Comment by md224 | Hacker News Reader

Comment by md224 | Hacker News Reader

Interesting experiment. Here's a thought: what if users could break down arguments into smaller "atomic" propositions (yes, this is somewhat inspired by the now defunct philosophy of Logical Atomism).

For example, taken from an actual debate on Saysaw about policing cyber-bullying:

"In theory, yes - but in practice, no. How on earth would we judge what an acceptable thing to say or post actually is? Do we have a mean-tweet checklist? Besides, if people aren't being mean in one place, they'll do it in another - that's the internet for you. People should just learn that their words, in whichever forum or situation they are said, have the power to upset or offend people."

This person's argument seems to consist of a few assertions, themselves debatable (in the literal sense of "can be debated on Saysaw"). These include:

1) We cannot have a reliable method for determining the "acceptability" of content

2) Censoring a bully in one channel of communication will lead them to continue bullying in another

These assumptions could be spun out into their own Saysaw debates, and these debates could be linked to other arguments on the site that use them as a premise. This would provide a rough way to estimate the plausibility of an argument: how much are its premises supported?

This may also have the effect of helping people to drill down to the core of a disagreement. A lot of debates are unproductive because people don't seem to understand what they're actually disagreeing about. Providing an avenue for breaking an argument down into its components may make it easier to locate the actual point(s) of contention.

Anyway, this is just brainstorming... not necessarily a workable system but something to think about.

I agree, if you're going to automate debate, you really need to have your software comprehend what an argument is. I have a design in my head of such a system that I've been wanting to develop for years but have never put it at the front of my priorities. Last time I was pondering implementing it, I looked around online and was disappointed to see that there were about 5 attempts to do structured arguing as a web service, but they were all FAR too informal and unstructured. I don't want a place to chat and type out paragraphs of pros. I want some hardcore step-by-step argumentation! Might as well talk about it, though, as it's interesting.

I'd go back to basics. Take Logic by W C Salmon (http://www.ditext.com/salmon/logic.html) and treat is a user spec. What you want, I think, is a way to structure arguments and catalog them like a wiki. So you have these things that people can interact with:

1. A premise

2. A conclusion

3. An argument

If I'll be allowed to model the data with Haskell data types to give a more formal description:

data Claim = Claim Text

data Premise = Axiom Claim | Conclusion Argument

data Argument = Argument [Premise] Claim

In more verbose English: an Argument is a list of Premises and a final conclusion which is some plain-text Claim. Claims are either facts or axioms that are taken as read. A Premise is either an Axiomatic claim or a Conclusion from another argument.

Only one conclusion can be drawn in an argument at once. If you want to make compound arguments that draw multiple conclusions, that's where the duplicity of a Premise comes in. It is either a plain-text claim, or a reference to another argument's conclusion that has been made earlier.

So putting that into practice:

humans_mortal_claim = Claim "All humans are mortal."

men_humans_claim = Claim "All men are humans."

men_mortality_argument = Argument [humans_mortal_claim ,men_humans_claim] (Claim "All men are mortal.")

Given those claims and argument I can construct a new argument:

plato_manhood_claim = Claim "Plato is a man."

plato_mortality_argument = Argument [plato_manhood_claim ,men_mortality_argument] (Claim "Plato is mortal.")

It's very simple, but it models the argument perfectly -- BUT, with the flexibility of simple English. You don't have to learn first-order logic here or try to parse the English into some formal representation (although I am currently working on such a library for Haskell: http://chrisdone.com/posts/attempto-controlled-english).

Now, that's the data model. The UI would be to represent arguments as squares on the screen consisting of a list of premises that you can add to and a final conclusion. When you add a premise, you can either pick from a catalogue of existing axioms/facts, or choose from existing conclusions of other arguments. The conclusion is a plain text field where you make an original, new claim, derived from the premises.

