This page contains my notes from weekly meetings of Stanford Effective Altruism. These notes are not meant to be a comprehensive description of what the meeting covered; it just includes the bits I wanted to write down. Comments reproduced here represent different individuals’ opinions, not necessarily my opinion or the consensus opinion of the group.

2016-05-29: Quantitative Models for Cause Prioritization

• Quantitative models (QMs) usually do the job, e.g., QALYs/DALYs
• If you rely too much on intuitions you’re prone to bias or contradicting yourself
  • I’d like to see a principled way of building robust comprehensive QMs
• We should distinguish between using QMs to make decisions and to find flaws in our thinking
• Michael should try to use his QM to predict lots of things in the near future
  • This question is largely already answered: in politics, sports, etc., QMs ridiculously outperform personal judgment
• For questions about values, you can directly input your values into the model, which you can’t do for empirical facts, so estimating values should be accurate
• Our intuitions translate poorly to numbers for small/big numbers
• Multiple models perform better than a single model
  • Single model has danger of overfitting
  • Can you “overfit” intuitions?
• Y Combinator mostly doesn’t use QMs
• In some cases your intuition is good and hard to model, e.g., judging character
• In addition to intuition and QMs, we have things we can communicate to others but can’t easily quantify
• A prior over interventions should relate to the distribution of income because starting a great intervention is like starting a successful company
• Quantifying your values is likely to not represent them correctly
• Model doesn’t work if you have inconsistent beliefs, so first you have to clarify/fix your inconsistencies

2016-05-22: Canvassing to Change Opinions

http://www.npr.org/sections/health-shots/2016/04/07/473383882/study-finds-deep-conversations-can-reduce-transgender-prejudice

• Canvassing seems effective and political campaigns over-invest in paid ads [citation needed]
• Campaigns often prefer volunteers over paid people because paid people often don’t actually do it
• Is the result affected by the types of people who talk to canvassers?
• How might this result generalize?
  • Probably easier to change sentiment than behavior
• Stanford PAW does dorm canvassing. We use a concrete ask like asking people to sign a Meatless Monday pledge and asking them to come to events. We get a decent feedback cycle by talking to them at events/emailing them
  • Conversations usually take 2-3 minutes
  • 60-70% of people who open their doors sign the pledge
• In my experience, it works well to show people you’re like them and you’re a reasonable normal person who changed your mind
• May be relevant that lots of people canvassed didn’t have established opinions on trans rights
• Charity Science found that canvassing is ineffective
• Companies often do cost-benefit analyses for ad campaigns, e.g., phone banks are expensive but still worth it
• If we did anti-animal welfare canvassing would it cancel out pro-animal welfare messages?
• Giving What We Can 10% pledge is a lot but could probably get students to donate 1%, maybe donate a bunch later
  • Stanford has students donate $20.16 presumably so they donate more later

How reliable is social science research?

• One report found that 36% of psychology studies successfully replicated (source)
• Definitions in social science are looser than definitions in, say, physics which makes replication harder
  • For something like veg ads there’s a clear measurable dependent variable
• Experiments are really important. In economics we much prefer randomized experiments over studies with large sample sizes but non-experimental design
  • When something happens randomly and you use that to establish causality, e.g., a city elects a new mayor who immediately hires lots more police officers, this tells you more about how police affect crime than running a big regression across cities

