A Representative Study on Human Detection of Artificially Generated Media Across Countries

This repository contains the code for our 45th IEEE Symposium on Security and Privacy paper A Representative Study on Human Detection of Artificially Generated Media Across Countries.

AI-generated media has become a threat to our digital society as we know it. These forgeries can be created automatically and on a large scale based on publicly available technology. Recognizing this challenge, academics and practitioners have proposed a multitude of automatic detection strategies to detect such artificial media. However, in contrast to these technical advances, the human perception of generated media has not been thoroughly studied yet. In this paper, we aim at closing this research gap. We perform the first comprehensive survey into people's ability to detect generated media, spanning three countries (USA, Germany, and China) with 3,002 participants across audio, image, and text media. Our results indicate that state-of-the-art forgeries are almost indistinguishable from "real" media, with the majority of participants simply guessing when asked to rate them as human- or machine-generated. In addition, AI-generated media receive is voted more human like across all media types and all countries. To further understand which factors influence people's ability to detect generated media, we include personal variables, chosen based on a literature review in the domains of deepfake and fake news research. In a regression analysis, we found that generalized trust, cognitive reflection, and self-reported familiarity with deepfakes significantly influence participant's decision across all media categories.

Survey

The code for our survey can be found in the webapp directory. We offer two modes to deploy the application. One for local development, one for testing the deployment in Docker. You will need to perform step 0 (data setup) regardless of the setup you choose.

0. Data

Put your data in webapp/data and webapp/image_data and do not change the directory structure. We also distribute some example data.

1. Local

You can locally deploy an instance of our webapp. This is probably the most convinient method for development. All you need to do is to install the package (in this case in editable mode and with test dependencies):

cd webapp
pip install -e ".[testing]"

Then you can build the database and deploy a development server:

FLASK_APP=dfsurvey.app.factory FLASK_CONFIG=default flask deploy build 
FLASK_APP=dfsurvey.app.factory FLASK_CONFIG=default flask run

For a full list of cli commands see here and full list of config options here.

The tests can be found under tests.

pytest

You will need a working version of Chrome and Selenium to run all tests.

2. Docker

We also offer a full dockerized environment which is used to deploy the survey. You can choose between three different configurations:

• Dev: Which deploys the local installation with a wsgi server (no-persistent database).
• Stage: Is an addition to the dev mode for testing the setup with a PostgreDB.
• Prod: Deploys the full setup with an outward facing NGINX which acts as a reverse-proxy for two backend-end services (one side-by-side; one rating based). This configuration needs further setup (described below).

Build and deploy

You can build and deploy the survey as follows:

Dev

docker-compose -f docker-compose.dev.yml build
docker-compose -f docker-compose.dev.yml up -d

Stage

Since stage is just an augment you have to provide both files:

docker-compose -f docker-compose.dev.yml -f docker-compose.stage.yml build
docker-compose -f docker-compose.dev.yml -f docker-compose.stage.yml up -d

Prod

docker-compose -f docker-compose.prod.yml build
docker-compose -f docker-compose.prod.yml up -d

4. Exporting Results

After you have run the survey you can export the results via the following command (substiute your own credentials where necessary):

FLASK_APP=dfsurvey.app.factory FLASK_CONFIG=export POSTGRES_USER=replaceme POSTGRES_PASSWORD=replaceme POSTGRES_DB=replaceme flask export json

Configure the production environment

The production environment is designed to be run on an actual domain server with a known domain. Thus, it requires some configuration upfront:

• Fill in your domain name in nginx/nginx.conf under server name. The setup will automatically generate a valid ssl certificate using certbot.
• Complete the prod.env file with your desired configuration.
• Enter your domain and mail adress in init-letsencrypt.sh.

Analysis

The notebooks for analysis can be found in the analysis folder. The data is available on osf.