But how do you make use of it and deal with real-world contention? So, when someone proposes an argument, I have a set of UI features to help me accept or refuse it. First, I have to go through each of the premises one by one and accept them. If they refer to conclusions of arguments I've already reviewed and accepted before, then they will be pre-accepted. Finally, I decide at the end whether THE PREMISES SUPPORT THE CONCLUSION, and NOT, whether I agree with the conclusion as-is (e.g. I can argue that the moon is made of cheese with valid argumentation, even though it is known to be false).

So you decide that, indeed, the premises support the conclusion, so you mark the argument as valid. HOWEVER, the argument is ONLY accepted if you marked all the premises as accepted. Because, as Salmon clearly states, an argument's conclusion is only considered true if all the premises that support it are true. Indeed, the opposite case may be true: you accept the conclusion, but the argument is invalid. This is the difference between argumentative validity and existential truth.

I believe dealing with both, and being explicit about it, is very handy. In one case pointing them out in the negative is helpful to avoid error. In the other case, saying "I agree with the conclusion even though the argument is invalid to me" or "I don't agree with the conclusion/one of the premises, but the argument itself is valid" is a good way to give encouragement to opponents, that you're not wholesale saying "you're just WRONG". You're saying "this part is not quite acceptable to me, but this part is okay, we just need to work on this bit before I can accept it."

So it may be and should regularly be the case that e.g.

1. I accept all the premises in this argument, but the argument is invalid or incomplete.

2. I accept none or only some of the premises, but the argument itself is valid.

3. I accept none of the premises, but the conclusion I accept.

Now, let's consider the cases of decisions made by a reviewer:

1. Argument is valid, conclusion accepted: this is perfect. This argument can now be considered part of the accepting user(s) knowledge-base.

2. Argument is valid, one or more premises not accepted: more work is needed. The user needs to contest one of the premises and start a new, separate argument, but linked to this premise, to address/resolve it. Some first-class support could be available, like a button to say "Please cite this premise with a source". And the user would get a notification saying "someone wants a citation", or whatever.

3. Argument is invalid: the user needs to start a separate argument to resolve it. One nice feature of the UI could be to select from a list of formal fallacies which the argument exhibits, each of which would have links describing what the fallacy entails. This would help users arguing sincerely to overcome fallacies.

4. Argument is valid, premises accepted, conclusion not accepted: the user has made a mistake in assessing the premises and the argument. You can't have an argument in which all the (accepted) premises support the conclusion, and yet disagree with the conclusion, if you're being sincere. There could be some helpful links about argument structure here.

People would just submit arguments and axiomatic claims into the ether, and other people can come along and accept them or reject them as part of their own personal knowledge-base. I think you would have to distinguish between public and personal "knowledge". It would be like Github, you fork people's arguments/claims by accepting them.

I'm not sure that this would be as awesome as I think it would be, because it'd require a great deal of sincerity. Just thinking about using it rules out a lot of contentious debates which are in essence people expressing their feelings rather than trying to reach the truth. However, it would certainly be an efficient way to nail down precisely where you disagree with someone, because you'd be able to point it out, in the UI. This here, this precise premise I do not accept. And either you both could split it off and argue that premise or just leave cancel the whole debate. Ultimately, it could be a nice knowledge-base. And a way for people who normally wouldn't know the first thing about arguing effectively to, with the AID of the system, develop an intuition for the structure of a decent, honest argument. I'd really try to downplay any "voting" aspect that gives credence to popularity. That would be my dream.

Sorry for spamming your face off. I'll keep this comment for later if I eventually go ahead and implement this.

Interesting ideas - definitely going to give this more thought. Saysaw wasn't initially intended to be such a formal process as this would alienate the majority of users (I can imagine my mum trying to use StackOverflow), but theres definitely value in something along the lines of what you described. It would be a completely different audience & business model though, probably similar to stackoverflow's, and would probably take a bit more than a week to build :-) Email info@saysaw.org if you want to discuss more

Wow, please do go ahead and make it happen :) I'd love to contribute to such a project.

