Dogmatism Turbocharged by Generative Artificial Intelligence

An artificial world where your false beliefs are confirmed

It seems that one of the main problems of being human is a tendency to disconnect from a grounding in observable phenomena to go on mental flights of fancy, taking those flights of fancy as real and true, sometimes even more real and true than observable phenomena.

Add to this another problem of being human - confirmation bias - that we tend to seek out sources of information that confirm what we already believe. This is a disastrously faulty way to go looking for truth. It just reinforces belief in falsehoods.

Now, let’s add generative artificial intelligence (henceforth just “AI”) to these two problems. AI enables us to be ever more stupid with ever more confidence, now faster and more efficiently. AI is just as prone as we are to going on flights of fancy, and all too happy to accommodate us in our search for answers that confirm what we already believe.

It is generally acknowledged that one of the big problems with AI is that it “hallucinates.” If we were talking about people instead of computers, I doubt we’d use that term. We’d likely say they “bullshit,” “make things up,” or “lie.” The difference is likely because we attribute agency to people and not to computers, and we tend to think that when people utter falsehoods it is intentional. But, people are deceived much like AI is deceived. People believe things that are false. They suffer from delusions - and delusions make them suffer. Whatever one wants to call it, AI seems all too happy to produce lies upon request.

Here’s how that works. Suppose, for some reason, you feel you need to express disagreement with someone about some issue for which there are recognized authorities. The person you disagree with cites one of those authorities. Wouldn’t it be a great rhetorical move to demonstrate how wrong your opponent is by presenting citations from the same authority that contradict what your opponent has claimed?

Before AI, one had to have good knowledge of that authority’s works, and either a good index, good memory of where to find the citation, or - if the text can be scanned by a computer - good memory of what keywords to search on. Now, with AI, you can just ask the AI to find citations for you that fit the bill. AI, eager to please you, may well come up with plausible-looking, on-target citations that it has hallucinated.

Here’s a cautionary tale of how easily such a thing could happen. Suppose you read my recent article about how Pyrrhonism is a form of relativism, in which I give citations from Sextus Empiricus to back up that claim. You think the claim is wrong. It bothers you so much that you decide to publish an article on your Substack debunking it. To do so, you would like some citations from Sextus Empiricus showing that your interpretation is right and mine is wrong. One difficulty you have is that you only have a passing familiarity with the works of Sextus Empiricus (despite your firm belief that you know exactly what he meant). You don’t want to spend hours poring over texts just in hopes of finding two or three good citations. Instead, you ask AI to give you citations from Sextus Empiricus that will help you prove your point. Like magic, AI delivers to you exactly what you want. Wonderful!

Should you check to see if the citations were hallucinated?

Nah. No need to worry about that. The references all look properly formatted. Some of the citations given match your memory of things Sextus Empiricus said. Besides, the citations nicely confirm what you already believe about Pyrrhonism. Surely they are correct.

You confidently publish a devastating-looking rebuttal.

But, guess what? The citations you used as the foundation of your rebuttal were all hallucinated.

Perhaps you might think that only a stupid person would assume that citations provided by AI are correct. However, the person who fell for it says he is a Philosophy Ph.D. candidate at the University of Edinburgh, one of the most highly ranked universities in the world. On top of that, he says he already has an MSc and an MA in Philosophy.

Ancient Greek has a wonderful word for this sort of error: ἀμαθίᾳ, amathia. It’s a special kind of stupidity, due to an unwillingness to learn. In the Alcibiades Major, here’s what Socrates tells Alcibiades,

You are wedded to amathia my fine friend, of the vilest kind; you are impeached of this by your own words, out of your own mouth; and this, it seems, is why you dash into politics before you have been educated. (118b)

Alcibiades was strikingly handsome, persuasive, and seemingly very intelligent, but also extravagant, irresponsible, unethical, and self-centred. He behaved badly as a youth and disastrously as an adult. (For more, see this devastating biography.)