2016-05-15: EA Global

• EA Global is valuable for meeting EAs
• Potentially valuable to bring a lot of smart people together at once, can have efficient conversations
• Coverage of conferences could be valuable for spreading EAs
• EAG probably not good for recruiting because random people at a conference aren’t necessarily welcoming or a good first impression
  • If you wanted to recruit people, you’d design the conference differently
• EAGx model is good because it increases the public profile of local groups, you can meet EAs doing niche projects; was good for someone like me who’s involved but not super involved
• EAG was valuable for meeting people who know specific things I wanted to know and talking to people about it
  • Could talk online but people are more likely to talk at a conference
  • Academics have this same need and they solve it with conferences
• Increasing the public profile of EA is good even if there’s bad publicity in the short term; makes it look like we’re doing things
  • The things we should look like we’re doing aren’t conferences
• Maybe someone should be working full time on press releases about EA stuff
  • Orgs do do their own press releases in may cases
• Influential social movements get influence with media outlets, e.g., Jacobin for socialism [citation needed]
  • No they don’t [citation needed]
  • They generally don’t found media outlets until they’re big [citation needed]
• EAG made it easier for me to learn important context when I was fairly new to EA
• There are people who are committed to EA but just don’t know EAs IRL
• You can’t prevent people from coming away with a bad impression but it’ll be better if the conference is more academic and less “populist” [citation needed]
• Conference should be general (not about a specific cause area) because it’s valuable to convince people to change causes
• Should flag talks according to how much background knowledge they require
  • Just synopses might be good enough or better because it’s more specific and not potentially condescending
• Getting people to hear about EA is useful so conference publicity is good; we may care too much about negative publicity because it feels disproportionately bad to us, but non-EAs don’t care that much [citation needed]
  • But it would be better to have different things for publicity vs. EA networking
• How to tell if a conference was useful to attendees? Feedback surveys aren’t that useful [citation needed]

2016-05-09: Values Spreading

• Spreading concern for other beings may be extra important because lots of other important changes are common interest to many values
• What proportion of donations are made with the goal of doing the most good?
• We should distinguish between values spreading to people in general and targeted values spreading to specific groups
• Can do easy targeted values spreading with Facebook/LinkedIn ads
• CEV: people’s values mostly change from things like social pressure, not thinking hard [citation needed]
• Might be harmful to spread consequentialism if people are bad at it
  • Making policy more consequentialist seems more likely to be good
  • Governments are already pretty consequentialist with things like foreign aid
• For memes to self-propagate, they need to replicate exponentially
  • It’s useful to get people to identify with things, e.g. vegetarianism has a clear identity and daily action
• Atheism movement was destroyed by feminism [citation needed]
  • But it basically served its purpose by making religion worthy of ridicule in certain circles

2016-05-01: EA and Humanities

• EA movement may have been better if we hadn’t focused so much on earning to give
  • But it may have been useful to have a clear message
• History is useful for understanding large-scale effects, social movements
  • Knowledge of history is useful but is being a historian necessary?
• Useful to have inter-field collaboration
• If we use lots of econ/CS-y terms, it makes it harder for people who could participate but just don’t know those terms
• At Yale EA we had 20+ people come to the first meeting but only a few math majors kept coming
• EA might attract techy people because that’s usually the kind of people who are interested in EA stuff
• Most effective thing might be to make EA events more fun
• Cause neutrality means you can’t revolve around a cause

Underrepresented fields

Numbers indicate how many people named this field as one where they thought the EA movement needed more people in it.

• Politics/policy (8)
• Marketing/media (6)
• Writing/journalism (5)
• History (3)
• Statistics (2)
• Sociology (1)
• Philosophy (1)

2016-04-26: Larissa MacFarquhar

This was a special discussion with Larissa MacFarquhar, not a regular meeting.

Comments by Larissa:

• I wrote /Strangers Drowning/ to show what being deeply morally committed looks like, and investigate why more people don’t do that.
• The EA community is generating new morally committed people.
• “Rational” types who haven’t really thought about morality are the easiest to persuade. It’s harder to persuade people with pre-existing moral commitments.
• There’s a second potential window to get people into EA after their kids go to college.
  • Stanford Distinguished Careers Institute helps retirees find ways to make a difference.
• Lots of non-EA foundations have the same disagreements as EAs about whether to support measurable or high-risk interventions.

Comments by other people:

• According to a Slate Star Codex survey, people in the EA movement for longer. tend to move away from global poverty and toward more unusual cause areas
  • A couple of people here started interested in global poverty and then shifted to other cause areas
• The EA movement under-prioritizes reaching out to highly influential people
• Lots of EAs agree that journalists (or other non-traditional-EA professions) are really valuable but that’s often not apparent to potential EAs

2016-04-24: Population Ethics

No, I didn’t forget to take notes today, but we didn’t discuss anything philosophically novel and I didn’t believe the discussion would be interesting in writing form.

