debugging one's thoughts, a system for

[gusl]

Here's a process for debugging one's thoughts:
* state your beliefs (data) in a formal language
* state hypotheses (i.e. something of the form forall x_1,..x_n, phi(x_1,...,x_n), e.g. explaining patterns in the data). (ignore all statistical connotations of the word "pattern": I'm talking about logical relationships)
* computer checks for consistency between beliefs and hypotheses, by trying to prove contradictions and absurdities from the premises above... also warn if unnecessary concepts are used, etc.
* now the user can revise hypotheses / concept definitions himself, or the machine could suggest a way out.

Go back to the first step to continue your theory-building / exploration for the purposes of understanding the stuff.

I think this would be similar to Socratic dialogue.

Comments

[tiedyedave]

It would be nice if one could produce these logical formalisms incrementally, rather than wholesale, as a kind of formalizing dialogue with the machine, with continuous feedback.

That is, one would sit down and type out a few assumptions in natural language, and the machine would present formalizations, possibly asking for disambiguation ("did you mean A if and only if B?"). Repeat, adding more assumptions for digestion, until the machine has presented a formalization to your satisfaction.

Then proceed to the checking step.

Of course, I envision the process of program compilation, and lots of other things, as eventually being addressed by this kind of feedback loop. Unfortunately, our understanding of natural language parsing and interactions will have to evolve noticeably before any of this becomes possible. :|

[gustavolacerda.livejournal.com]

I like your idea. By substitution of equals, you I like the following idea too:

http://www.optimizelife.com/wiki/index.php/ArguLogic
http://www.optimizelife.com/wiki/index.php/Underspecification_in_Argumentation

I don't think I follow your idea about program compilation.

[bondage-and-tea.livejournal.com]

I'd like the hypotheses to allow existential quantification as well, e.g. Ax.Ey.P(x,y).

Which leads to an interesting question. What makes an interesting hypothesis?

[gustavolacerda.livejournal.com]

Things of the form Ax.Ey.P(x,y) are not excluded by my definition. phi need not be a predicate.


<< What makes an interesting hypothesis? >>

It is an interesting question. And you're in the right place to explore it. Have you spoken to Alison about this?

[gustavolacerda.livejournal.com]

Interesting questions are those:

* in which a positive or negative answer would have significant/important implications (theoretical or practical impact)
* about which one has inconsistent intuitions (it should be A, but OTOH B...), essentially creating a dispute in one's head.
* the answer is not obvious
* through which one could potentially learn a lot, by exploring it

Different emphasis on the above answers give scientists different profiles.

[gustavolacerda.livejournal.com]

sorry, I mean DiaLogic