This gets into something that I think derails a lot of debate when dealing with people of faith. Most people, including religious people, try to argue as if both sides of the argument are starting with the same assumptions, when they're really not. If they accept the same premises for a specific argument, then sure, they can have a meaningful debate about it, but if they don't, and they very frequently don't when it comes to debates involving people of different faiths (including atheism for the sake of semantic convenience), then the debate will just be a shouting match. This idea is often referred to as presuppositionalism, that people have fundamentally different, and frequently irreconcilable, ways of looking at the world, and any argument where the arguers have different presuppositions is therefore futile.

This is relevant, because it seems that the system you propose would make presuppositions very explicit, which would help people with different presuppositions to realize why they are disagreeing instead of just yelling at each other till they go hoarse. It sounds like a very cool system, and I'd love to see it in action.

I think it was a long time ago that I discussed a similar tool with friends (essentially a way of mapping out an argument, and listing out premises/arguments/conclusions). The feedback I got most often:

"Yeah that would be cool, but that feels like a chore, and I wouldn't really use it".

I wish sawsay the best of luck, and some points to perhaps keep in mind:

- Make the platform fun. This is going make or break the service -- at-least for popular use.

- Spend a lot of time iterating/evolving the UX so what is essentially a hard task, is broken naturally into very simple parts.

Woah! This would be so cool. So here's my tl;dr of this:

A post would include:

- Premise (a set of aximotic claims, or previous conclusions from other posts)

- A list of arguments.

The reply to this, would:

- point out which premise you disagree with

- Add new premises (which may invalidate previous arguments)

- Add new arguments based on your premises

We can also perhaps maintain a global pool of premises for that debate so when a new comer comes in, he can look at all the points raised by everyone, and taking them into account, post his or her own argument. Perhaps premises maybe remove-able from this pool if a consensus is reached that it is invalid (or perhaps when the original poster is convinced, and he or she chooses to retract it).

Did i get it right?

I actually came here to say this. This breaking-down process to turn complex arguments into a tree of simple statements that can be debated more objectively and with less controversy was the core of a similar platform that existed a few years ago, but which has since been discontinued, called Ideagraph.

I collected all the remains of the website I could find some weeks ago here: http://imgur.com/a/bAmcV. Unfortunately, there doesn't seem to be a web archive version.

I've been hoping the concept would surface again, and this Saysaw is quite close, but doesn't take the concept of distributed debate quite as far.

EDIT: To be perfectly clear, since the blurbs in the link above make no mention of the working mechanism of the site: Ideagraph worked by allowing each statement to be voted up or down, or subdivided into its component ideas. This could happen multiple times, creating a deep tree structure, where only leaf nodes could be voted on; the score of the parent nodes would be automatically calculated from its children, all the way to the top (the high-level arguments). Also, any node could be linked to another part of the tree, which they describe as "Facts concluded in one debate by one group of people are reused in other debates." The whole thing was built on top of MediaWiki, to provide the wiki features of distributed editing, edit history, etc. (and also because the idea came from an attempt to improve Wikipedia discussions). But mediawiki obviously that wasn't the ideal platform to build such a system, and I believe it's one of the reasons the project was eventually discontinued.

Thanks md224, yes thats a great idea, something like you described is being planned right now. Definitely a challenge to implement well while keeping it an easy/fun/accessible interface. Typical users would get totally lost in something like stackoverflow.

This is an awesome idea! I like thinking of arguments/debates as almost like a logical/mathematical argument (thanks to a lot of people, but most recently Day9 of Starcraft community fame) and this would make that explicit and therefore quite cool, in my opinion.

I've had this exact same thought process come back several times over the years, like an itch that ought to be scratched, but I haven't ever been able to nail down how I think the interface should actually work to facilitate the idea.

Yes I think it would be cool if debates could be broken into smaller issues which faced off against each other. See also http://idebate.org/debatabase

I have thought about this before and came to the same conclusion. If you want a serious debate site, then you have to dig deeper to resolve the finer points that make up a broader issue.

Reply on news.ycombinator.com