I did what I think is the largest piece of research on current and aspiring AI safety folk in Australia & New Zealand. I wanted to understand their career barriers so that I could then optimize my organization's tactics to remove them.
Caveats
- I am not a trained social scientist nor a statistician. There will be errors.
- I’ve budgeted a few hours to write this up. So this is mostly about me posting the results for others to use. I hope they’re useful.
The process
- I conducted 35 long form interviews with local Technical and Governance folk (about an hour each).
- From this qualitative data I identified a long list of barriers to impact.
- I then clustered the long list of barriers into 15 high level ones.
- I also found two reasonably common symptoms of mental-ill health. Not nice TBH.
- I used this list of 17 barriers / symptoms of mental-ill health in a survey. My goal was to measure how intense or common each barrier to impact was using Likert scales.
The survey scales
- Likert scale for Barriers
- Not at all (0)
- Slightly (1)
- Somewhat (2)
- Moderately (3)
- Very (4)
- Extremely (5)
- Likert scale for mental health related questions
- Never (0)
- Rarely (1)
- Sometimes (2)
- Often (3)
- Always (4)
Demo breakdown of survey responders
By career stage
| I am not currently working in an AI Safety role, but I aspire to in the future | 20 |
| I am currently working in an AI Safety role | 17 |
Their ideal or current career stream
| Ideal or current career path | Count |
| Technical work (e.g. research) | 21 |
| Governance and Policy | 13 |
| Movement Building | 3 |
Results
Barriers
Overall barriers to impact (high to low)
| Barrier | Intensity |
| How limited have you been by a lack of *local*, well-paying AI Safety jobs? | 3.11 |
| How much has a lack of mentorship hindered your progress in AI Safety? | 3.03 |
| How confused are you about the amount of upskilling you'd need to be successful in your most desired AI Safety job? | 2.72 |
| How difficult have you found creating and following a career plan (including upskilling) for AI Safety? | 2.56 |
| How limited have you been by a lack of high-quality *local* training programs for AI Safety careers? | 2.49 |
| How challenging have you found navigating the funding ecosystem in AI Safety (this could include grants, PhD programs, etc.)? | 2.47 |
| How unclear are you about where your skills could be best applied in AI Safety? | 2.36 |
| How limited have you been by a lack of strong *local* (i.e. in-person) networks for AI Safety careers? | 2.33 |
| How much have financial limitations hindered your pursuit of an AI Safety career (e.g., taking unpaid time off for upskilling)? | 2.24 |
| How confused are you about the types of paid jobs available in AI Safety? | 2.11 |
| How much has a lack of quality feedback on your AI Safety work hindered your progress? | 2.08 |
| How much has a lack of collaborators for AI Safety projects hindered your progress? | 2.06 |
| How difficult is it for you to stay up-to-date on relevant AI Safety opportunities (jobs, funding, networking, training)? | 1.97 |
| How difficult have you found accessing career advice specific to AI Safety? | 1.75 |
| To what extent have you experienced a desire to contribute to the *local* AI Safety movement, but been confused about how you can contribute (e.g. through volunteering)? | 1.03 |
Governance & Policy career stream: barriers to impact (high to low)
| Barrier (Governance) | Intensity (out of 5) |
| How limited have you been by a lack of *local*, well-paying AI Safety jobs? | 3.46 |
| How confused are you about the amount of upskilling you'd need to be successful in your most desired AI Safety job? | 3.08 |
| How much have financial limitations hindered your pursuit of an AI Safety career (e.g., taking unpaid time off for upskilling)? | 2.92 |
| How limited have you been by a lack of high-quality *local* training programs for AI Safety careers? | 2.77 |
| How challenging have you found navigating the funding ecosystem in AI Safety (this could include grants, PhD programs, etc.)? | 2.69 |
| How difficult have you found creating and following a career plan (including upskilling) for AI Safety? | 2.69 |
| How much has a lack of mentorship hindered your progress in AI Safety? | 2.62 |
| How limited have you been by a lack of strong *local* (i.e. in-person) networks for AI Safety careers? | 2.54 |
| How unclear are you about where your skills could be best applied in AI Safety? | 2.46 |
| How confused are you about the types of paid jobs available in AI Safety? | 2.31 |