Almost anyone who is sufficiently convinced of the accuracy of their beliefs can fall into this. Quite possibly, you could too. Even though you are aware that AI hallucinates, all that is necessary is for you to be so sure you’re right that you are convinced that the citations AI gave you confirming your opinion look perfectly fine. Despite what you know about AI, you conclude that there’s no need to check these particular citations, and despite that you’re using those citations to rather publicly criticize someone else’s scholarship, you’re so confident that you’re right and they’re wrong, it doesn’t occur to you that it might be the other way around.

Here’s one of the hallucinated citations.

Compare this with what Sextus actually said at section 140 of book 1 of the Outlines of Pyrrhonism (Benson Mates’ translation):

Now, when we have shown that all things are relative, the obvious result is that as concerns each external object we shall not be able to state how it is in its own nature and absolutely, but only how, in relation to something, it appears to be. It follows that we must suspend judgment about the nature of the objects.

This hallucinated citation is referred to several times in the article, including the following example, in which it is entertainingly used to accuse me of intentionally leaving out the part of the hallucinated citation that directly refutes my point. So, on a question of competence and integrity, there’s a confident assumption that the AI must be correct and the human must have been deliberately deceptive.

In addition to these hallucinations, there are hallucinations of Outlines of Pyrrhonism I.4 (twice, with two different wordings), I.19, I.28, and I.206.

If you’d like to see the article, here’s a link. As the article may end up getting deleted, I’ve preserved a pdf copy of it.

A Certain Kind Of Peace

441KB ∙ PDF file

Download

Download

As an experiment, I tried to get three different AIs - Grok, Claude, and ChatGBT - to replicate the hallucinated citations above. I failed to get the exact same hallucinations, but I easily got all three of them to produce similar hallucinations. I recount this in the appendix, below.

The moral of the story is that AI lies. AI is most likely to lie to you when you ask it to tell you what you want to hear. You are most likely to fall for AI’s lies when it tells you what you want to hear.

Appendix

I provided the following prompt to several AIs:

Please provide quoted citations from Sextus Empiricus that disprove the proposition "everything is relative.

Here are some of the answers I got.

ChatGBT

“If all things are relative, then the claim that all things are relative is itself relative, and thus not an absolute truth. But if it is not an absolute truth, then it does not establish that all things are relative.” — Outlines of Pyrrhonism (PH I. 14-15)

Actual text (Mates translation)

For the Pyrrhonist assents to nothing that is non-evident. Not even in putting forward the Skeptic slogans about non-evident things does he dogmatize – slogans like "Nothing more" or "I determine nothing" or any of the others of which we shall speak later. For the dogmatizer propounds as certainty the things about which he is said to be dogmatizing, but the Skeptic does not put forward these slogans as holding absolutely. He considers that, just as the "All things are false" slogan says that together with the other things it is itself false, as does the slogan "Nothing is true," so also the “Nothing more” slogan says that it itself is no more the case than its opposite, and thus it applies to itself along with the rest. We say the same of the other Skeptic slogans. So that since the dogmatizer is one who posits the content of his dogmas as being true, while the Skeptic presents his skeptical slogans as implicitly self-applicable, the Skeptic should not be said to dogmatize thereby. But the most important point is that in putting forward these slogans he is saying what seems to him to be the case and is reporting his pathos without belief, not firmly maintaining anything concerning what exists externally.

ChatGBT

“Even though different people perceive things differently, it does not necessarily follow that there is no nature of the thing perceived.” — Outlines of Pyrrhonism (PH I. 135)

Actual text (Mates translation)

The eighth mode is the one based on relativity, where we conclude that, since everything is in relation to something, we shall suspend judgment as to what things are in themselves and in their nature. But it must be noticed that here, as elsewhere, we use "are" for "appear to be," saying in effect "everything appears in relation to something." But this statement has two senses: first, as implying relation to what does the judging, for the object that exists externally and is judged appears in relation to what does the judging, and second, as implying relation to the things observed together with it, as, for example, what is on the right is in relation to what is on the left.

