Example Debate with AI Researcher about Popper
This is a fictional dialog which stands on its own but also follows my article Error Correction and AI Alignment. The main goal isn’t realism; it’s to illustrate some of Critical Fallibilism’s thinking about rationality and debate.
The Debate Begins
Elliot Temple: Do you agree that your conclusions about AIs are premised on Karl Popper being wrong?
AI Researcher: Yes.
ET: So will you debate Popper’s arguments?
AR: Of course.
ET: Have you read much Popper?
AR: Some.
ET: Do you believe you personally read enough to reach accurate conclusions about Popper?
I’ll now split the debate into separate branches based on the answer to this question.
Branch 1
ET: Do you believe you personally read enough to reach accurate conclusions about Popper?
AR: Yes.
ET: Have you read Popper’s replies to his critics? (Volume 2.)
AR: No.
ET: OK. I wanted to ask if you found any errors in Popper’s replies or if you had an argument that he hadn’t already covered in that book.
AR: Well, I haven’t read it.
ET: So then why do you think you’ve personally read enough to reach accurate conclusions about this?
AR: Well, I read two of his books. They were long. That’s a lot.
ET: If someone read two books related to AI Alignment and then wrote a refutation, and that was the most and best attention AI Alignment ever got from critics, would you be satisfied?
AR: No.
ET: Do you see my point?
AR: I guess so.
ET: So let’s continue on the basis that you haven’t personally read enough Popper to properly evaluate it.
Which brings us to:
Branch 2
ET: Do you believe you personally read enough to reach accurate conclusions about Popper?
AR: No.
ET: So you’re relying on secondary sources? There are some other thinkers who did read enough Popper to reach accurate conclusions, who judged him negatively, who you are relying on?
AR: Yes.
ET: OK. I suggest we start with the topic of whether induction is a useful thinking method that humans use. Are you or your sources familiar with Popper’s claim to have refuted induction?
AR: Yes.
ET: Great, to begin with, can you provide your single best source which refutes Popper?
AR: [link]
ET: Reviewing this, I don’t see any quotes of Popper followed by explanations of errors in Popper’s text.
AR: Yeah, it just explains why induction is necessary.
ET: Do you have any source which directly addresses Popper’s arguments and reasoning?
AR: I’ve got great sources about induction.
ET: Popper explained why that thinking is wrong. You need to give a rebuttal.
AR: Where did Popper explain that?
ET: Wait, so actually you aren’t familiar with Popper from personal reading or from reading any secondary sources that present and then refute Popper’s ideas?
AR: I didn’t say that.
ET: If you were ignorant of Popper and therefore neutral about his ideas, that would be OK and in that context I could help you learn about Popper. But if you’re claiming to have already reached a conclusion about Popper being wrong, then it’s inappropriate to ask me what Popper’s main points were or where he wrote them.
AR: Can we stop doing so much meta discussion?
ET: I was trying to. I wanted to get topical arguments from you. I’ve been asking for refutations of Popper.
AR: I’d rather talk about AI alignment. I have lots of great points to say about that.
ET: Do you understand that I disagree with some of your premises, so talking about your higher level ideas based on those conclusions doesn’t make sense?
AR: Maybe if I tell you about how dangerous AIs are, you’ll find it persuasive.
ET: If you don’t tell me how your AI-danger claims are compatible with non-inductivist epistemology, or address my concerns and convince that induction can work, then I don’t see how you could persuade to me.
AR: Well, it’s not my job to persuade everyone. I can give arguments that persuade many people, and if you aren’t persuaded, that’s OK, you can’t expect me to persuade more than e.g. 80% of people even if I’m right. So persuading you personally doesn’t matter.
ET: So, in short, you can’t and won’t debate me or Popper, and you don’t know of any intellectuals who wrote down arguments that address our points?
AR: You’re so picky and meta; no wonder no one will debate you. That seems like a pattern with you, so maybe you should consider that it’s a you-problem.
ET: Do you honestly think that’s a rational argument?
