Empowering Journalists in El Salvador with AI to Combat Misinformation and Disinformation
June 11, 2024
In this success story, we explore an innovative tool built by Omdena in collaboration with IREX that leverages machine learning, natural language processing (NLP), and AI agents to combat the spread of misinformation and disinformation in El Salvador. By empowering journalists and media practitioners in El Salvador with cutting-edge tools, we’ve developed a system designed to validate the truthfulness of news items before they are shared with the public.
Here’s how we did it!
The Problem
The Challenges of Misinformation and Disinformation in the Digital Age
The advent of the internet and social media has made it easier than ever to spread information with just a click of a button. While this has many positive aspects, it also presents a significant challenge: the rapid dissemination of false information, whether inadvertently (misinformation) or intentionally (disinformation). This issue is particularly prevalent in areas such as politics, religion, and economics, where the impact on society can be severe.
The negative consequences of misinformation and disinformation are far-reaching:
- Societal Divides: False information can create or exacerbate divisions within society, leading to polarization and conflict.
- Manipulation: Individuals may be more susceptible to manipulation when exposed to misleading or deceptive information.
- Detrimental Actions: Acting on false information can lead to behaviors that benefit the perpetrators at the expense of the majority in society.
- Violent Extremism: In extreme cases, misinformation and disinformation can fuel violent extremism, posing a threat to public safety.
- Erosion of Trust: The proliferation of fake news erodes public trust in media, institutions, and the very notion of truth.
To address this complex challenge, innovative solutions that harness the power of technology, particularly machine learning and AI, must be developed. These solutions can help journalists and media practitioners validate information before it is released to the general public, mitigating the spread and impact of false information.
The Background
The Impact of Misinformation in El Salvador
In El Salvador, the spread of misinformation and disinformation has had a significant impact on society, particularly in the realm of politics and public opinion. A study conducted by the University of El Salvador found that 68% of Salvadorans have been exposed to fake news on social media platforms, with 45% admitting to having believed and shared such content.
The consequences of this widespread misinformation are evident in the country’s political landscape. In the 2019 presidential elections, false information circulated on social media, targeting candidates and influencing voter opinions. A survey by the National Democratic Institute revealed that 74% of Salvadorans believed that fake news had a moderate to significant impact on the election results.
Moreover, misinformation has exacerbated social tensions and polarization in El Salvador. False stories related to crime, immigration, and economic issues have fueled divisions and mistrust among different segments of society. A report by the International Crisis Group highlighted how misinformation campaigns on social media have contributed to the stigmatization of certain communities, such as those living in gang-controlled areas.
The COVID-19 pandemic has further exposed the dangers of misinformation in El Salvador. False claims about the virus, its origins, and potential treatments have spread rapidly on social media, leading to confusion and potentially harmful behaviors. A study by the Universidad Centroamericana José Simeón Cañas found that 62% of Salvadorans encountered fake news related to COVID-19, with 28% believing and acting upon such information.
The Importance of Fact-Checking in Journalism
In the era of fake news, fact-checking has become an essential practice for journalists and media outlets. By verifying the accuracy of information before publishing or broadcasting it, media practitioners can help combat the spread of misinformation and disinformation, maintaining the integrity of their profession and the trust of their audience.
However, with the rapid pace of news dissemination in the digital age, manual fact-checking can be a time-consuming and resource-intensive process. This is where AI-powered tools can make a significant difference, automating and accelerating the verification process, and enabling journalists to keep up with the ever-increasing flow of information.
The Partner
In our mission to combat misinformation and disinformation in El Salvador, we partnered with IREX, a global development and education organization that works to empower youth, cultivate leaders, strengthen institutions, and extend access to quality education and information. IREX has extensive experience in media literacy and countering disinformation, making them an ideal partner for this project.
IREX’s expertise in developing and implementing media literacy programs, as well as their deep understanding of the challenges posed by misinformation and disinformation, were invaluable in shaping our AI tool. Their insights into the needs of journalists and media practitioners, particularly in the context of El Salvador, helped ensure that our solution was tailored to address the specific challenges faced by our target users.