2016-04-17: Universal Basic Income

Group 1 discussion

• GiveDirectly’s Universal basic income (UBI) experiment could be good for getting more donations by doing something novel
• $1/day is surprisingly low
  • Would be interesting to have multiple cohorts with different amounts of money
• If income is a positional good, it matters how connected the cohorts are
• Some things generalize between Kenya and the developed world, e.g. how much not having to work affects your happiness
• If you say work makes people happy and you have a job at Google maybe you should be skeptical of that
• Have real-time happiness surveys found that people are happier with work or leisure?
• Rent could decrease with UBI because people don’t have to live near jobs
  • Empirically, moving is hard because people don’t usually move to places with more jobs
• Could be stigma on UBI as “parasitism”
• GiveDirectly has found that cash transfers create conflicts between neighbors [citation needed]
• How do we detect that UBI makes people happier/less happy?
• UBI may have better/different effects if it’s for life, and results of a 10 year study may not generalize
• This study would convince me¹ UBI is good if unemployment doesn’t increase

Group 2 discussion (notes by Caroline)

• what will the effect of UBI be on prices?
• what about wages?
• will there be jobs with no one who wants to do them?
• what about migrants?
  • if migrants are given UBI, lots of people will migrate
  • Swedish immigration partly caused by generous welfare state
  • Alaska Permanent Fund
    • doesn’t really incentivize moving there
• what will happen to rent?
  • will it cause homelessness?
• will people still want to work minimum wage jobs?
  • evidence from Indian reservations
  • they get a UBI from casino money
  • many choose not to work and live below poverty level
• how do jobs affect happiness?
  • unemployment makes people really unhappy
  • but partly because stress of looking for jobs
    • retirement is better than unemployment
  • majority of people retire as soon as they can
    • but a lot of them are blue collar workers with physical health issues
  • what will Kenyans do with more leisure time?
    • no Netflix, etc.
    • could mean they spend more time on social interaction than us
• people love Social Security
  • could phase in UBI in America by lowering retirement age
• [long off-topic argument about shipping routes and whether we should sell the United States to Singapore]
• how much can we generalize from Kenya to other countries?
  • studies in India would be useful due to different culture
  • also Ghana
    • (most African countries don’t have good payment infrastructure)

2016-04-10: Jacob Steinhardt on AI Safety

• Experts know things that are difficult to communicate to outsiders
• A lot of GiveWell’s early success came from the ability to find arguments for/against something
• My responsibilities as advisor for the Open Philanthropy Project
  • Open-ended conversations on early investigations
  • Referrals to relevant contacts
  • Represent Open Phil at the Puerto Rico AI conference
• I’m somewhat skeptical of MIRI’s agenda
  • Some problems they’re working on seem disconnected from AI safety
• Both safety-concerned AI research and AI safety research are important so personal fit is the most relevant consideration
• Some machine learning problem domains require a way to robustly handle low-probability bad events, e.g. self-driving car crashes
• Could be worried about poorly organized coalitions building AGI with kinda bad values designed by committee or something
• OpenAI: Elon Musk is concerned about authoritarian government/surveillance

2016-04-03: Does development make people happier?

• GiveWell charity comparison measures utility in percent income increase
  • Isomorphic to converting income increase to some other metric
  • It’s bad for the final metric to be something that’s questionably good
• Do life satisfaction survey questions lose something in translation?
  • You could ask bilingual people the same questions in two languages and see if they answer differently
• GDP per capita is a bad metric of how much richer people got
  • GDP per capita growth looks similar to median income growth
• How to account for hedonic set point in different places?
  • Look at difference in prison vs. not-prison happiness across countries
• People in urban China are much healthier than rural China
• Health in China has gone up a lot but self-reported health has not
• High uncertainty on what affects happiness means maybe we should be less confident about global poverty
  • More clear that disease is bad
• Income and happiness over time are correlated in most countries, but not US or China because of rising income inequality

2016-03-27: Monetary Policy

Today I make lots of empirical claims without citations so just add [citation needed] to everything.

