Definition / requirements
Similar to the claim evals, these requirements are use to give a score to prospective claims. However, rather than scoring the strength, clarity, or utility of claims, these requirements score text on whether their validity as claims.
- Claims must not contain conjunctions, including “and”, “but”, and “or”
- Claims must not must not contain “because” or “until”
- Claims must not commas
Claim extraction algorithm
Claim extraction is going to be one of the primary algorithms in Wicker. It takes arbitrary text as input and outputs a suggestion for how the text could be decomposed into claims. Part of its function can be to check input text for the definitional requirements of a claim.
Splitting on ‘and’
It is tempting to split claims on ‘and’, e.g., “he loves chocolate and blueberry juice” → “he loves chocolate” + “he loves blueberry juice” since each of these could be debated separately. There are several considerations here. First, as in this example, the meaning might be such that the sentence cannot be split, e.g., if it meant “he loves the combination of chocolate and blueberry juice”. Second, this relates to the question of adjectives and variants, e.g., “he lived in the big, red barn” → “he lived in the barn” + “the barn was big” + “the barn was red”. The point is just that even when you don’t use the word ‘and’, you can still be listing things that are part of a conjunction. You could have said “the barn is big and red, and he lived in it”, in which case it would’ve been more clear.
Hidden relational claims
Sometimes a “support” or similar relational claim will seem to be superfluous. Consider 01 | It is unlikely that UBI will be implemented before 2028 as critiqued by 02 | Advancements in AI will necessitate UBI. Since 02 alludes to the implementation UBI directly, a relational claim such as 03 | 02 critiques 01 seems less necessary. This has something to do with how 02 is phrased — if it had said 12 | There will be significant advancements in AI by 2028, then the relational claim would certainly be required to fill in the details (Toulmin warrant) about job loss and how these things might connect.
Are these two claims?
Is 02 actually making 2 claims? No. Though it implies both that AI will advance significantly, which could be its own claim, and that these advancements will necessitate UBI, that’s not actually what it’s saying. You can tell because you couldn’t actually separate the two claims out. Well, maybe this is wrong. You could split it into something like 12 | There will be advancements in AI by 2028 and 13 | 12 will necessitate UBI by 2028. You could theoretically want to attack either of these in order to critique 02.
The claim builder
See The claim builder.
Claim shortnames
In the claim builder and elsewhere, it is desirable to reference claims without having to include their full text. Therefore, a method of generating shortnames could be devised. In its most basic instantiation, you could imagine an LLM coming up with a two-word name for any claim. Then, wherever the claim is referenced, its shortname could be used by default along with a little “expand” widget that reveals and hides the full claim text.
The power of emojis
Adding emojis to claim shortnames could be quite powerful since emojis are very information dense. They can sometimes represent subtle meanings and sentiments in the space of a single character. They may also endear new users to an otherwise potentially daunting platform by communicating “humanness” and “chillness”.