Outline

1. The Problem
   • Classic discussion format
   • Aumann's agreement theorem
2. The Rules
   • An example run
   • Example debate
3. The Debate(s)
   • Find your belief(s) to debate on
   • Debate in pairs
4. The Feedback
   • Questions

Estimated duration: 90 min

Intro

• Georg Jähnig

  • Less Wrong Berlin since 2015
  • interests: Disagreement techniques, Double crux

• Yes/No Debate

  • since 2018
  • live workshops at
    • LW Berlin
    • European Less Wrong Community Weekend
    • Czech EA
  • online debates at reddit.com/r/YesNoDebate

Sounds familiar?

• Discussing for a long time and still
  • no (big) change of mind
  • no (big) learnings

Sounds familiar?

• Discussing for a long time and still
  • no (big) change of mind
  • no (big) learnings
• happens even when both are curious and willing to understand

Sounds familiar?

• Discussing for a long time and still
  • no (big) change of mind
  • no (big) learnings
• happens even when both are curious and willing to understand
• likely because:
  • we like to talk about our favourite arguments / what convinced us
  • but that's not necessarily what the other cares about / what might convince them

Sounds familiar?

• Discussing for a long time and still
  • no (big) change of mind
  • no (big) learnings
• happens even when both are curious and willing to understand
• likely because:
  • we like to talk about our favourite arguments / what convinced us
  • but that's not necessarily what the other cares about / what might convince them
• communication / discussion unfortunately seen as
  • space for the speaker
  • not as service for the listener

Classic format: "I talk, you talk"

• A and B get their speaking time (hopefully the same)

Classic format: "I talk, you talk"

• A and B get their speaking time (hopefully the same)
• while A speaks, B hopefully listens

Classic format: "I talk, you talk"

• A and B get their speaking time (hopefully the same)
• while A speaks, B hopefully listens
• when then B speaks, B hopefully also addresses what A said

Classic format: "I talk, you talk"

• A and B get their speaking time (hopefully the same)
• while A speaks, B hopefully listens
• when then B speaks, B hopefully also addresses what A said
• ➡️ no built-in checks
• ➡️ risk of talking at cross-purposes
• ➡️ risk of no / too weak alignment

Classic format: "I talk, you talk"

• A and B get their speaking time (hopefully the same)
• while A speaks, B hopefully listens
• when then B speaks, B hopefully also addresses what A said
• ➡️ no built-in checks
• ➡️ risk of talking at cross-purposes
• ➡️ risk of no / too weak alignment

The fix: Put in the checks

• ➡️ questions!

Classic format: "I talk, you talk"

• A and B get their speaking time (hopefully the same)
• while A speaks, B hopefully listens
• when then B speaks, B hopefully also addresses what A said
• ➡️ no built-in checks
• ➡️ risk of talking at cross-purposes
• ➡️ risk of no / too weak alignment

The fix: Put in the checks

• ➡️ questions!
• make sure they get asked
• make sure they get answered

Aumann's agreement theorem

• Robert Aumann (*1930)
  • mathematician, game theory
• If you know what I know & we're both rational, we must agree.

Aumann's agreement theorem

• Robert Aumann (*1930)
  • mathematician, game theory
• If you know what I know & we're both rational, we must agree.
• Same information & Same reasoning → Same conclusions

Aumann's agreement theorem

• Robert Aumann (*1930)
  • mathematician, game theory
• If you know what I know & we're both rational, we must agree.
• Same information & Same reasoning → Same conclusions
• Same input & same code → same output

Aumann's agreement theorem

• Robert Aumann (*1930)
  • mathematician, game theory
• If you know what I know & we're both rational, we must agree.
• Same information & Same reasoning → Same conclusions
• Same input & same code → same output

↪ If I disagree with someone, at least 1 of these is true:

Aumann's agreement theorem

• Robert Aumann (*1930)
  • mathematician, game theory
• If you know what I know & we're both rational, we must agree.
• Same information & Same reasoning → Same conclusions
• Same input & same code → same output

↪ If I disagree with someone, at least 1 of these is true:

• I lack (relevant) information.
• I am not rational.

Aumann's agreement theorem

• Robert Aumann (*1930)
  • mathematician, game theory
• If you know what I know & we're both rational, we must agree.
• Same information & Same reasoning → Same conclusions
• Same input & same code → same output

↪ If I disagree with someone, at least 1 of these is true:

• I lack (relevant) information.
• I am not rational.
• They lack (relevant) information.
• They are not rational.

• Bert: "People should be allowed to carry guns in public."

• Ana: "Should people be allowed to carry rifles?"
• Bert: "Yes."

• Ana: "Should people be allowed to carry hand grenades?"
• Bert: "No."

• Bert: "Should people be allowed to carry knives?"
• Ana: "Depends. Yes for small knives (maybe blades <10cm). No for blades >20cm."

• Bert: "So people shall not carry hand grenades, while the police have them?"
• Ana: "False premise: The police do not have hand grenades."

