The following is a pitch for a new technology project (and eventually, a public benefit org). It’s still more of a twinkle in my eye than it is a real thing, but nights and weekends continue to be poured into it.
Spectra - Democratic Interoception
So far, the internet has generated more misunderstanding and division than it has dispelled. Our goal is to reverse that.
Goal 1: Diffuse outrage
Social feeds reward inflammatory political posts: allegations of hypocrisy, unfair characterizations, and worse. You can ask Spectra if such claims are accurate, and it will check by referencing opinion surveys, contextualising them with demographic data, and returning an answer. Usually, little hypocrisy exists, and instead, the beliefs of a few loud extremists are being misattributed to larger groups.
Goal 2: Build understanding
From our filter bubbles, it can be difficult to understand what fellow citizens actually believe. Do you want to know the president’s approval rating in your city? Spectra can take the latest national poll and use demographic markers to localize its results for you. Want to understand how views on states’ rights have shifted over time? Spectra can find and harmonize relevant surveys across time to try to answer that. Are you curious about a new issue, such as AI tutors in schools? Spectra can tell you when there is a hole in the available survey data and design one for you to finance the running of.
Goal 3: Align leadership
The project’s brass ring is to change the information landscape so that more aligned leaders run for office and win. First, by making constituents’ will legible, Spectra helps journalists, pundits, and activists call out unaligned behavior from politicians. Second, by creating common knowledge about which political platforms would be widely popular, Spectra will alter the dynamics of closed primaries so that they are more likely to nominate candidates popular with all voters, not just those of one party.
Approach
Answering these questions requires a rich synthesis of surveys, voter files, and census data. Much of this data already exists in disparate and spread-out formats, but one must be a trained data scientist to navigate it. The project’s first phase is to collect and integrate what exists now. Our more lofty goals will require additional surveys to be run, as existing ones generally don’t dive deep enough into the moral and psychological factors that act as downstream predictors of preferences and beliefs.
Principles
-
Open: Methodology will be transparent and auditable, and the code will be open source.
-
Accessible: A thin layer of AI will aid the creation of queries and aggregations.
-
Trustworthy: Results will be delivered as aggregations of the data with only minimal AI “interpretation” on top.
The first step is to combine the US census, community survey, and electoral data into a database and an LLM interface to query it. Stay tuned!