Inoculating against pandemic misinformation

Real-time misinformation detection and debunking through crowd-sourced expertise.

Dr Tristram Alexander

The worldwide response to covid-19 has shown us that one of the chief obstacles in battling the pandemic is not the science of the response, or even the development of a vaccine, but how to communicate reliable health information to the general public.  This challenge was identified by the World Health Organization as an "infodemic".  In the case of an emergency, people are eager for information, but might not have the literacies and skillsets needed to assess the quality of the information.  This creates opportunities for the spread of misinformation which can have direct health consequences, as populations reject scientific recommendations such as mask wearing, or try dangerous treatment suggestions.

This is a global problem, but successfully tackling it requires targeting individual examples of misinformation as they appear.  The scale of this problem is immense.  Misinformation is easily shared on social media platforms, and facebook has 2.8 billion users, and Twitter 300 million users.  Such misinformation is pernicious, as even after being debunked through a central fact-checking repository, its impacts on behaviour can remain.  The challenge is to transparently identify misinformation and verified information in real-time, and make explicit the evidence against misinformation, inoculating users against misinformation as it appears.

The most disadvantaged in our communities have been disproportionately affected by the pandemic, while also being most at risk to succumbing to misinformation.  Our solution seeks to give those seeking public health information in a time of crisis the tools to distinguish between misinformation and reliable information.  

Our solution is public facing, and ultimately seeks to provide information to the English-speaking general public.  Our solution dynamically identifies influential content as it appears in the network, assigns it a reliability score and shows the connection between misinformation and debunking information.  This addresses a current gap in present fact-checking services: in our solution information is shown in the context that it is seen by the target audience, and the argument type and debunking information is directly connected to this misinformation, so increasing the chance that the reader is inoculated against the misinformation.

Our solution also serves public health providers in English-speaking countries.  We are seeking to integrate with the government public health systems in the countries where we deploy, to provide real-time identification of the prevalence of misinformation, and a synthesis of current public health recommendations.   This will allow health providers to address their messaging to the identified information gaps.

The solution provides free and accurate access to information without discrimination to all stakeholders. The information is intended to negate the spread and impact of misinformation and ultimately improve the health of the population during a health crisis. 

Our solution will provide a free-to-use dashboard, via a website, showing a continuously updating snapshot of misinformation and public health information.  Information will be displayed in its original form, so it is familiar to users of the social network, with a reliability measure and in the case of misinformation, the argument type, debunking information, and network prevalence.  This will empower users to obtain and understand health information, and reduce the impact of pandemic misinformation.

All technical developments will appear in peer-reviewed publications, and be presented at international conferences.

The training data used to develop the argument type classifier will be used to produce a white paper.  This will identify the primary sources of misinformation in the current pandemic, and inform any government-driven misinformation inoculation program.

This tool will be the world’s first automated public health fact-checker, seeking to counter the negative impacts of misinformation.

The resource will be for use by the general public, and a valuable tool for national health sectors and global health organisations such as the WHO. This will directly influence public health policy, public health recommendations and public health behaviours of the population during a health crisis.

Overarching goal: To reduce the negative impact of misinformation on health during a health crisis by achieving the following objectives:

1. Identify and expose all major sources of health misinformation in a health crisis (validated by our own manual coding);
2. Enable more effective inoculation against misinformation by revealing the methods inherent in misinformation (measured by misinformation prevalence);
3. Be a major source of truth for the health sector and the general public during a health crisis (measured by user types, numbers and use of information e.g. public health information, media etc);
4. Influence positive health behaviours OR mitigate negative health behaviours during a health crisis (measured in partnership with MoH and the WHO and including measuring the impacts on public policy, public health recommendations and where relevant, data on vaccine uptake, disease testing, and other behavioural indicators).