The Goal
The main objective of this project was to develop a robust AI tool that would empower journalists and media practitioners in El Salvador to validate the truthfulness of news items before sharing them with the public. By leveraging machine learning, NLP, and AI agents, we aimed to create a system that could analyze news items against existing patterns of real and fake news, as well as validate them using processes employed by credible news verifiers.
By achieving these goals, we aimed to empower journalists and media practitioners in El Salvador to combat misinformation and disinformation effectively, ultimately contributing to a more informed and harmonious society.
Our Approach
Step 1. Data Collection and Analysis
Gathering Local News and Fake News Datasets
To develop an effective AI tool for validating news items in El Salvador, we needed a comprehensive dataset that included both authentic local news and examples of fake news in Spanish. Our data collection process involved two main strategies:
- Scraping Local News Articles: We scraped news items from various online newspapers across El Salvador, ensuring a diverse and representative sample of authentic local news.
- Augmenting with Fake News Datasets: To complement the scraped local news, we gathered several Spanish-language fake news datasets from third-party sources on Kaggle and GitHub. These datasets were used to augment the authentic data and provide examples of false information.
Once the data was collected, we performed extensive exploratory data analysis (EDA) to identify patterns and characteristics of false news information. This process involved techniques such as data visualization, statistical analysis, and text mining to uncover insights that would inform the development of our AI models.
Determining Data Authenticity
To ensure the quality and reliability of our dataset, we employed various methods to determine the authenticity of the collected news items:
- Fact-Checking: We manually fact-checked a subset of the news items to verify their accuracy and truthfulness.
- Reverse Engineering: In some cases, we used reverse engineering techniques to trace the origin and spread of news items, helping to identify potential sources of misinformation or disinformation.
By carefully curating and validating our dataset, we laid a solid foundation for the development of accurate and effective AI models for news validation.
Step 2. Model Building
Leveraging AI Agents for News Classification and Validation
At the core of our AI tool are sophisticated machine learning models and AI agents that work together to classify news items and validate their truthfulness. Our approach involved two main components:
- Machine Learning for Pattern-Based Classification: We used the MFEND algorithm, a state-of-the-art machine learning technique, to classify news items based on patterns learned from our curated dataset. This allowed the system to quickly identify potential instances of fake news based on their similarity to known examples of misinformation or disinformation.
- AI Agents for Process-Based Validation: To further enhance the accuracy and reliability of our tool, we developed a suite of AI agents that mimic the processes used by credible third-party news verification outlets. These agents perform tasks such as:
- Determining how many times a news item appears in other credible sources
- Confirming if the headline aligns with the content of the news article
- Identifying the number of times a headline appears in other media outlets, which could indicate a misinformation campaign
By combining pattern-based classification with process-based validation, our AI tool provides a comprehensive and robust approach to news verification.
Integrating Credible Verification Sources
To ensure the credibility and trustworthiness of our AI tool, we integrated several well-respected news verification websites into our validation process, including:
- Verifica.EFE
- DB-Known Fakes
- Voz Publica
- Infodemia
By cross-referencing news items with these sources, our AI agents can provide users with additional confidence in the validity of the information, as well as identify potential sources of misinformation or disinformation.
Step 3. Stakeholder Engagement
Collaborating with Key Partners to Understand Current Validation Processes
To ensure that our AI tool effectively addresses the needs of journalists and media practitioners in El Salvador, we engaged with various stakeholders throughout the development process. These collaborations allowed us to gain valuable insights into the current practices and challenges of news validation, which informed the design and functionality of our tool.
Some of the key stakeholders we engaged with include:
- FUNDE
- APES
- Moon Shot Team
- Accion Ciudadana
- Disruptiva Magazine
- Independent Digital Newspaper “Voz Pública”
By working closely with these partners, we were able to incorporate their expertise and feedback into our AI tool, ensuring that it aligns with the real-world needs and processes of news verification in El Salvador.
Step 4. Model Deployment
Making the AI Tool Accessible and User-Friendly
To maximize the impact and usability of our AI tool, we focused on creating a seamless and accessible deployment process. Our team of ML engineers and software engineers developed a robust backend infrastructure that allows users to interact with the AI models and agents through intuitive APIs.
Key aspects of our deployment approach include:
- Flask APIs: We served the ML models and AI agents using Flask APIs, enabling clients to easily make API calls and receive predictions on the classification and truthfulness of news items.
