Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio.
This is: EA Survey 2020: Cause Prioritization, published by David_Moss on the AI Alignment Forum.
Summary
Global Poverty is the highest rated cause overall
We found support for longtermist and meta causes to increase with higher self-reported engagement in EA
We also observed higher support for neartermist causes in non-male respondents across engagement levels, though there was no gender difference in support for longtermist causes among more engaged respondents
Comparing ratings across separate EA surveys, we observe a decrease in support for global poverty over time, and an increase in support for animal welfare and AI risk
Full Scale Responses
We asked respondents to evaluate a series of different causes[1] on the same 5 point scale as used in previous years.[2]
Comparing mean ratings across causes
As the same participants rated multiple causes, we used a linear mixed model, with respondent as a random factor, to compare the mean ratings of different causes.
This confirms that the mean rating for Global Poverty is significantly higher than all other causes, followed by Cause Prioritization and Reducing Risks from AI.
AI Risk and Cause Prioritisation receive very similar mean ratings to each other and are the second most highly ranked.
Biosecurity, Climate Change, EA Movement Building and Existential Risk (other) all receive similar ratings (although Biosecurity and Existential Risk (other) do differ significantly[3]).
Animal Welfare and Broad Longtermism are the next highest ranked, followed by Nuclear Security and Meta (not movement building) and Mental Health.
Top Cause Percentages
As in previous years, we report the percentages of respondents rating each cause as the ‘top cause.’ Of course, this is essentially just looking at one level of the responses shown in the graph above. While this leads to simple headline findings (e.g. which cause has the most people rating it ‘top’), it is likely more informative to look at the full range of responses.
One result that may be of particular interest, however, is support for Biosecurity (and pandemic preparedness), given the pandemic. This increased from 4% in 2018 and 2019 to 6% in 2020.
Relationships Between Causes
We explored relationships between ratings of different causes by conducting an exploratory factor analysis. This procedure aims to identify latent factors underlying the data (for example, support for longtermism might be related to support for a number of different longtermist causes).
Across a series of different models, we identified three factors underlying the data, with the same causes associated with them to similar degrees. These were:
Note: Broad longtermism and Animal Welfare only very weakly loaded onto their respective factors.
Exactly how to interpret each factor is, of course, somewhat open to debate, but we think that responses to the causes in each of these different groupings can, to a significant degree, be thought of as reflecting common factors, such as support for existential risk reduction causes or neartermist causes.
Predictors of cause ratings
To simplify analysis, we reduced the ratings of the individual causes above into the three groupings identified by our EFA above (‘longtermist’ ‘meta’ and ‘neartermist), by averaging the scores for each of those categories. As before, we used a linear mixed model to account for the fact that respondents each rated multiple causes. We provide the raw average ratings for each individual cause in the tables in Appendix 2.
Another decision we took to simplify the model was to only examine the influence of people’s level of self-reported engagement with EA and gender.
The first plot below simply shows the mean ratings for each of the three broad cause areas across engagement levels. As we can see, average support for Near-termist causes declines with increa...
view more