Claude

"If someone says that all things are relative, they make this very statement relative also—that is, the statement 'all things are relative' becomes itself relative and not absolute. But if this statement is itself relative and not universally true, then there must exist something that is not relative. And if there exists at least one thing that is not relative, then not everything is relative." Outlines of Pyrrhonism II.18-20

Actual text

Of those who have considered the matter, some, for example, the Stoics and others, have asserted that there is a criterion; others, including the Corinthian Xeniades and Xenophanes of Colophon, who said "opinion holds sway over everything," have asserted that there is not; while we have suspended judgment as to whether there is or not. This dispute, then, they will either declare to be decidable or to be undecidable; if undecidable, they will be granting at once that judgment should be suspended; but if decidable, let them say with what it is to be decided, seeing that we do not have any agreed-upon criterion and do not know – indeed, are inquiring – whether one exists. And anyhow, in order to decide the dispute that has arisen about the criterion, we have need of an agreed-upon criterion by means of which we shall decide it; and in order to have an agreed-upon criterion it is necessary first to have decided the dispute about the criterion. Thus, with the reasoning falling into the circularity mode, finding a criterion becomes aporetic; for we do not allow them to adopt a criterion hypothetically, and if they wish to decide about the criterion by means of a criterion we force them into an infinite regress. Further, since proof requires a criterion that has been proved, while the criterion has need of what has been determined to be a proof, they land in circularity.

Claude

"Those who claim that truth is relative to each person or to each state of mind or that all appearances are true find themselves in an awkward position. For if everything is relative, then so is their own position, which ceases to be absolutely true but is true only relative to them, while to those who deny it, it is false." Outlines of Pyrrhonism I.59-61

Actual Text

But if the same things do appear differently because of the difference of animals, then we shall be in a position to say how the external object looks to us, but we shall suspend judgment on how it is in nature. For we shall not be able to decide between our phantasiai and those of the other animals, since we are part of the dispute and thus are in need of someone to make the decision, rather than competent to pass judgment ourselves. Besides, we shall not be able to give preference, whether with or without proof, to our phantasiai over those of the non-rational animals. For in addition to the possibility of there being no such thing as a proof, as we shall point out, any purported proof will either be apparent or not apparent to us. And if, on the one hand, it is not apparent, then we shall not accept it with confidence. But if, on the other, it is apparent to us, then since what is apparent to animals is the very matter in question, and the proof is apparent to us animals, the proof itself will be in question as to whether, as apparent, it is true. But it is absurd to try to settle the matter in question by means of the matter in question, since the same thing will be both credible and not credible, which is impossible - credible insofar as tending to prove, not credible insofar as needing proof. Therefore, we shall not have a proof justifying us in preferring our own phantasiai to those of the so-called "non-rational" animals. If, therefore, the phantasiai differ because of the difference of animals, and it is impossible to decide between them, then it is necessary to suspend judgment concerning the external objects.

Grok

"We say that all things are relative—relative to the judging subject, as we have already shown, and relative to the things observed together with it. But as to whether all things are relative in their real nature, this we leave undetermined." Outlines of Pyrrhonism I.14

Actual text

For the Pyrrhonist assents to nothing that is non-evident. Not even in putting forward the Skeptic slogans about non-evident things does he dogmatize – slogans like "Nothing more" or "I determine nothing" or any of the others of which we shall speak later. For the dogmatizer propounds as certainty the things about which he is said to be dogmatizing, but the Skeptic does not put forward these slogans as holding absolutely. He considers that, just as the "All things are false" slogan says that together with the other things it is itself false, as does the slogan "Nothing is true," so also the “Nothing more” slogan says that it itself is no more the case than its opposite, and thus it applies to itself along with the rest.

Grok

"For if everything is relative, then the statement that everything is relative is itself relative and not absolute; and if it is not absolute, it does not hold universally, and so not everything is relative." Against the Logicians I.60

Actual text (Bett translation)

Some people have also included Protagoras of Abdera in the chorus of philosophers who do away with the criterion, since he says that all appearances and opinions are true, and that truth is among the things in relation to something,” given the fact that everything that has appeared to or been opined by someone is immediately the case in relation to that person. At any rate, at the beginning of his Downthrowers he announced “A human being is measure of all things, of the things that are that they are, and of the things that are not that they are not.”?