No organization or institution can escape the rise of artificial intelligence, and it’s no different for the Wisconsin Policy Forum. Like other organizations, we have had to contend with AI as it extends its digital reach into more domains. Our staff and I are seeking to harness its great potential for gathering, analyzing, and synthesizing information. However, in keeping with our traditions at the Forum, we insist on using it thoughtfully and carefully.
As with any opportunity to move in a new direction, it’s helpful to think about our history before rushing ahead. On the one hand, the Forum has welcomed innovation in the past, employing steps such as coding and interactive dashboards to better inform the public. As recently as 2018, we delivered data on the 248 largest cities and villages in Wisconsin by mailing readers a book of tables. Today, we provide that same information on all 607 cities and villages in the state using an online data tool that anyone can access instantly.
Yet whatever means we use to deliver our findings, the thing that matters most is whether people find them to be credible. For 113 years, the Forum has built a reputation for producing accurate and fair research. As you will see in this newsletter, this summer we won three national awards for our work from the Governmental Research Association – the most of any organization. All of that work could be undone by an ill-advised or shoddy study. With its tendency to hallucinate and misreport information, AI could make this worry a reality.
That’s why we have established principles within the Forum about how to do our work, whether it involves AI or not. We seek to be accurate, transparent, and accountable in our work, and to produce research that is trustworthy, useful, nonpartisan, and easy to understand. In using AI, every member of our staff knows that he or she must abide by these principles, use this tool carefully, and bear the responsibility for any of its mistakes.
As AI develops, so do its possible uses, and we are holding an ongoing discussion about the most and least accepted ways for the Forum to use AI. Here are a few examples of what we currently see as good uses of AI:
- To help produce code for data analysis tools such as R or Excel. We can test this code, which we employ to wrangle and analyze data. Once the script checks out, we can use and rely on the code itself rather than AI.
- To transcribe and summarize notes from the many stakeholder and expert interviews that we do for our research (while respecting any expectations of confidentiality).
- To review reports and studies to find errors and typos and to critique them from different perspectives.
When we use AI models to examine drafts, however, we generally avoid using their output directly. Given the tendency of these models to introduce errors, letting an AI model produce anything that a Forum reader will look at directly seems reckless and unacceptable to me at this stage. The rapid progress in the capabilities of AI models has astonished me just as it has so many others. Yet they must still demonstrate themselves to be trustworthy for each new task and responsibility. The importance of our work on behalf of our members and the residents of Wisconsin demands that we use AI with both wisdom and care.