
ChatGPT only generates sequences of lexical symbols (that is, letters and punctuation) based on very complex set of rules that have been tuned to generate new sequences that have same features as those in the input corpus. Even if ChatGPT does it so well that it mostly looks like what a human, sometimes even an expert could have written, it is still nothing but sequence of lexical symbols.
Without any actual reasoning or understanding, everything in the response is completely, totally superficial: it only looks like a proper response, but it is not. That is why the responses are so often nonsense in some way or another.
I argue that such text does not provide an argument. ChatGPT is just a very advanced symbol sequence generator even if the output looks like an argument. It looks like that because the whole point of ChatGPT is to mimic what people have written.
It is more likely that assuming that such text contains an argument worth attacking is the actual fallacy here.
The problem with the philosophical ideal of judging every argument on its merits is that a human lifetime is not long enough to do it, by very many orders of magnitude. Like it or not, you will reject (i.e., ignore) the vast majority of human output based not on intrinsic merit but other criteria.
Any time you start to read someone's writings, decide they are mentally ill, and leave the rest unread, you're rejecting their other arguments based only on (what you believe to be) a property of the person who made them. Any time you tell someone that that person's work is not worth their time, you are for practical purposes arguing ad hominem. I find it impossible to see that as wrong, because it's unavoidable.
Nothing in that argument changes if you replace Gene Ray by ChatGPT. I've read enough of ChatGPT's writings to form the opinion that nothing it writes is worth reading (except as humor, or as a window into its "mind"). I would tell others the same, and it's up to them to decide whether they trust my opinion.
I do not dismiss ChatGPT arguments because I know them to be invalid. I decide to not bother to engage with them because experience has shown them to usually not be worth the effort. It is an argument from efficiency.
And while ad hominen is a recognized fallacy, there is another, less recognized fallacy: that any being in the universe has unlimited time and/or energy to faff around with the firehose of nonsense some people or programs can generate. Maybe call it "the fallacy of omnipotence"?
I understand that ignoring the time and energy required to deal with an argument is convenient when modeling thought, and lets us get at other interesting questions, but in reality nobody sane lives that way. And any philosophy that out-right refuses to deal with that fact will in some sense be fundamentally unhuman.
It's obviously a fallacy to dismiss an argument without reason just because you distrust the source. Like dismissing the argument of a known liar might be a good heuristic in terms of where not to waste your time, but it's ultimately a fallacy as they are still capable of telling the truth and might even do so even if it's rarely done.
The more interesting question with regards to AI generated text is rather who is the author and what is the argument. Because so far the AI doesn't do logic so if there are followup questions you're almost inevitably required to steelman the argument.
Yes, either an ad hominem or the genetic fallacy that Futilitarian pointed out, depending on your reasoning for ignoring the arguement. If you argue against the virtues of a chat bot (e.g. it hallucinates) then I would say it is ad hominem.
To understand this its best to think about what exactly ChatGPT is doing, its been trained on lots of writing, including arguments, to learn how to respond to inputs. A very simple form of this would be if I got a bunch of experts (or not so experts) to write a bunch of arguments to question X, then put them in a hat, mix them around and draw one of them as my response. If you argued that you can ignored the arguments because I drew them from a hat then you are committing a fallacy (especially if we only used experts)
We can make this closer to ChatGPT by splitting arguments into parts adding labels of what kinds of arguments can follow this part (e.g. all As have B, can be followed by arguments of type: "X is A" or "B implies C"), and what kinds of arguments this follows (e.g. requires A to be defined). Then if I randomly draw an argument part that assumes nothing (or only that I'm trying to answer a particular type of question), I can then follow this by collecting all parts that could be the next argument parts and therefore is something that would follow the first thing I drew from my hat. Do this a number of times and I have an argument that maybe none of the experts proposed, but built out of valid parts that experts have considered. Again, ignoring this argument because it came from drawing your answers from a had would lead to a fallacy.
If you are still thinking that this is a random process and has no understanding, it is entirely possible for a human to make a valid argument based on repeating what they have heard experts say. Sometimes mixing up similar arguments from different experts. This means you cannot immediately ignore randomly generated arguments because they came from a machine. If you say you ignore all arguments where the speaker has ignorance, then you forget that even experts don't lack ignorance (they just have a lot less than the average person).
Overall if you have an argument generated by ChatGPT, it is valid if someone can read that argument and can champion it as true, and it is invalid if someone can read that argument and say it is wrong because of these reasons. The main reasoning towards automatically ignoring it is opportunity cost to validate that it didn't just make convincing garbage, but that is a different question to the one you asked.
That chat-gpt created it, I suppose not actually, though obviously it depends on context (it can be cheating). But becasue chat-gpt agrees with it? Absolutely 100% convinced we should not pass over out critical faculties to robots.
Is attacking an argument because it is machine generated an ad hominem fallacy?
No. People use calculators to show that 7 plus 5 equals 12, but no one believes that equation is true because a calculator said so. The fallacy in that example would be an appeal to authority. Here, the source of truth is not the calculator, but rather in decades of human thought, starting with Bertrand Russel’s proof that 1 plus 1 equals 2.