• Fed is scared of inflation so it wants to raise rates but Fed Up says unemployment is more important
• Raising rates could hurt returns for pension funds, hurting future generations
  • This only matters for low-risk investments like bonds
• Lobbying to keep rates low could make Fed over-reactive if inflation accelerates
  • Open Phil: even if so, Fed Up creates more debate which is good
    • This is fully general argument for lobbying for anything
• Yellen doesn’t want to raise interest rates
  • Yellen wants to raise interest rates
• Useful to have a Schelling fence against using public opinion to shape monetary policy
  • Other groups already do this
• More transparency in Fed appointments is probably good
• Fed Up’s race argument fell flat, I¹ have low confidence that Fed Up is effective
• Interest rates track expectation Fed raising federal funds rate
• Making the dollar stronger is good for US but possibly net bad because international debtors owe more. Fed incentives are misaligned here
  • Fed has claimed that it cares about other countries’ economies
• Other countries like Brazil have worse unemployment and people in charge of monetary policy probably aren’t as good so trying to shift policy there may be better
  • Maybe Brazil is a bad example
• Lots of groups are lobbying on interest rates
• Fed has institutional bias to do what it’s always done rather than new things
• Fed wants to be credible because if Fed isn’t credible, interest rates won’t track federal funds rate
• We should target Spain or Nepal because they’re the largest countries with lots of unemployment
• Taylor rule suggests that the federal funds rate should be two percentage points higher than it currently is

2016-03-13: Cultured Meat

• It’s better to tax animal meat than subsidize cultured meat but probably less politically feasible
• Easier to add new subsidies than remove subsidies because of lobbying pressures
• Subsidies are harder than getting grants or other methods of furthering cultured meat
• Cultured meat hate is analogous to GMO hate
  • Anecdotally, a lot of GMO hate is really Monsanto hate
• Cultured meat people and the Good Food Institute are market testing some better things to call it
• Cultured meat could taste better than animal meat because we can customize it
• Open Phil’s writeup on why they’re not funding cultured meat used 2004 cost-effectiveness estimates which are inaccurate; conversation notes have an expert claiming he won’t get funding any time soon but got $2.5M two months later [citation needed]
  • Given how few people are working on it, it’s too hasty to conclude that we can’t reduce costs
  • Claims plant-based meat is crowded; it currently has ~$200M funding, and there’s potentially at lot more room
  • Chicken nuggets don’t resemble chicken that much, we can more easily culture stuff like that than fully-formed chicken parts

2016-03-06: Politics

• High-impact policy areas: foreign aid, immigration policy, wars + international stability, science + research funding, economic growth, industrial agriculture
  • Other high-impact areas: US poverty, trade, criminal justice
• Lobbying might be efficient or might not, we disagree a lot
  • Swift Boat campaign was effective
• A study found that if you ask people about candidates before an election policy predicts votes much better than personality traits²
• Funding is a trailing indicator: candidates who do well get money, not the reverse [citation needed]
• Maybe more effective to try to influence future candidates. Might not be too hard to predict who future presidential candidates are
• Disrupting speakers might influence rhetoric a lot, e.g. Black Lives Matter protesting candidates
• Could be good to mobilize for World Bank appointments or other powerful non-elected positions
• If Fed Up-like campaigns worked, you’d expect big bond holders or other people with an interest in keeping interest low would fund them
  • Unless tragedy of the commons

2016-02-28: Scientific Research

• Could we predict in advance that life-saving research projects were valuable?
  • Smallpox eradication was clearly important when there was a big international project on it, but not obvious before
  • Easy to tell that high yield crops would be useful
• What is valuable and physically possible but we don’t know how to do?
  • Nanotechnology
  • In vitro meat
  • Anti-aging? (probably physically possible but unclear)
• Hard to get funding for projects that take more than a few years [citation needed]
• Should we fund the development of ideas? How?
• In vitro meat might not get much industry funding because it would be easy for competitors to replicate once it’s marketable, so it might not be profitable
  • Hampton Creek raised tons of funding for something in the same space as in vitro meat
• In vitro caviar could be effective because it’s really expensive anyway so you don’t have to drive the price down to be competitive
• Fish need better PR because no one cares about them
• Should research how to make animals feel less pain

Other group discussion (notes by Daniel Filan)

