Building a Fact-Checking Startup: Lessons Learned

Addressing the Challenge of Online Misinformation

Following the 2016 U.S. presidential election, a project was initiated with the goal of developing a solution to combat the proliferation of false information online.

The core idea was to create a fact-checking system, leveraging semi-automation, capable of identifying potentially inaccurate or questionable statements.

This system would then provide relevant, high-quality contextual information to support or refute those claims.

The Initial Vision

The underlying belief was that by utilizing technology to encourage individuals to prioritize truth, facts, statistics, and data in their decision-making processes, a more reasoned and rational online environment could be fostered.

The aim was to shift the online discourse away from exaggeration and towards evidence-based discussion.

Progress and Remaining Obstacles

Over the course of five years of dedicated effort, Factmata has achieved a degree of success in this endeavor.

However, significant hurdles – encompassing both economic and technological challenges – remain before this field can fully mature.

These barriers must be addressed to ensure the continued development and effectiveness of automated fact-checking technologies.

Significant Hurdles Encountered

It became apparent early on that the automation of fact-checking presents a remarkably difficult research challenge. Initially, a key obstacle was establishing a clear definition of what constitutes a “fact” requiring verification. Subsequently, consideration was given to the development and ongoing maintenance of comprehensive, current databases of facts, essential for evaluating the accuracy of assertions.

For instance, while the widely-utilized Wikidata knowledge base appeared promising, its update frequency proved insufficient for verifying claims pertaining to swiftly unfolding events.

Furthermore, operating as a for-profit fact-checking entity presented difficulties. The majority of journalism and fact-checking organizations operate on a nonprofit basis, and social media companies generally favor collaboration with nonprofits to mitigate potential accusations of bias.

Challenges in Defining Quality

Beyond these logistical concerns, constructing a business centered around assessing what is considered “good” is inherently intricate and subject to interpretation. Definitions are perpetually open to debate.

What was frequently labeled as “fake news” often manifested as extreme political partisanship, while instances deemed “misinformation” frequently represented dissenting viewpoints.

Consequently, we determined that identifying “bad” content – specifically, toxic, obscene, threatening, or hateful material – offered a more viable path from a business perspective.

We focused on detecting “gray area” harmful text, which is content that platforms are uncertain about removing but require further evaluation. To accomplish this, we developed an API that assesses the harmfulness of comments, posts, and news articles based on factors like hyperpartisanship, controversiality, objectivity, and hatefulness, among 15 other indicators.

Expanding Beyond Fact-Checking

Recognizing the value in monitoring the evolution of claims online concerning pertinent corporate matters, we expanded our offerings. In addition to our API, we created a SaaS platform designed to track rumors and evolving “narratives” across any topic, be it a brand’s products, governmental policies, or vaccines related to COVID-19.

The complexity of this undertaking is considerable. A significant lesson learned was the limited reach of $1 million in seed funding within this domain. The creation of training data for validated hate speech and false claims is not a standard labeling task; it necessitates subject-matter expertise and careful consideration, both of which are costly.

Development and Product-Market Fit

In reality, the development of the necessary tools – including browser extensions, website demonstrations, a data labeling platform, a social news commenting platform, and real-time dashboards displaying our AI’s output – was comparable to launching multiple startups concurrently.

Adding to the complexity, achieving product-market fit proved to be a challenging process. After extensive development, Factmata has transitioned to focus on brand safety and brand reputation.

We now offer our technology to online advertising platforms seeking to improve their ad inventory, brands requiring reputation management and optimization, and smaller platforms in need of content moderation. Reaching this business model required considerable time, but we are now experiencing consistent customer trials and contracts, and are projected to achieve $1 million in recurring revenue by mid-2022.

The Challenges of Building Socially Responsible Media

The path to establishing a business with positive social impact within the media landscape is fraught with obstacles. Significant change remains difficult as long as online advertising, search engine rankings, and newsfeed algorithms prioritize virality and attracting user attention.

Small organizations are unlikely to overcome these challenges independently and will require both financial assistance and supportive regulations to succeed.

The Need for Regulatory Intervention

Effective regulation is crucial, demanding that governing bodies enact robust legislation. While platforms like Facebook and Twitter have made advancements, online advertising infrastructure lags behind, and newer platforms lack motivation for improvement.

Currently, there's insufficient incentive for companies to actively moderate content beyond what is legally required; concerns about reputation or user retention are often inadequate. Even strong advocates for free speech acknowledge the necessity of financial incentives and prohibitions to compel platforms to invest in reducing harmful content and fostering a healthier online environment.

Envisioning an Alternative System

While eliminating undesirable content entirely is unrealistic, we can develop a system that actively promotes higher-quality material.

Algorithms, despite their imperfections, hold considerable potential. They can be utilized to automatically evaluate online content based on its quality and “goodness.” These resulting “quality scores” could form the foundation for new social media platforms that operate independently of advertising revenue, instead focusing on promoting – and financially supporting – content that benefits society.

Resource Requirements and Government Support

Developing these new scoring algorithms will demand substantial resources. Even the most innovative startups will likely require funding in the tens, or even hundreds, of millions of dollars.

Collaboration between multiple companies and nonprofit organizations will be essential, each contributing different versions of these algorithms for integration into user newsfeeds.

Government involvement can take several forms. Firstly, it should establish clear guidelines defining “quality,” relieving companies from the burden of creating their own policies.

Furthermore, government funding is vital. Such funding would prevent companies from compromising their objectives and encourage transparency and public scrutiny of their technologies, including potential biases and flaws. These technologies could even be made freely available for public use, ultimately serving the public good.

Leveraging Emerging Technologies

Positive steps have been taken by platforms to invest in the advanced technology needed for effective and sustainable content moderation. The advertising industry has also made progress, adopting brand safety algorithms like those from Factmata, the Global Disinformation Index, and NewsGuard.

Cryptocurrency and token economics offer a promising new approach to funding and incentivizing high-quality, fact-checked media. For instance, experts within tokenized systems could be incentivized to fact-check claims and efficiently scale data labeling for AI content moderation, reducing the need for substantial upfront investment.

Looking Ahead

The realization of the original vision for Factmata – as a technological cornerstone of a fact-based world – remains uncertain. However, we are proud of the effort undertaken and optimistic that our experiences can guide others in the ongoing effort to combat misinformation and disinformation.