EDIA: Stereotypes and Discrimination in Artificial Intelligence

This is an adaptation of the original version of EDIA, which will allow you to build and run the tool from a docker image.

[Paper] [HuggingFace🤗 Demo]

Language models and word representations obtained with machine learning contain discriminatory stereotypes. Here we present the EDIA project (Stereotypes and Discrimination in Artificial Intelligence). This project aimed to design and evaluate a methodology that allows social scientists and domain experts in Latin America to explore biases and discriminatory stereotypes present in word embeddings (WE) and language models (LM). It also allowed them to define the type of bias to explore and do an intersectional analysis using two binary dimensions (for example, female-male intersected with fat-skinny).

EDIA contains several functions that serve to detect and inspect biases in natural language processing systems based on language models or word embeddings. We have models in Spanish and English to work with and explore biases in different languages ​​at the user's request. Each of the following spaces contains different functions that bring us closer to a particular aspect of the problem of bias and they allow us to understand different but complementary parts of it.

You can test and explore this functions with our live demo hosted on HuggingFace🤗 by clicking here.

Setup and Use

Prerequisites: -Linux OS. -Have DockerEngine installed. -At least 12 Gb of RAM (final generated container uses ~9.78 Gb).

# --- 1. Clone this repo ---
$ git clone https://github.com/fvialibre/dockerized_edia.git && cd dockerized_edia 

# --- 2. Give execute permissions to .sh files. ---
$ sudo chmod +x setup.sh build_image.sh run_image.sh

# --- 3. Setup work environment ---
# This will create the necessary folders and download the data files to be able to generate the docker image.
$ ./setup.sh

# --- 4. Build the docker image ---
## Option A: Via local app/
$ ./build_image.sh <docker_image_name>

## Option B: Downloading the latest image from Docker Hub
$ sudo docker pull fvialibre/edia:tool

# --- 5. Run the docker image --- 
# Create a new <container_name> and run it. The EDIA tool will be available at the address http://localhost:<host_port> in few minutes.
$ ./run_image.sh <docker_image_name> <container_name> <host_port>

How to stop/delete and restart the created <container_name>

# ---  Stop the <container_name> container ---
# (Stop EDIA tool)
$ sudo docker stop <container_name>

# --- Start/Restart the <container_name> container  ---
# (Start/Restart EDIA tool)
$ sudo docker start/restart <container_name>

# --- Remove the <container_name> container  ---
# (After this you need to run the ./run_image.sh command to use the EDIA tool again)
$ sudo docker rm <container_name>

Permanently remove the EDIA docker image from your local

$ sudo docker rmi <docker_image_name>

Tool Configuration

Before execute ./run_image.sh, you can configure the tool through host_config/tool.cfg file. This contains the below configuration parameters:

Name	Options	Description
language	es, en	Changes the interface language
embeddings_path	your_embedding_file.vec	Path to word embeddings to use. You can use your own embedding file as long as it is in .vec format, and host it the host_data/ directory created automatically when you run ./setup.sh. If it's a .bin extended file, only gensims c binary format are valid. if this field is left empty, the tool will use two default embedding files according to the configured language.
nn_method	sklearn, ann	Method used to fetch nearest neighbors. Sklearn uses sklearn nearest neighbors exact calculation so your embedding must fit in your computer's memory, it's a slower approach for large embeddings. Ann is a approximate nearest neighbors search suitable for large embeddings that don't fit in memory
max_neighbors	(int) 20	Select amount of neighbors to fit sklearn nearest neighbors method.
vocabulary_subset	mini, full	Vocabulary necessary for context search tool
available_wordcloud	True, False	Show wordcloud in "Data" interface
available_logs	True, False	Activate logging of user's input. Saved logs will be stores in host_logs/ folder.

Resources

Videotutorials and user's manual

• Word explorer: [video] [manual: es | en]
• Word bias explorer: [video] [manual: es | en]
• Phrase bias explorer: [video] [manual: es | en]
• Data explorer: [video] [manual: es | en]
• Crows-Pairs: [video] [manual: es | en]

Citation Information

@misc{https://doi.org/10.48550/arxiv.2207.06591,
    doi = {10.48550/ARXIV.2207.06591},
    url = {https://arxiv.org/abs/2207.06591},
    author = {Alemany, Laura Alonso and Benotti, Luciana and González, Lucía and Maina, Hernán and Busaniche, Beatriz and Halvorsen, Alexia and Bordone, Matías and Sánchez, Jorge},
    keywords = {Computation and Language (cs.CL), Artificial Intelligence (cs.AI), 
    FOS: Computer and information sciences, FOS: Computer and information sciences},
    title = {A tool to overcome technical barriers for bias assessment in human language technologies},
    publisher = {arXiv},
    year = {2022},
    copyright = {Creative Commons Attribution Non Commercial Share Alike 4.0 International}
}

License Information

This project is under a MIT license.