Notes | Tutorial 39

Date: 2020-09-11

1. Paths Forward (review)
2. efficiency
3. helpful things
4. epistemic stuff
5. during tutorials
   1. how to respond to literature cites
   2. how do you write literature to make bridging args easier?
   3. todo: write some early stuff about policies
   4. SI murders ppl
   5. PF - community representatives?
   6. Why do things fail? Generally:
   7. To read / references / related

Paths Forward (review)

PF is not a method. PF is something have, not something you do. It has been primarily (completely?) developed by Elliot Temple.

If someone has PF it means they have general methods for correcting errors. The presence (and adherence to) those methods is what provides PF. That presence is the solution to a very particular problem (and the related problems/questions):

If you are wrong about something:

• Are you always able to find that out? Why are you generally able to find out if you’re wrong about something?
• Can you be confident that this works for anything you could possibly be wrong about?

Important 28min: if you disagree with something, you should have thresholds and limits for things that will change your mind. you should know what they are and be able to tell ppl if they ask.

It’s not specific to one domain (e.g. research into a particular topic), it applies to all elements of one’s life.

The core of PF is about discussion. There are - broadly - two ways to find and fix an error. You can do it in your own head (thinking about it alone) or via discussion. Discussion is important because that’s the primary way of getting a variety of ideas. Even reading a book is a simple discussion (sort of) - the author says a bunch to you.

Note 62m: don’t blur (too much) active discussion w/ passive stuff like books/videos. two way convo has customised/personal stuff. books aren’t.

65: different experience when trying to read a book (and only the book), vs when you have the book + a fan of the book who can answer qs, etc. they can provide error correction -> (can) prevents compounding errors. Available advocates.

How you do discussions matters b/c it affects whether they’re high/low quality and whether they’re productive or not. Unproductive discussions are ones where people don’t learn anything, don’t resolve disagreements, do things around goals other than furthering the discussion and learning, the ppl talk cross purposes, etc. If you have lots of unproductive discussions it’s unlikely you have good PF because these discussions don’t answer the core question.

59min: important special case: someone else knows you’re wrong and wants to tell you. If you have PF you should ALWAYS be able to deal with this.

efficiency

• cite material
• develop a library of criticism
  • link ppl to that
  • 4min: use proxies, don’t need to do everything yourself
• allow conversations to go deep to find root causes of disagreements
• fail early (don’t have long breadth-first discussions by default)
• 8min: let opponents pick questions / topics

helpful things

• debate policy
• track record of public debate / discussions
• write & publish stuff
• 6min: backups - e.g. http://curi.us/2068-my-paths-forward-policy

epistemic stuff

• if something isn’t public it hasn’t been contributed to the project of human knowledge
• two way street

during tutorials

• key issue - avoid fooling yourself / suffering from your own biases
  • your biases need to be criticizable too
• what do you expect of other ppl re trust
  • if you’re holding yourself to a lower standard than you expect from others
    • issue – 18min in
  • if you don’t trust yourself -> easier for other ppl to trust you
    • your reasons are general beyond yourself

PF is alternative to social hierarchy - how can you show your stuff is worthwhile. main method for allocating attention is social status. social status based methods make errors / fool yourself. PF says ‘let’s figure out how to tackle and solve these problems’ build ‘real’ intellectual status rather than social status.

• poor reception of PF
• changes rules of game (from social to rational)
• 20-25min

– closeness to higher status ppl / accomplished ppl - ppl start treating them like a peer

proxies - popper’s status transferrable in some cases?

• tricky, proxy has to supply bridging args
  • minimal bridging args makes this easier / hard to reject badly

re popper and bridging:

• can start with “can you answer the famous arguments”
  • low difficulty hurdle to jump - should have recorded answers already
• ignoring something mainstream more dangerous than ignoring things that are obscure

how to respond to literature cites

basic method: read until the first important error.

reactions:

• dismissal - that’s not an important error
• holistic - you need to read everything to understand
  • ppl disagree with the method itself?
    • contradicts “PF takes too long”; the person then demands high effort
• the book but the idea is good
  • someone needs to write a good version
    • from scratch
    • copy just good paragraphs
    • extract good bits and have short explanatory commentary
  • why isn’t there a better version?
    • popular stuff should have good resources

meta arg: the fact the material is bad reflects badly on your school -> provides evidence about that group’s intellectual processes

people don’t like damning big things for “small” reasons.

47 min: if you ask someone:

would you be surprised to be wrong?

It’s a bit pointless to argue, then.

can label claims they think are weak/strong.

how do you write literature to make bridging args easier?

• link to different layers of abstraction
• “incoming” -> lower layers of abstraction
• consequences -> higher
• peers - sorta applications, but can be “sibling”

bridging args often about application, so example literature makes it easier maybe?

todo: write some early stuff about policies

• maybe also psuedo ‘guide’ for ppl starting?

SI murders ppl

too complex for most LW people

• imagine talking to a smart 15 yo, how much do they know about SI? probs need to start more basic; LW ppl don’t know enough about SI necessarily to apply.

note: universality of computation vs of mind -> rip superint.

breakpoints: different dimensions are a type of breakpoint.

• e.g. rational vs real numbers? aleph numbers?
  • no / maybe sorta
• more like traits - the stuff we were talking about in yes/no sessions. length vs green-ness of cucumbers

PF - community representatives?

http://curi.us/2064-less-wrong-lacks-representatives-and-paths-forward

Why do things fail? Generally:

• there’s an object problem
• then there’s a meta problem
• the combo is what breaks things

To read / references / related

• https://fallibleideas.com/paths-forward
• Gladwell’s new yorker article on college rankings – the order of things.
• http://curi.us/2065-open-letter-to-machine-intelligence-research-institute
• http://curi.us/2204-alisa-discussion

• https://www.elliottemple.com/essays/using-intellectual-processes-to-combat-bias

• read other PF stuff on FI and curi.us - note questions, etc

---

You can leave a comment anonymously. No sign up or login is required. Use a junk email if not your own; email is only for notifications—though, FYI, I will be able to see it.

---