| How difficult is it for you to stay up-to-date on relevant AI Safety opportunities (jobs, funding, networking, training)? | 1.85 |
| How difficult have you found accessing career advice specific to AI Safety? | 1.77 |
| How much has a lack of collaborators for AI Safety projects hindered your progress? | 1.69 |
| To what extent have you experienced a desire to contribute to the *local* AI Safety movement, but been confused about how you can contribute (e.g. through volunteering)? | 1.38 |
| How much has a lack of quality feedback on your AI Safety work hindered your progress? | 1.23 |
Technical career stream: barriers to impact (high to low)
| Barrier Technical | Intensity |
| How much has a lack of mentorship hindered your progress in AI Safety? | 3.3 |
| How limited have you been by a lack of *local*, well-paying AI Safety jobs? | 2.81 |
| How confused are you about the amount of upskilling you'd need to be successful in your most desired AI Safety job? | 2.55 |
| How much has a lack of quality feedback on your AI Safety work hindered your progress? | 2.5 |
| How difficult have you found creating and following a career plan (including upskilling) for AI Safety? | 2.5 |
| How limited have you been by a lack of high-quality *local* training programs for AI Safety careers? | 2.33 |
| How challenging have you found navigating the funding ecosystem in AI Safety (this could include grants, PhD programs, etc.)? | 2.3 |
| How unclear are you about where your skills could be best applied in AI Safety? | 2.25 |
| How limited have you been by a lack of strong *local* (i.e. in-person) networks for AI Safety careers? | 2.15 |
| How much has a lack of collaborators for AI Safety projects hindered your progress? | 2.1 |
| How confused are you about the types of paid jobs available in AI Safety? | 2 |
| How much have financial limitations hindered your pursuit of an AI Safety career (e.g., taking unpaid time off for upskilling)? | 1.90 |
| How difficult is it for you to stay up-to-date on relevant AI Safety opportunities (jobs, funding, networking, training)? | 1.8 |
| How difficult have you found accessing career advice specific to AI Safety? | 1.65 |
| To what extent have you experienced a desire to contribute to the *local* AI Safety movement, but been confused about how you can contribute (e.g. through volunteering)? | 0.75 |
Governance vs Technical (differences in barriers to impact)
| Barrier | Governance | Technical | Absolute difference | Proportional difference |
| How much has a lack of quality feedback on your AI Safety work hindered your progress? | 1.23 | 2.50 | 1.27 | 0.68 |
| To what extent have you experienced a desire to contribute to the *local* AI Safety movement, but been confused about how you can contribute (e.g. through volunteering)? | 1.38 | 0.75 | 0.63 | 0.59 |
| How much have financial limitations hindered your pursuit of an AI Safety career (e.g., taking unpaid time off for upskilling)? | 2.92 | 1.90 | 1.02 | 0.42 |
| How much has a lack of mentorship hindered your progress in AI Safety? | 2.62 | 3.30 | 0.68 | 0.23 |
| How much has a lack of collaborators for AI Safety projects hindered your progress? | 1.69 | 2.10 | 0.41 | 0.22 |
| How limited have you been by a lack of *local*, well-paying AI Safety jobs? | 3.46 | 2.81 | 0.65 | 0.21 |
| How confused are you about the amount of upskilling you'd need to be successful in your most desired AI Safety job? | 3.08 | 2.55 | 0.53 | 0.19 |
| How limited have you been by a lack of high-quality *local* training programs for AI Safety careers? | 2.77 | 2.33 | 0.44 | 0.17 |
| How limited have you been by a lack of strong *local* (i.e. in-person) networks for AI Safety careers? | 2.54 | 2.15 | 0.39 | 0.17 |
| How challenging have you found navigating the funding ecosystem in AI Safety (this could include grants, PhD programs, etc.)? | 2.69 | 2.30 | 0.39 | 0.16 |
| How confused are you about the types of paid jobs available in AI Safety? | 2.31 | 2.00 | 0.31 | 0.14 |
| How unclear are you about where your skills could be best applied in AI Safety? | 2.46 | 2.25 | 0.21 | 0.09 |
| How difficult have you found creating and following a career plan (including upskilling) for AI Safety? | 2.69 | 2.50 | 0.19 | 0.07 |
| How difficult have you found accessing career advice specific to AI Safety? | 1.77 | 1.65 | 0.12 | 0.07 |
| How difficult is it for you to stay up-to-date on relevant AI Safety opportunities (jobs, funding, networking, training)? | 1.85 | 1.80 | 0.05 | 0.03 |
Mental ill-health
How often have you felt lonely or disconnected while pursuing an AI Safety career?