Similarly, someone can not argue that maybe, just maybe, 7 plus 5 does not equal 12 simply because the source of the answer is a calculator. Such an argument would simply apply the inversion of the appeal to authority, though just as fallacious.
The "machine-generated" problem is just the calculator problem writ large. The proponent cannot say they are right because artificial intelligence wrote the argument, as that would be a 21st-century version of the appeal to authority.
The suspicions surrounding artificial intelligence arise from its track record, not from something inherent in the technology. Examples include instances where the machine just made stuff up. Such instances provide a legitimate basis for challenging a machine-generated argument.
Short answer
No, it is not an ad hominem attack to dismiss "an argument" generated by ChatGPT. This is because only a little investigation will show that any "text" generated by ChatGPT cannot contain any arguments at all. Ever. To think otherwise is to massively misunderstand what language is and how it works. Asking this question is like asking:
Is dimissing an argument because it's machine generated an ad hominem fallacy if the argument is not an argument but a bonobo monkey?
Full answer:
It is a mistake of epic proportions to assign an argument to something which contains no argument. Take, for example,a set of symbols which has been generated to look like human speech.
It would be absolutely fallacious, immoral in fact, to say of such a text that it contained an argument which is either valid or sound. This is because what you would be looking at is not evidence of the intent of a speaker to influence a listener or reader's stock of assumptions in some way. This will not be immediately clear without an example.
Let's consider the following piece of supposedly machine generated text:
The man punched John. Then he punched him - harder. So they were both violent.
We cannot consider this argument to be either sound or valid. The reason is that it means nothing. It means nothing because there was nobody at the creating end of this argument who had an intention to influence a listener. There is no creator who is using a combination of context and shared knowledge to make sure that the intended interpretation of their utterance influences the listener's stock of assumptions in the intended way.
For a start, who is the referent of the word John meant to be? John Lennon, my friend John from the pub? Your uncle? We don't know. You can assign a referent to the word John if you wish. But you cannot argue the case, or make out that your assignment is correct. You might say, "Oh, but we can say it refers to some man called John". But, erm, no you can't. It could easily be retorted that it refers to an orangutang, or a tiger and there's nothing we can meaningfully say or do to counteract this.
I can hear someone say: "Well, we can nonetheless say that the argument is valid, even if we don't know who the word John refers to and whether they actually hit anyone" But, erm, no we can't. Not nearly. The reason is that we cannot know who the words he and him are meant to represent. That's because they're not meant to represent anything or anyone. And that's because there was no communicator meaning them to be interpreted in any particular way. You could interpret he to mean John, and the word him to mean the man. But, so what? Someone else can say that the assignment should be the other way round. You could argue the point, but it is a meaningless argument. You cannot clarify who the pronouns refer to. There was nobody intending them to refer to anyone. Furthermore, I have a female friend called John. And so could I easily say that in this case neither he nor him refer to John. And I could further argue that therefore the argument is not valid because I can argue that the word both refers to John and the man, and John has not been shown to be violent. But that would be ridiculous drivel too!
As nobody can definitively define any referents to any of the seeming referring expressions in the 'text' that was generated, one cannot truthfully say that it has any meaning or that it is valid or invalid or sound or not sound. To assign such a meaning to it would not only be idiotic it would be mendacious and any assessment of the text as representing a valid argument would be equally fallacious.
The words that speakers use to communicate are only meaningful because of both the communicative intention of the speaker and the successful interpretation of that intention by the listener. The language in any given utterance is normally hugely underdetermined as well as being multiply semantically and syntactically ambiguous. Language itself it not a code produced by an enigma machine for which a listener has a key. It is just an archeological trail, which given the context and shared knowledge can lead a listener to a correct interpretation of the speaker's intentions.
No speaker, no communication. No communication, no argument. You wouldn't assign validity to a camel, or soundness to a cloud. You can't assign an argument to a jumble of ChatGPT generated symbols designed to to look like an artefact of human communication.
I have just asked a chatbot, Falcon. I tried 3 times and got 3 different answers: no it isn't; yes it is; no it isn't an ad hominem fallacy, but it is an ad hominem attack. All three answers were supported (not necessarily strongly) by reasons.
I think the OP's question turns on the exact definition of ab hominem. For example:
I think it is fair to say that the Chatbot argues inconsistently. I don't think that it is really an ad hominem to point out that a person has a track record of lying and arguing both side of the question on different days. If a clever lawyer argues the case two different ways, we may know that they can't both be right, even if we can't spot the precise fallacy. Maybe this is what Falcon "meant" the third time: an attack, but not a fallacy?
To add to the collection of answers I'd say ... technically ... an ad hominem (attempts to) discredit an argument by attacking the person (literal translation) which in 9 outta 10 cases takes the form of maligning a person's moral character e.g. that the arguer has a police record, that the arguer beats his dog, that the arguer is adulterous, etc. That is to say "the argument is no good because the arguer is bad"
I don't know if an ad machina fits into this mold perfectly. Can a machine be bad??
