tl;dr: I'm exploring the use of AI/ML to help think-tanks and research organisations with policy recommendations and research communication. I've been working on features such as searching a library of documents, extracting relevant information from large research reports, searching the web for relevant information, and adapting policy briefs to specific scenarios. I'm looking for feedback on whether this is a useful problem to be working on and how to tailor a solution that meets specific needs.
As with many others, the release of ChatGPT has me considering the implications of how AI/ML will be used in society.
It barely needs to be mentioned that there are many nefarious use-cases for ML in both the near-term and long-term future, but I don't see that being any reason to limit ourselves from capturing the benefits that are on offer for EA-aligned initiatives.
I've been playing around with GPT-3 for about a month, trying to design a system that can perform tasks relating to policy recommendations & research communication.
My rationale is the following: budgets for comms/marketing people in think-tanks are scarce, and researcher time is best spent doing original thinking, rather than mass research translation. I'm also reasoning that because policy research orgs are quite a narrow (and presumably unprofitable) market segment, they're unlikely to have a specific solution developed for their needs.
I currently work in public service in Melbourne, and I know a few people in EA-related think-tanks, so that's why this has occurred to me to work on. However, I still don't feel like I have a clear understanding of specific problems that are faced by think-tanks and similar research organisations.
I'm mainly just working from the perspective of trying to help spread good ideas by making them easily integrated with current events and broader context. However, I'm keen to understand whether that's truly a useful problem to be working on, and how I should be tailoring a solution that meets specific needs of researchers.
Up to now, I've worked on the following features:
- Searching a library of documents & extracting relevant information from large research reports
- Searching the web for relevant information
- Adapting policy-briefs to specific scenarios
- Creating a social media post that combines the recent events in a news article with the relevant points of a research report.
- E.g. Here is a "LinkedIn post" that I generated which combined a think-tanks report on aged care, and a recent article on escalating COVID rates: "COVID-19 is taking its toll on aged care facilities, with more than 100 people dying in the last week. The government needs to take urgent action to improve the workforce and ensure the safety and wellbeing of older Australians. This can be done by increasing the number of qualiﬁed care workers, improving work conditions, and increasing funding to aged care facilities. Means testing people receiving aged care services would also make it fairer for those who need it most. Funding could be raised through a Medicare-style levy on taxable income or by reducing generous tax breaks for wealthy older Australians. Including more of the family home in the pension assets test would also save the budget up to $2 billion a year."
- Creating a policy brief for a specific politician with regard to a specific topic.
- E.g. A task might be to "Develop a policy brief for the Minister for _____. Draw on the findings within the report titled ______. Critically analyse the current initiatives and programs by relevant departments such as _____. Also see if there is any recent media articles which are relevant to the issue, and can be used to provide extra fuel for the proposed arguments."