From time to time, I see a tool to present a discussion as a tree with arguments for and against it.
Unless it is a school essay, arguments don't go that way.
It is usually harder to encompass what a node (an atomic fact) is and what a link is (it usually goes beyond "support" and "counter"). Very often, this structure is not a tree. Maybe a DAG with weighted edges, but if it were that straightforward - knowledge graphs would simply work.
Instead of rehashing the same tree approach, we should adopt something closer to an LLM-embedding approach - for a given statement, we should have "relevant statements" with an additional dimension if it supports, counters, expands, provides an example, and so on. In this case, it wouldn't even be a DAG.
Concur, I discovered UMAP when looking for a way to dimension reduce and visualize embeddings, and it also works on non-embedded data too. Interesting idea to think about it applied to arguments in a debate... especially in conjunction with the work around using LLMs to infer knowledge graphs
> Uniform Manifold Approximation and Projection (UMAP) is a dimension reduction technique that can be used for visualisation similarly to t-SNE, but also for general non-linear dimension reduction. The algorithm is founded on three assumptions about the data
1. The data is uniformly distributed on Riemannian manifold;
2. The Riemannian metric is locally constant (or can be approximated as such);
I've experimented with constructing arguments as actual DAGs before here: http://concludia.org/ If you are strict about logical force and premises leading to lemmas and conclusions, I think it works pretty well.
There's a lamport paper lying around somewhere that also talks about representing arguments and proofs as DAGs.
I think it would be fun to include a framework where you could use logical fallacies to support common arguments that include them and ask people who support those arguments to replace them with non-fallacious arguments as a more nuanced way of debating and getting around that all criticism of popular arguments is somehow a strawman.
Unless it is a school essay, arguments don't go that way.
It is usually harder to encompass what a node (an atomic fact) is and what a link is (it usually goes beyond "support" and "counter"). Very often, this structure is not a tree. Maybe a DAG with weighted edges, but if it were that straightforward - knowledge graphs would simply work.
Instead of rehashing the same tree approach, we should adopt something closer to an LLM-embedding approach - for a given statement, we should have "relevant statements" with an additional dimension if it supports, counters, expands, provides an example, and so on. In this case, it wouldn't even be a DAG.