Year 1 

We will focus on our technology development, including further refinement of our training data and machine learning algorithms and the development of a front-end for our technology.  We will engage with the Australian Ministry of Health, and develop a deployment plan for the pilot of our technology.  Our target is the population of Australia (25 million)

Year 2

We will launch our technology as a pilot in Australia.   Feedback from the Australian pilot will be used to develop the interface.  We will engage with Ministries of Health in Canada and New Zealand and seek to deploy into these markets.  Our target populations expand to include these countries (total population 78 million).

Year 3

We continue to expand our development program, deploying into the larger markets of the United States and United Kingdom.  We will connect with the World Health Organization to integrate our technology with their information services.  Target population 473 million.

In the technology development phase of our solution we are manually validating the success of our automated methods using already collected training data.

In the pilot phase we will partner with academic psychologists at the University of Sydney to measure the impact of our misinformation inoculation strategy.  We will also seek feedback through small-scale trials on the ease-of-use of our technology interface.  Our goal is to have better uptake and success than the current World Health Organization interface.

In the pilot phase we will also test the success of our system at correctly identifying misinformation dynamically, and classifying argument type.  This will continue to be carried out using manual validation.

In the Growth phase our project manager and industry consultant will manage the outreach to health organizations.  Impact will be measured by our success in obtaining buy-in from our target partners.  

In the Scale phase we will be looking at expanding our target organizations and integrating with the World Health Organization.  Our impact will be determined by the degree to which we are successful in seeing uptake by these organizations and influence on public policy.

Our team is set up to overcome the technical challenges of this problem.  We see barriers to deployment.  Our plan is to use the success of the Trinity Challenge to connect with business and government partners to develop a pilot solution to be deployed first in Australia.  With success we will demonstrate a technology that is shown to be useful to the public health sector, and so trigger further collaborations and opportunities for deployment.  We hope that through the contacts enabled by the Trinity Challenge we can then deploy into international markets.

There is a cost barrier to developing a front-end for our technology.  We have focused on developing the engine for processing and analyzing social media data.  We would use some of the resources provided as part of success in the Trinity Challenge to develop a public-facing interface to our engine.  We would do so in consultation with public health partners.

We see a barrier to migrating our technology to non-English-speaking domains.  Our automated methods are immediately applicable to different languages, but our important manual coding and validation steps would require expertise in these languages.  We see this as a longer term goal, beyond the three year plan.

University of Sydney

We are looking for funding support to enable the development of a pilot for deployment in Australia.  This funding, and pilot development, will enable expansion into additional markets, to bring the solution to scale.  We have some connections, through the University, to the World Health Organization, so funding support would ensure that the necessary partnerships were secured and the project could take place.  While early indications are very favourable from both the Ministry of Health and WHO, neither have discretionary funding at this stage to invest in such a project.

Beyond the grant, we are looking to expand into other languages, once solid measurable impacts are demonstrated from our deployment into English-language markets.

We see misinformation as a global, pressing problem, and in the course of the pandemic we have seen the damage it can do.  We believe our approach can make a difference to the global health response but we need to turn our proof of concept into a deployable form.  The Trinity Challenge will enable us to do this.

We are in the early stages of partnership development, and see a real value in partnering with Ministries of Health (MoH) in the target markets.  We are focusing on the Australian MoH in Year 1, while developing our partnerships with Ministries in New Zealand, United States, Canada and United Kingdom.  This will ensure the development of user-friendly and appropriate tools for both the health sector and general public.

Likewise, as our solution is deployed, we will collaborate with health stakeholders at the global level, such as the World Health Organization (WHO).  Direct and ongoing engagement with the WHO will ensure that the output is suitable for use at the global level, and so address the significant issues of concern raised by the multilateral organisation, supporting their mission to curb misinformation.

Additionally we will be seeking engagement with the general public, who are also target users for the tool, to gauge user-friendliness and effectiveness.  We are seeking to engage with social media partners to explore opportunities for integration into platforms such as facebook and Twitter.  We believe their feedback on the project would be invaluable.