- Docker and Streamlit Cloud: We deployed the AI tool on both Docker and Streamlit Cloud, providing flexibility for our partners to run the application either locally or on the cloud, depending on their preferences and infrastructure.
By offering multiple deployment options and user-friendly APIs, we ensure that our AI tool can be easily integrated into the existing workflows of journalists and media practitioners in El Salvador.
Step 5. User Interface Development
Creating an Intuitive and Localized User Experience
To further enhance the usability and accessibility of our AI tool, we developed a user-friendly interface that allows journalists and media practitioners to interact with the backend models and agents seamlessly. Key features of our user interface include:
- Streamlit-Based Development: We built the user interface using Streamlit, a powerful framework that enables rapid development of interactive and visually appealing applications.
- Spanish Language Support: Recognizing the importance of localization, we designed all interfaces in the Spanish language, ensuring a native and intuitive experience for our target users in El Salvador.
- Seamless Backend Integration: The user interface seamlessly consumes the services provided by the backend APIs, allowing users to input news items, receive classification and validation results, and view the sources used for verification.
- PDF Report Generation: To facilitate the sharing and documentation of news validation results, our user interface provides users with the option to generate a PDF copy of the report, which can be easily distributed or archived.
By prioritizing user experience and localization, our AI tool empowers journalists and media practitioners in El Salvador to effectively combat misinformation and disinformation, without requiring extensive technical expertise.
Overcoming Challenges in Developing the AI Tool
While developing our AI tool for news validation, we encountered several challenges that required innovative solutions and collaborative efforts:
- Data Scarcity: One of the primary challenges was the limited availability of labeled data specific to the Salvadoran context. To overcome this, we employed data augmentation techniques and leveraged transfer learning from models trained on similar tasks in other languages.
- Language Barriers: As our target users are primarily Spanish speakers, we had to ensure that all components of the AI tool, including the user interface and documentation, were properly localized. This required close collaboration with native Spanish speakers and thorough testing to ensure the accuracy and clarity of the translations.
- Infrastructure Constraints: Deploying the AI tool in El Salvador meant working with limited infrastructure and resources. To address this, we optimized our models for efficiency and developed a flexible deployment strategy that could adapt to various local constraints.
By proactively addressing these challenges and fostering a collaborative environment, we were able to successfully develop and deploy our AI tool, empowering journalists in El Salvador to combat misinformation and disinformation effectively.
The Outcome
- Successful Development and Deployment: We successfully built a comprehensive AI tool for news validation, including a robust backend infrastructure and a user-friendly frontend interface, which can be deployed both locally and on the cloud.
- Demonstration to Partner Organization: We had the opportunity to demonstrate a deployed version of our AI tool to one of our key partners, showcasing its effectiveness and potential impact in combating misinformation and disinformation.
- Effective Collaboration: Throughout the project, we fostered a strong collaborative environment, working closely with other contributors at Omdena to leverage our collective expertise and ensure the successful development of the AI tool.
Time Frame
The entire project from planning to deployment only took 2 Months!
Benefits and Applications
Empowering Journalists and Strengthening Democracy
The AI tool developed in this project has the potential to revolutionize the way journalists and media practitioners combat misinformation and disinformation, offering a wide range of benefits and applications:
- Enhancing News Credibility: By providing a reliable and efficient means of validating news items, our AI tool helps journalists and media outlets maintain their credibility and build trust with their audience.
- Saving Time and Resources: Automating the fact-checking process with AI enables journalists to quickly verify information, saving valuable time and resources that can be allocated to other essential tasks, such as investigative reporting.
- Promoting Informed Decision-Making: By reducing the spread of false information, our AI tool empowers citizens to make informed decisions based on accurate and trustworthy news, strengthening the foundation of democracy.
- Fostering Social Cohesion: Combating misinformation and disinformation helps mitigate societal divides and prevent the polarization that can arise from the spread of false information, promoting a more cohesive and harmonious society.
- Countering Violent Extremism: By curbing the dissemination of misleading or deceptive information, our AI tool contributes to the prevention of violent extremism that may be fueled by such content.
- Adaptability to Other Contexts: While initially focused on El Salvador, the methodology and technology behind our AI tool can be adapted to address similar challenges in other countries and regions, expanding its potential impact on a global scale.