Overall
| Total | % | |
| Often (3) | 12 | 33.33% |
| Sometimes (2) | 12 | 33.33% |
| Rarely (1) | 7 | 19.44% |
| Never (0) | 3 | 8.33% |
| Always (4) | 2 | 5.56% |
By career stream
| Governance and Policy | % | Movement Building | % | Technical work (e.g. research) | % | |
| Always (4) | 0 | 0.00% | 0 | 0.00% | 2 | 9.52% |
| Never (0) | 1 | 7.69% | 1 | 33.33% | 1 | 4.76% |
| Often (3) | 5 | 38.46% | 2 | 66.67% | 5 | 23.81% |
| Rarely (1) | 3 | 23.08% | 0 | 0.00% | 4 | 19.05% |
| Sometimes (2) | 4 | 30.77% | 0 | 0.00% | 8 | 38.10% |
How often have you experienced negative emotions (low confidence, self-doubt, fear of failure) while pursuing an AI Safety career?
Overall
| Grand Total | % | |
| Often (3) | 13 | 36.11% |
| Sometimes (2) | 12 | 33.33% |
| Always (4) | 5 | 13.89% |
| Rarely (1) | 5 | 13.89% |
| Never (0) | 1 | 2.78% |
By career stream
| Governance and Policy | % | Movement Building | % | Technical work (e.g. research) | % | |
| Always (4) | 2 | 15.38% | 0 | 0.00% | 3 | 14.29% |
| Often (3) | 4 | 30.77% | 2 | 66.67% | 7 | 33.33% |
| Sometimes (2) | 5 | 38.46% | 0 | 0.00% | 7 | 33.33% |
| Rarely (1) | 1 | 7.69% | 1 | 33.33% | 3 | 14.29% |
| Never (0) | 1 | 7.69% | 0 | 0.00% | 0 | 0.00% |
This seems broadly in line with previous EA Survey results on the main barriers to involvement in EA.
Too few job opportunities / too hard to get an EA job, no close EA friends, and clear how to contribute were among the most commonly mentioned barriers. Lack of a local group and mentorship were also commonly mentioned.
Thanks David; that is good to know from a data standpoint, but also saddening from a human one.
Do you have any thoughts as to the extent online networks (like chat groups, twitter, the EA Forum, online books clubs, or discussion groups like the virtual programs) can influence the feelings of loneliness or disconnectedness?
I have spent about 3 years living in places with no/minimal in-person EA community since I learned about and decided to get involved in EA. I think that online interaction pales in comparison with in-person interaction, but it is a heck of a lot better than nothing. Having regular interaction with a fairly consistent group of people has helped me mitigate feelings of loneliness or disconnectedness, and I think it has been similar for a few other regulars. (although I must admit that time zones are challenging)
Hi Joseph, I'm sorry to hear you've had these challenges. I think movement builders like me should consider this a priority.
"I think that online interaction pales in comparison with in-person interaction, but it is a heck of a lot better than nothing."
I agree with this. I am also exploring opening coworking spaces in Sydney and Melbourne because I think they have a strong theory of change AND will help the symptoms of mental-ill health. I think it gives people an environment to connect but also opens up independant work as a potential career path (since you'll have like-minded people around).
What do you think?
As long as those locations have a sufficient number of interested people to support a physical co-working space, then it seems like a promising potential.
While visiting the EA Forum and reading posts in Slack workspaces helps me feel a sense of connection to the broader EA community, having a group of a few regulars who have a video chat once a month or so has probably been the best for relaxed networking and for forming a sense of community. Seeing the same people repeatedly over time isn't a novel idea, but I mention it because it is so different from the anonymous mass of strangers in most online interactions.
Thanks for flagging that Joseph, i'll put it in the old memory bank for when I'm considering new initiatives. I appreciate all your feedback so far.
Wow thanks for doing this research. Glad to see I'm not alone in what I'm feeling but also not at the same time? On the bright side EA has grown rather rapidly considering it was established in 2009.
Executive summary: A survey of AI safety professionals in Australia and New Zealand reveals significant barriers to impact, including lack of local jobs and mentorship, as well as high rates of loneliness and negative emotions among practitioners.
Key points:
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.
The amount of 'often' and 'always' (edit: on the mental health questions) was higher than I expected. I think this is important, especially if it's representative.
I think it would be helpful to see, qualitatively, what causes these negative emotions.
I'd have selected 'often' as well btw.
Hi quila! I'll be honest - many of the interviews had a tinge of sadness to them. It has motivated me to explore opening co-working spaces in Sydney and Melbourne. Partly because I see a strong theory of change, but also because I think it will help improve the wellbeing of aspiring AI folk (there is ofc overlap between these two things!).
Are you from ANZ?
I meant that I think there should be some larger dedicated effort to solve this directly (maybe by one of the large meta-EA orgs).
No