The new slate star codex survey is out, and they asked people what the probability AI would destroy humanity is. The mean response was 14%, with a median response of 5%. Overall, a fairly reasonable distribution of responses, with a few crazies here and there. Code here.
I used grouping functions and correlation coefficients to determine the zero order associations between variables and percieved AI risk. AIRISK here is measured as the % chance the respondent claims they think AI will destroy humanity by 2100. p-values were assessed with the kruskal test.
Race (yes): p < .01
Middle Easterners and Indians most concerned about AI Risk. Blacks least concerned.
Sex (no): p = .48
Gender (yes): p < .001
Trans women and non-binary most concerned, trans men and cis people least concerned.
Sexual Orientation (yes): p < .001
Bisexuals and other divergents most concerned, Hetero/homosexuals least concerned.
Age (yes): p < .001, (r = -.129)
IQ (no): p =.46, r = .02 (note: the average is 140. Very selected sample.)
SAT scores (yes): p = .013, r = .05 (note: the average is 1485. Very selected sample.)
Relationship style (yes): p < .001
Polyamorous people more concerned than monogamous people.
Field (yes): p < .001
Artists, AI researchers, and mathematicians most concerned of all fieldsmen. Economists, neuroscientists, hard scientists, and statisticians are not so concerned.
Political Ideology (yes): p < .001
Conservatives less concerned than others.
Political Spectrum (1 - far left, 10 - far right | yes): p < .01
Far left very concerned, others not so much.
Having different political opinions from 10 years ago (yes): p < .001
Changing opinions predicts differences in general.
“Unhappiness” (derived from mood, anxiety, life satisfaction, romantic satisfaction, job satisfaction, social satisfaction, suicidality | yes): p < .001
r = .08
Having an inner voice (yes): p < .001
Preference for Humanities over STEM (1 - like STEM, 5 - like humanities |yes): p < .001
People who like humanities are less likely to believe in AI risk.
Thinking you had long COVID (yes): p < .01
Trusting the media (yes): p < .001
r = -.101 (more trust ←→ less risk)
BMI (yes): p < .001
Controlling for variables:
All variables were tested to see whether they held water controlling for media trust / unhappiness / BMI / age.
The following did not pass significance after these 4 controls:
Race
Gender
Sex
Having an inner voice
Interest in STEM
IQ
SAT
The following still passed significance after these 4 controls:
Moral Views (theory: Eliezer influence)
Sexual Orientation (theory: confounded by openness and mental illness)
Being poly (theory: confounded by the degree to which you are down the LW rabbit hole)
Profession (theory: interest in AI causes stronger positions on this issue, not sure why mathematicians are so afraid)
Political ideology (theory: leftists more concerned with acceleration, rightists not as much)
Political spectrum (theory: leftists more concerned with acceleration, rightists not as much)
Changing political views (theory: confounded by openness)
Unhappiness (theory: crazy person → crazy beliefs)
Trust in media (theory: people who don’t trust media will take anti-establishment positions)
BMI (theory: covaries with anxiety/status seeking)
Age (theory: AI risk is a new concern, older individuals are not as interested)
Having long COVID (p = .03) (theory: crazy person → crazy beliefs)
Who believes AI Will Destroy Humanity?
I am curious if there is a correlation between taking Adderall regularly and worrying about AI risk.
Hi Sebastian, I realized that I once saved links to 2 articles you wrote on Substack about IQ & race but they seem to have disappeared, are they still available anywhere ? Thank you