• Ana: "Do more people die from guns than from drugs?"
• Bert: "I don't know."

Questions about the rules?

Caveats: When to use & when not

Caveats: When to use & when not

Yes/No debate works when:

• talking with 1 other person
• you have a clear starting point:
  • a statement that you disagree about
• you both believe
  • there is an objective truth about the belief
  • that can be found using correct reasoning and maximum information
• both are willing to change their mind, at least partly
• where Scout Mindset is allowed

Caveats: When to use & when not

Works (probably) not at:

• (public) debates where it is
  • not about information exchange between participants
  • but about representation of your belief
• and you may not change your mind (in public)
• where Soldier Mindset is expected

An example debate

It is acceptable to campaign for veganism while eating meat myself.

What to ask?

1. Check for agreement with underlying claims:
   • Do you agree that many people die from Covid?

What to ask?

1. Check for agreement with underlying claims:
   • Do you agree that many people die from Covid?
2. Check for definitions / same understanding:
   • With "Covid vaccine", do you mean all vaccines, incl. non-mRNA ones?

What to ask?

1. Check for agreement with underlying claims:
   • Do you agree that many people die from Covid?
2. Check for definitions / same understanding:
   • With "Covid vaccine", do you mean all vaccines, incl. non-mRNA ones?
3. "Whatabout" questions are encouraged:
   • Would you also recommend getting the flu vaccine?

What to ask?

1. Check for agreement with underlying claims:
   • Do you agree that many people die from Covid?
2. Check for definitions / same understanding:
   • With "Covid vaccine", do you mean all vaccines, incl. non-mRNA ones?
3. "Whatabout" questions are encouraged:
   • Would you also recommend getting the flu vaccine?
4. "Why" workaround – probe for an answer:
   • Is one of your reasons to not get vaccine the lack of long-term data?
     (If not, feel free to tell me the actual reason.)

What to ask?

1. Check for agreement with underlying claims:
   • Do you agree that many people die from Covid?
2. Check for definitions / same understanding:
   • With "Covid vaccine", do you mean all vaccines, incl. non-mRNA ones?
3. "Whatabout" questions are encouraged:
   • Would you also recommend getting the flu vaccine?
4. "Why" workaround – probe for an answer:
   • Is one of your reasons to not get vaccine the lack of long-term data?
     (If not, feel free to tell me the actual reason.)
5. "Can you" trick:
   • Can you give me an example?
     (If yes, feel free to do so.)

What to ask?

1. Check for agreement with underlying claims:
   • Do you agree that many people die from Covid?
2. Check for definitions / same understanding:
   • With "Covid vaccine", do you mean all vaccines, incl. non-mRNA ones?
3. "Whatabout" questions are encouraged:
   • Would you also recommend getting the flu vaccine?
4. "Why" workaround – probe for an answer:
   • Is one of your reasons to not get vaccine the lack of long-term data?
     (If not, feel free to tell me the actual reason.)
5. "Can you" trick:
   • Can you give me an example?
     (If yes, feel free to do so.)
6. Lay out your whole reasoning – but step-by-step:
   • Covid causes many deaths, correct?
   • From this follows it is dangerous, correct?

Find ideas you'd like to debate on

• list of EA-related ideas:
  

Find ideas you'd like to debate on

• list of EA-related ideas:
  
• your own ideas
  • add an elaboration in 2-3 sentences
  • belief tree
  • falsification criteria (what would change my mind?)

Get matched

1. pick a memory card
2. find the matching person
3. compare your ideas
4. pick an idea to debate

Debate

Your feedback

• How was it?
• Could you understand your counterpart's belief?
• Did you update your belief?

Rinse & repeat

• get matched again
• and debate
• feedback

Insights from previous debates

• Asking is harder than answering
  • crafting a question forces you to realize what actually makes you believe what you think:
    • What are the relevant facts for you?
    • In what way do they lead to your belief?

Insights from previous debates

• Asking is harder than answering
  • crafting a question forces you to realize what actually makes you believe what you think:
    • What are the relevant facts for you?
    • In what way do they lead to your belief?
  • Put yourself into your partner’s mind
    • remember what they have answered before
    • come up with a reasonable follow-up question.

Insights from previous debates

• Asking is harder than answering
  • crafting a question forces you to realize what actually makes you believe what you think:
    • What are the relevant facts for you?
    • In what way do they lead to your belief?
  • Put yourself into your partner’s mind
    • remember what they have answered before
    • come up with a reasonable follow-up question.
• in spoken debates: Asking is harder than in written ones
  • because: more time

Insights from previous debates

"It was surprisingly good to have a focal point to respond to. Even if they posted a wall of text explaining their position, the question was the only thing I felt required to address, so it took less mental energy to debate than it would have taken in a different format."

– debater from reddit.com/r/YesNoDebate

Yes/No Debate more

• reddit.com/r/YesNoDebate
• twitter.com/YesNoDebate
• yesnodebate.org incl. Newsletter