Problem
EthicsNet’s goal was to allow for users to annotate websites, videos, and photos for ethical values in order to create a dataset that could be used to train machine learning models to understand morals. EthicsNet already had a Chrome extension, which allowed users to annotate different online sources. However, the purpose and process of annotating videos and text was unclear, and therefore, did not motivate users to complete it. In addition to the annotation suite, EthicsNet also wanted a new way for people to generate personalized AI tokens that could be shared with different AIs like ChatGPT that are based on their moral values.
Solution
We worked with EthicsNet to design the Personalization Suite, a brand new dashboard for creating personalized tokens which represent positions on ethical issues. A token is created by filling out a series of ethics questions, which are encoded along ethical axes, such as utilitarianism vs deontology, or hierarchy vs egalitarianism. These tokens are used in ongoing research by the EthicsNet team, with the eventual goal of being used by AI models to better align themselves with human ethics. We also supplemented EthicsNet’s Annotation Suite with better designed questions, allowing users to submit data in the form of text, images, and videos, annotated with their opinions on the ethics of the content in the data. This will also be used in the ongoing research to augment machine learning with ethical values.
Design
The idea of creating personalized AI tokens is a revolutionary idea that has not been done before, which made it both fun and challenging to design. Our goal was to design personalization questions that could encourage users to share their morals and background in a meaningful and engaging way. We went through several rounds of iterations of designing different question templates that could be reused for different types of ethical questions.
EthicsNet users also range from a variety of countries and backgrounds. Because of this, we aimed to make the questions on the personalization form universal. For example, we used a political compass on the demographics form to better rate political ideology.
Moreover, it was a big challenge to tackle addressing data privacy issues. We had to carefully consider how much personal data we should be asking for from users and prioritize transparency with users. Based on GDPR laws, we also designed a system for admins to be able to delete data upon a user’s request.
Tech Stack
EthicsNet is built from an existing React app in TypeScript with TailwindCSS. As part of our project, the underlying infrastructure was migrated from Create React App to Vite. The frontend is hosted on Firebase. The backend is in JavaScript using Express, and is hosted on AWS Serverless. We also connect to their existing MongoDB cluster.
Features
Personalization Suite
In order to generate a personalized token based on a user's background and values, we created an interactive questionnaire with a series of demographic and ethical questions for users to answer. We also provided users the ability to control what data is shared with EthicsNet.
User Dashboard
We created a user dashboard to allow users to easily copy, paste, and edit personalization tokens. The dashboard also allows users to view their annotation statistics, which summarize trends in annotations and encourages users to annotate more.
Annotation Suite
To ensure that the annotation process is understandable and compelling, we helped modify the annotation suite so that users can more easily annotate various text, image, video, and scenarios with clear understanding of what each field is asking for. Users can also view previous annotation activity.
Admin Dashboard
The admin dashboard allows EthicsNet admins to view statistics and trends about EthicsNet users. Furthermore, admin can view, edit, and delete users’ responses, and manage personalization questions displayed using a content management system.
