Better Discourse
- Many discussions between clever informed people don't seem to change their minds fast
- eg AI, Housing, LGBT+, trans, nuclear power
- Positive sum answers are possible
- If you don't believe this then you wouldn't expect progress. But I do, so why aren't we moving faster
- How do we agree more quickly on what we agree on and so make progress on that? My theories:
- Discourse has no checkpoints
- The same discussions happen week in week out. It's hard to say "no, we've already discussed that, it's not up for debate any more"
- Evidence
- Twitter
- AI, age gap, trans, race science,
- Counter evidence
- Narratives do shift, eg Trump, DEI
- But
- How could we give discourse that we care about checkpoints/dashboards
- UK dashboard
- London Housing Dashboard
- UK immigration dashboard
- US policy outcomes dashboard
- AI dashboard
- On manifold? More broadly
- Most people don't want to actually change their mind
- What forms of great discourse exist?
- Metaculus comment section
- Internal discussions at some organisations
- Community notes
- How could the algorithm get used elsewhere
- When have I convinced anyone of anything?
- What attemptedly good forms of discourse exist
- Twitter
- No checkpoints
- People get torn apart, hard to be scout mindset
- Youtube debates
- Focus on unimportant topics
- Are people shifting to better opinions?
- What are better opinions?
- Find big inaccurate markets on manifold and organise debates on them
- See if the markets shift?
- See if related markets shift?
- Youtube "differing opinions videos"
- Rarely seem to make progress
- Disagreement and ambiguity are more sparky than consensus
- Look at the youtube shows which focus on consensus. The episode with Flat Earthers and Scientists asks questions like "the other side are ill informed" and "I am smarter than the other side" rather than "enquiry is important" "planes fly" etc
- Understanding the future is really hard for everyone
- It's not clear that there are shortcuts to future knowledge. Political experts are worse than random. Expertise in general is bad at predicting the future. How many Discourse problems are solved mainly by waiting?
- What I think about
- How to store discussions
- What I work on
- Viewpoints - a tool to get opinions on complex topics.
- Viewpoints allows people to quickly swipe on statements and allows the questioner to find consensus. Likewise people can make comments to say what they are uncertain of in the statements.
- Writing up discussions - trying to put discussions in a way that can be stored
- What would success might look like.
- A conference run with better discourse processes
- Manifest
- Every time you discuss with someone and disagree put it in the app.
- Matchmaking based on some agreement, some disagreement
- Websites working more using consensus tools
- EA Forum LessWrong - anyone else running one would feel like a minor win
- Solving real world problems
- Helping build a block of Housing
- Running a set of debates
- Why do people like debates
- The discussion involves lots of contextual information. How two people talk about something tells you a lot more about it.
- Find two real people disagreeing on the disagreements in the text
- Estimaker Find run a debate for each node.
- Have people vote on each node
- Are they recording the values from the Pause.AI poll
- A process in search of a theory
- I am wary of my work because it involves a process I'm really interested in but doesn't necessarily yield results.