When developing a project like this one, and particularly when outlining the ethos of Folk Evidence, it is essential to consider ways in which this resource could be misused. A system that could potentially be relied upon for treatment advice, even within the constraints of a medical professional, has to be reliable beyond doubt. Here I aim to outline a selection of the more significant potential issues with the repository, but this is by no means an exhaustive list. I would be grateful – if any other issues might arise – if readers would get in touch to bring such issues to my attention.
Abuse of Source Forums for Advertising
One particular issue I foresee for the source data involved is the risk of misinformation and data manipulation. By planting data or false anecdotes on forums, advertisers or actors within companies could falsely skew the data to force their own products to be recommended by the program. This would mislead the models used, and by proxy the users. In order to mitigate against this, there are a few techniques that would need to be explored.
Firstly, sources could be weighted by community trust, user history, and network credibility, rather than the volume of sources alone, which is much harder to falsify. Secondly, anomaly detection software could flag and discount unnatural posting patterns or commercial keywords. Finally, when delivering data and suggestions within a site UI, sources would need to be transparent, making any suggestions fully source-able.
Harm Reduction and Health Responsibility
Another issue that could impact the feasibility of Folk Evidence is the potential for the information it provides to be used unsafely. Even if the intent of the site is informative, it is difficult to ignore the level to which readers may interpret anecdotal reports as medical advice. The risk of this can be reduced through a few strategies.
Disclaimers are a major part of the provision of this kind of information under any circumstances, and Folk Evidence would certainly be in need of one. Prominently displaying that this information does not qualify as medical advice, as well as encouraging consultation with healthcare professionals, could help remind site users of its less reliable nature.
Privacy and Consent
Online anecdotal data often comes from forums such as Reddit, or social platforms where users’ privacy must be protected. This can be an issue when trying to quickly scour large amounts of data from the internet, so a few key steps must be included in the data collection process.
To begin with, any data must be anonymised, stripped of usernames and identifying details, as well as grouped so that anecdotes can be presented aggregately, allowing privacy to individual posts, summarising sentiment and trends rather than specifics. The site itself would have to be carefully reviewed legally to ensure compliance with data policy and any applicable laws, which would be its own step in creating the directory.
Algorithmic and Sampling Bias
Another issue with the very nature of online forums is their bias. Reddit, Facebook and other niche forums don’t represent all demographics, leading to biased perceptions of what works (more on this in later posts). This one is harder to solve easily, as most solutions would involve using more widely originated data, particularly non-english forums, but this would add a level of complexity to the software that would be difficult to work around. Alternatively, weightings could help bring greater strength to underrepresented perspectives, but this too opens up further issues in the execution of such a weighting. The best way to deal with this at the moment would be to publish software methodology, so that others can critique and improve it with this problem in mind.
Commercial Conflicts of Interest
There is a concern that if Folk Evidence grows, companies may wish to influence which treatments appear trustworthy through sponsorships and investment into the system. This is an unlikely situation the program would be in, but it is still important to recognise and mitigate against it.
This would mainly be in the form of terms of use and mission statements, including independence clauses forbidding the acceptance of advertising or sponsorship from entities selling anything that could be mentioned on the platform. This could also include full disclosure of any funding or affiliations, and even making data pipelines open source so they can be publicly audited.
These are the main concerns that spring to mind when considering the feasibility of Folk Evidence as a project, even without any understanding of the software involved. As far as I can imagine, these would be the most likely problems when building this project, but if there is anything you as the reader feel has been omitted or could be done to help these problems more effectively than the solutions I have suggested, please get in touch.