• AI safety strategy
  • Daniel Dewey has a paper about how to make sure that people actually use friendly AI
• Unsexy research
  • Unsexy research is probably undervalued, and where we should look to find high value things.
  • This sort of attitude is common in startups
• What things could help science?
  • Tools
    • Good, easy-to-use numerical methods have been good
    • Same thing but for inference is possibly really valuable
  • Structure
    • William Shockley wrote a paper about why some scientists are much more productive than others
    • Basically, there are like 8 independent things you need to do to write a paper
    • If you can get a bit better at all of them, this is a huge productivity boost
    • “On the statistics of individual variations of productivity in research laboratories”
    • Could be good to get scientists in congress to stop dumb things in allocation of research funding
    • Good paper on whether the FDA is making it too easy or hard to sell drugs
    • “Is the FDA too conservative or too aggressive? A Bayesian decision analysis of clinical trial design”
  • About the field of metascience
    • Tend to be in clinical/social science
    • Could be valuable to have research in math/physics/cs
  • People/orgs/websites to look at:
    • John Ioannides
    • Stanford METRICS (website)
    • Centre for Open Science
    • Reproducibility project
    • Science exchange (company?)

2016-02-21: EA Community

• If we spent a year looking for weird ideas, how likely are we to find something good?
  • The group has a variety of opinions on this point
• If you do identify a good cause for which no organization exists, it’s hard to do anything with that knowledge
  • You can write about it and get others interested, maybe someone can do something
• If campus EA groups protest something we could get attention and potentially associate EA with good things
  • Fossil Free Stanford is trying this but isn’t attracting as much media attention as Black Lives Matter; some people think it’s trying too hard to get publicity instead of focusing on what they care about
  • Experience Poverty went poorly
  • But Live Below the Line is similar to Experience Poverty and hasn’t gotten serious negative PR
  • Black Lives Matter has an advantage because it’s about identity politics
• Likely to get good publicity: write about some popular issue from an EA perspective, e.g. how much good Bush did with his global health policies
  • Gets people thinking about global health as an important consideration for presidents
• We should maybe care more about EA ideas getting good publicity than the EA movement itself
• People have negative affect attached to math, so we should avoid math in PR pieces
• Why people leave Stanford EA
  • I come when topics are interesting
  • Too much public display of affection
  • It’s bad for my schedule
  • Meetings are boring
  • Organizing logistics of donating to charity is not our comparative advantage so we shouldn’t spend so much time discussing it
• Talking about politics is fun but maybe not the greatest topic
• Getting people to show up isn’t an end in itself; we want people to donate and do effectively altruistic things
• It’s important for communities to accommodate people with kids (maybe not for SEA though)
• Sometimes we assume a disagreement exists because we believe other people haven’t considered our point of view
  • Example: assuming the best part of Elon Musk coming to EA Global is to get him more EA, not because he could teach us something
• We could read GiveWell writeups to learn about their methods, discuss disagreements, etc.

2016-02-14: MIRI

Background

• AGI could easily kill everybody
• Bostrom explains why in Superintelligence
• “Core” EA focuses on robust evidence, which you can’t get for AI safety so we need other ways to assess MIRI’s effectiveness

Group Discussion: Measuring Effectiveness

• Count number of publications
• Could fund PhD fellowships to publish more papers
• Look at who MIRI is working with, who’s giving it grants
• MIRI could be net harmful for AI safety
  • MIRI caused OpenAI to be created [citation needed] which is probably bad
• A visible weak AI could make MIRI more palatable
• Early researchers usually do not become well-recognized after a field grows (cf Robin Hanson)
• Getting highly cited is a good signal because it means the org is interesting and more likely to get top researchers on board
• EAs are bad at management. They should read How to Run a Business 101
• MIRI has exceeded expectations—a few years ago I wouldn’t have thought it would have this much influence

Other Group Summary

• What makes academia successful?
• Evidence for MIRI will always be weaker than AMF, so to prefer MIRI you have to care about the size of outcomes