AR: How can I reason with someone like you. This is a waste of time. Bye.
Branch 3
Let’s back up and continue in a different way from an earlier part.
ET: Great, to begin with, can you provide your single best source which refutes Popper?
AR: [link]
ET: OK, I see a few arguments here. But I’ve got a question. Did the author read Popper’s replies to his critics?
AR: I’m not sure.
ET: I think it’d be good to present arguments as either pointing out an error in Popper’s replies or making a new point that Popper’s critics in that book didn’t make. Does that make sense?
AR: Sure.
ET: So do you know of any Popper critic who ever did that?
AR: No.
ET: So doesn’t that indicate that no one has done a good job of attempting to refute Popper?
AR: Does anyone ever live up to your very high standards?
ET: Yes. Popper replied to his critics. I also liked Hazlitt’s book-length refutation of Keynes, but I’ve been unable to find any rebuttal to it from Keynesians. Regardless, do you have an explanation of what lower standards I should have and how/why they’d be adequate to resolve disagreements correctly? Do you have a written methodology to offer, which is easier but good enough?
AR: You can read books like Yudkowsky’s Rationality: From AI to Zombies and try to do techniques from it.
ET: I’ve read that. It’s nothing like a specific method, a flowchart, a specific set of standards, etc. It’s a bunch of scattered ideas and techniques. Which is fine. Overall I enjoyed the book. It has some useful tips in it. But it doesn’t provide clear methods for reaching conclusions about ideas. It doesn’t have anything like an organized, comprehensive, complete approach or solution.
AR: I don’t think what you want exists. Philosophy just isn’t that developed.
ET: Well, I’ve been working on it (building on Popper) and I think I’ve got solutions which are adequately developed for people to start using. I’ve been disappointed by the lack of interest in rational debate methodology by others. Since you have no alternative solutions, why don’t you try out my methods or critique them?
AR: Learning your ideas seems like too much work, especially since they build on Popper so I’d have to learn more about his ideas too. So I’ll just content myself with no solutions to rationality and go back to AI research.
ET: Got any colleagues who want to study rationality ideas which try to solve problems you lack solutions for?
AR: Nah, they’ll all estimate you’re unlikely to be right and not want to risk wasting their time.
ET: Want to try my method of reading until the first error?
AR: No, that isn’t already a standard, mainstream method.
ET: Got written, objective criteria for how to estimate what is worth the time or not?
AR: Nah, we all just try to do our best.
ET: So if you’re biased, you’ll fail?
AR: We also try our best not to be biased.
ET: You’re basically trusting your rationality, and if you’re not as good as you think, your approach is bad. And this is all avoidable. You could do things like check out my ideas which provide answers that your own literature doesn’t have.
AR: If you want more readers and attention, just write good things and you’ll get popular.
ET: You think truth reliably leads to popularity?
AR: Better ideas are way more likely to get popular. Instead of expecting any individuals to debate you or evaluate your ideas, you should just let humanity as a whole collectively judge your work by how much social status and popularity it gets you.
ET: You want me to become popular while being ignored by people like you until after I’m popular?
AR: Other people seem to manage that.
ET: Usually by social climbing. Sometimes by luck.
AR: If you’re so smart, then beat them at social climbing.
ET: That’s not a rational, truth-seeking attitude.
AR: I don’t see any way to win with you.
ET: Yeah, I agree, I think you lost…
AR: Rude! Bye!
Conclusion
One of the reasons it’s hard to debate with people is they don’t believe in making decisive arguments about each specific issue. Instead, they believe in adding up how strong the weight of evidence and arguments is for each side (minus some points for criticism). In that approach to epistemology, you just have to (arbitrarily) estimate that your side is stronger than Popper’s, not actually quote specific things Popper said and decisively refute them. Everything is about how much you like each side, not about reaching conclusions, issue by issue, until you untangle things. They basically think reaching clear conclusions is too hard and have given up on it, so they generally aren’t even trying to analyze and engage with specific text from critics.