Applying logical techniques to actual arguments
This discussion was created from comments split from: Can you recommend a textbook on logic?.
Post edited by Sascha on
Howdy, Stranger!
Categories
- 3K All Categories
- 152 Research & Reading
- 692 The Zettelkasten Method
- 7 Knowledge Work
- 100 Writing
- 464 Software & Gadgets
- 154 Workflows
- 730 The Archive
- 15 Plug-In Showcase
- 88 Resolved Issues
- 225 Projects Logs and Journals
- 83 Project: Zettelkasten.de
- 53 Critique my Zettel
- 171 Random
- 373 Introduce Yourselves!
Comments
For those interested in applying logic techniques to actual arguments, take a look at this post, which I just came across today:
New Publication: Identifying Flawed Reasoning in Contrarian Claims about Climate Change
(For context: p3 is a "hidden premise", an assumption that is not communicated while presenting the argument)
I didn't read the paper. So, the authors might need a bit more supporting reasoning to arrive at this hidden premise.
But this hidden premise looks quite fallaciously injected. The argument is perfectly consistent with the claim that greenhouse gases have a significant impact. It might be that other factors have a greater influence. Or: Greenhouse gases might on their own have the biggest influence, but the interaction of other factors leads to a different temperature direction.
If someone has the full text, I'd like to take a closer look.
I am a Zettler
The post has a this link to a pre-press version (which I haven't read except for skimming the beginning):
https://skepticalscience.com/docs/Flack_2026_deconstruction.pdf
I don't want to argue either climate science or whether each instance in the post is perfectly correct. The idea is the approach, not the details. However, in the one case you quote, the premise P1 ("current climate change is driven by natural factors") can only be true if the effect of human factors (greenhouse gases) can be ignored (P3). That qualification is not included in the argument. If P3 could be ignored then the argument might reach a true conclusion. Since P3 has not been shown to be correct, P1 cannot be shown to have a truth value of True, and so the conclusion has not been demonstrated to have a truth value of True.
Yes. We will not discuss climate science because this is (sadly) politically loaded and therefore against the forum's rules.
The goal of my interjection is to zero in on an example to practice the actual argument analysis.
Is it the assumption that greenhouse gases are just caused by people? If CO2 is a greenhouse gas, then greenhouse gases are not only introduced by humans.
I'd sharpen up the hidden premise p3 to p3':
Civilisation's influence on climate is negligible. (or something like that)
The primary weakness of p3' is that it merely contradicts the premise of the other side of the argument.
But the notion that there is a hidden premise is quite weak because the negation of civilisation's contribution strength is obvious.
I am a Zettler
I find P3 also interesting (I replace the topic-specific terms with variables):
"G doesn't have much of an X effect".
The authors of the above mentioned paper define on page 4: "Deconstruction involves assessing the quality of reasoning in misleading arguments, identifying logical fallacies."
Here's how I'd deconstruct P3:
P3 is not a valid premise because it's incomplete. The premise introduces a quantitative value: the strength of an effect. What is missing is the connection between the quantitative value and the truth value. How does the quantitative value determine the truth value of the premise?
When a paper about logic-checking claims to reveal "hidden premises", I'd expect those premises to be valid.
Maybe like this?
Phrased this way, we can get into a much more nuanced debate. For example:
@tomp said:
I think I found the catch in the post and the referenced paper (Flack, 2026).
We don't see "actual arguments" in the paper. :-(
The claims in the table aren't actual claims, but a classification of claims (Coan et el (2021), https://doi.org/10.1038/s41598-021-01714-4).
The arguments in the table are the result of a methodology that started out with a selection of actual arguments ("examplars" from a previous dataset, p. 13), but then "sought to identify the most representative argument for each taxonomy claim" (p. 16).
The process for finding hidden premises is less clear. Are they the result from evaluating examplars? Or have they been constructed from the most representative arguments?
The goal of those hidden premises is clear (p. 16, emphasis added):
I'd like to try it with our example:
P1 and P2 mention natural factors. But as the main debate is about the relevance of other factors, we'd need to include those as well in the argument."Greenhouse gases" doesn't deliver, because they are also emitted by natural processes. As far as I know a common term is “anthropogenic factors”.
P1 claims a causal relationship that ignores anthropogenic factors. In order to make it work, we could add that a) current climate change is driven only by natural factors or b) that anthropogenic factors currently play only a minor role.
I think this hidden premise should make the argument work:
Same idea as Sascha's, but a different way to get there. :-)
Looking for hidden premises is a good idea, even if the paper doesn't provide the best examples.
The main point is that any of those other, non-stated premises may or may not be true but since none of them has been shown to be true, then the conclusion of the original argument has not been shown to be true.
The original argument was basically "If (a and b) then c; a is True and b is True therefore c is True". All the alternatives mentioned above show that a's truth value is not known to be True, the if premise has not been shown to be adequate, or both. So the conclusion that c is True is not warranted.
At any rate, I posted the link because I thought it had a direct tie-in to one of the topics that arose in the original thread, namely, a deeper study of Logic.
From my point of view, the main value of this kind of study is to learn how to recognize fallacies and whether the premises are adequate to support the claimed conclusions, were their truth values to be established. These goals do not require a great depth of study. Beyond that, in real life the variables involved are rarely Boolean: they usually do not have simple True/False values. We need to make progress in the face of that situation.
I agree. :-) Thanks for posting!
Another aspect I found interesting are the names of the claims in the taxonomy.
How do you choose a title or filename for a zettel about a proposition? For example, "heading into ice age" is too short to make sense on its own, but it works nicely as a shortcut for the predictive claim "we're heading into an ice age, no matter what humans do".
And it's a nice example of a hierarchical outline, where a short title like "4.3 too hard" works because we know the parent claim.
Yes.
That's why I liked the distinction between fact-checking and logical checking in the paper.
Yes. Fortunately, many useful tools and technical terms exist already. But they require training. I find the textbooks interesting but challenging.
Have you come across the concept of "fuzzy logic"? It's often used in a context of classification. For example, is a man whose height is 5 ft 10 in (178 cm) "tall"? In a fuzzy logic approach, one sets up what are essentially templates, e.g., "short", "medium" "tall", "very tall" and evaluates the given height against each template. The template with the highest matching score is considered the best fit, and the score indicates how good the fit is. The approach replaces boolean reasoning. So it can be applied to situations in which boolean reasoning is classically used, such as using fuzzy instead of Boolean unions and disjunctions.
The process is not at all the same as Bayesian statistics.
One aspect of fuzzy logic is especially interesting to me. Bart Kosko showed that the parameters of any artificial neural net (ANN) can be converted to a set of fuzzy logic rules: each of them (ANN and fuzzy controllers) is a universal function approximator, The fuzzy rule set has a much lower dimensionality than the ANN. This is especially interesting these days because a large language model is an ANN. IOW, a fuzzy controller backed by a rule set derived from a trained LLM should be able to perform exactly the same way at a much lower cost in computer resources.
Yeah> @tomp said:
I'm aware of fuzzy logic where linguistic variables like "tall" are vague, but the function that maps them to truth values is well determined. I haven't heard of "templates" in this context.