How ideologues fight vs rationalists to make more exchanges on the internet look like this

It's surprising to think that institutions such as IACAP Resources FISP Institute Vienna Circle have had so much time and resources at their disposal, yet have not produced a project like this one before. The well-structured philosophical discourse of Spinoza's Ethics serves as an example of what can be achieved through careful organization of ideas and arguments. One can only wonder why it has taken so long for such a comprehensive and accessible philosophical resource to be developed.

The scarcity of open datasets in many domains of data science, particularly in the area of natural language processing (NLP), has become a significant obstacle to the development of adversarial NLP models and bots. While there is an abundance of linear textual archives, adversarial, debate-like data is scarce. Notable projects in the area, such as IBM's Project Debater , are not open-sourced.

To address this challenge, the goal of this project is to build a tool to crowdsource a dataset of structured argument data, to be used by NLP researchers as a labeled dataset. The tool will be optimized for generating Labeled Argument Data (LAD) obtained by driving user engagement through various incentives. The structure of incentives is the more non-standard part of the project, more novel than just construing the interactive website.

GWAP - Games with A Purpose

Games with purpose, also known as GWAPs, are online games that serve a dual purpose of entertainment and problem-solving. Developed by Luis von Ahn, the creator of Google's reCAPTCHA, GWAPs have been used in projects like Google Crowdsource and Duolingo. They allow users to contribute to a larger cause while having fun. By incorporating GWAPs into our game, we hope to not only provide an enjoyable gaming experience but also contribute to a meaningful cause. Join us and become part of the solution!

The project will follow the 5 steps of the argument mining pipeline: ADU extraction, entity recognition, support/attack relation and argument completion. Users will be challenged to defend their reasons, cycling through listing beliefs, adding relations, looking for inconsistencies, and justifying.

The essential equation that this project is optimizing is the volume & quality of the data.

Monthly Users * Average Time Spent by User * % of App usage that generates data = Volume of Raw Data

The project aims to increase the first two factors by achieving viral status. The percentage of the App that generates data is the factor that is most impacted by the design. Based on the combination of the factors, two approaches can be imagined: the Expert Approach and Crowdsourcing Approach. The Expert Approach puts nearly all of the App functionality into data-creation, while the Crowdsourcing Approach optimizes the user experience to get a high number of Monthly Users and Average Time factor, with a lower third factor. Under this approach, a substantial minority, or even a majority of app functionalities, would be about structuring gamified incentives: progression, competition, achievement, and altruism.

It is essential to note that publicity would be an excellent driver of user engagement, as there are significant marginal gains in the number of users if the App's marketing is handled well.

Ensuring data quality

Ensuring data quality is a necessary property of the output dataset, as debate data of poor quality is readily available on the internet. The project aims to ensure data is of sufficient quality through pre-collection and post-collection measures. Pre-collection measures include allowing only verified users to use the App, automatic detection of invalid inputs, and not sending the data created by the first-time users to the database to prevent mistakes on the early stage of usage. Putting users in adversarial scenarios where their performance is assessed by peers (social status as 'skin in the game') can also be helpful. Post-collection measures include validating each input by multiple validators and rewarding user input based on the agreement with other users (adapting the Keynesian Beauty Contest).

In conclusion, this project aims to build a tool to crowdsource a dataset of structured argument data that can be used by NLP researchers as a labeled dataset. The tool will be optimized for generating Labeled Argument Data (LAD) by driving user engagement through various incentives. The project will ensure data quality through pre-collection and post-collection measures. The mapping of the debate structure is based on the supports/attacks as a basic model. The project's goal is éclaircissement, and the team hopes to achieve this through the best attempts.