2016-02-07: Global Health Unmeasurables

• Saving human lives is generally easier than improving them
• Curing chronic pain looks promising
  • Lots of chronic pain is fixable by exercise [citation needed] but not neglected or tractable
• Living in a malaria-prone area makes you afraid of mosquitoes, so there are indirect quality of life benefits to reducing malaria
• Is malaria eradication bad for mosquitoes?
  • Successful eradication could make ecological intervention more palatable
• Malaria effects: fever, nausea, aching joints, swollen organs; can cause other chronic illnesses
• Why might the EA approach to global health be fatally flawed?
  • Saved lives were net negative
  • Not enough funding on research
  • Negative flow-through effects
  • We should have spread Western culture instead of resisting colonialism
• If you give people nets for free, they’re more likely to buy nets later because they’re more aware that nets are important [source: “Short-Run Subsidies and Long-Run Adoption of New Health Products: Evidence from a Field Experiment”]

2016-01-31: Political Reform

• Some countries tend to follow others, e.g. US follow UK on social policy. May be better to focus reforms on leading countries
• People psychologically prefer command and control over Pigovian tax because they don’t like thinking about tradeoffs
• Open borders in China–authoritarian government can control things, plus large population can absorb a lot of people
  • Chinese culture requires that immigrants assimilate, so would probably not create cultural divides like in the US
• Important to have people who understand AI research and know what’s dangerous
  • Probably fewer than 100 people would re-implement AlphaGo given a low-level description
• Measuring political change
  • It matters how fast transitions are; depends on the issue and is hard to predict
  • How impactful is lobbying? Silicon Valley firms spend more as they grow which suggests lobbying is rational [citation needed]
  • Professional lobbyists probably have information on the impact of lobbying
• Could fix North Korea by uniting with South Korea
  • Worked well for Germany, maybe not for Yemen
• Can use $1 million to bribe Representatives on issues they don’t much care about
  • More corrupt countries are easier/less risky to bribe

2016-01-24: Criminal Justice

(Here BFF refers to the Bronx Freedom Fund.)

Beginning

• Unclear how cost-effective BFF actually is. How much does bail cost? How long do people get out of jail for?
• BFF claims to reduce guilty rate from 90% to 50% by reducing plea bargains
  • Could be a selection effect
• Would BFF grantees have taken bail bonds? What can BFF do that markets can’t?
• Potential benefits to government for not having to pay for incarceration
• How is 97% court appearance rate plausible? Why would the government even bother with bail if that were true?
• Charity might be more effective than private corporations because charities can be more discriminating (less legal restriction)

Flow Through Effects

• Theory: use BFF to raise concern for bail reform
• Why do people take pleas if it’s bad? Is their legal counsel doing its job?
• It would be hard to convince people to eliminate bail altogether, maybe easier to reduce or eliminate for minor offences
• Most European countries have reduced bail and lower court attendance rates [citation needed]
• Lost of people who don’t plead guilty go on trial and are found not guilty might actually be guilty
• 97% figure comes from BFF’s form 990 so it’s at least correct on the face
  • Not showing up for court is really bad for you so you might expect almost everyone to do it
• If BFF chooses grantees who are particularly trustworthy/honest/etc. then it’s a weaker case for bail reform

Expected Value of BFF

• Effects: Pre-trial jail time, more acquittals, grater time working, increased trial costs
• Only 7% of people released without bail fail to show up
  • Could be strong selection effects
• Bigger flow-through effects on helping first-world people
• A lot of people think of (accused) criminals as the outgroup, which is psychologically harmful for criminals. Possible really good to lift that psychological burden
• Bail on average prevents 15 days of pre-trail jail time
  • 365/15 * $790 spent/year = $19K per person-year of jail prevented
• Costs government $450/night to keep BFF people in jail
• About half of big law firms do pro bono work on things like this so it’s pretty crowded [citation needed]

2016-01-17: Org Structure for Progress on Hard Questions

• Brian Tomasik has apparently not studied physics because if he had, he would care more about protons than about electrons
• Effective Altruism Outreach wants questions for essay contests or research
  • They should curate topics to focus on importance rather than interestingness
  • Cause prioritization is important and neglected among EAs, but maybe not the best for an essay contest because it’s dominated by value differences
• How useful are academic conferences?
  • Very useful in math/CS because you learn about colleagues’ work and get new interests
  • People sitting in a room isn’t as good as divvying tasks and sending people off separately
  • Academic conferences can incentivize people to do research because they know there’s demand
• Combine people who we trust and are good at new areas along with domain experts
• It would be useful for SEA to have someone who knows the numbers on topics

References