Sentient Futures
Back to proposals

Two reviews: (1) AI X-Risk expert assessment (2) AI consciousness cruxes

commitment
10 hrs/week
flexibility
Open to proposals
mentored by
Max Miller
Max MillerIndependent

The project

AI X-risk expert assessment

Hop on to dig into open avenues. I think AI X-risk concerns need more weighted evidence so that other concerns e.g. AI and animal welfare don't get neglected. These avenues are:

  1. Looking into the implications of conceptual work on instrumental convergence (there has been conceptual work on this topic which hasn't had much elaboration).
  2. Updating threat models and instrumental convergence based on the latest AI models (e.g. Fable): whilst Fable is a massive boost, again, we need to ground the "hype" in existing frameworks. We've got those frameworks, but just need to update them.
  3. Broader: writing up the extent of AI attack vector (Nuclear, biological etc) autonomy based on the latest models.
  4. Extending our section on AGI forecasting with more in-depth critiques.
  5. Less clear but: Finally, our report was focused on strict AI X-risk concerns, but I'd like to counter them with the AI moral consideration dimension. Relatedly, see project 2.

Findings would be added into the 200pp assessment, on Karlsruhe University's website, and arXiv (or another repository).

AI consciousness cruxes

We've (our SF incubator team) applied to the Models of Consciousness conference in Copenhagen and would love to have some help continuing that work.

  • This work involves: systematic review on AI moral consideration (I am very versed in systematic review and have transferable skills people can greatly benefit from!), and, optionally, interviews (here as well, transferable skills from my Psych degree and outreach/part. recruitment strategy).
  • A great opp. for mentees to get familiar with AI consc. straight from expert interviewees (I've found this fascinating).

Intermediate findings would be published as a pre-print and hopefully shared at the conference as a poster (October).
Given the nice grounding of both projects, I'm currently looking for funding too!

The mentee's role

All projects involve going through a systematic review process (AI-assisted, very accessible though). For 1, a. we have an existing but likely outdated literature collection:

  1. Determine paper databases
    1. Scopus, IEEE Xplore , the Association for Computing Machinery (ACM), PubMed, EBSCOHost and Web of Science, arXiv
    2. Any way of searching other sources?
  2. Determine boolean search string for your topic (effort) (use AI)
  3. Store in reference manager (Zotero, Mendeley...) (effort if haven't used ref. Manager before)
    1. Remove duplicates
    2. Retrieve missing metadata: notably Abstract
  4. ASReview:
    1. 10 "warm-up" articles
    2. Screen up to 100 abstracts (3h)
  5. Read and summarise papers

Project 2 may require more than 10h/week for a deadline (e.g. conference).

Who I'm looking for

  • Intrinsic motivation/taking ownership: Even though both projects depart from pre-existing work, their scopes are open. This isn't consultant work, I'd prefer for whichever project to fit into your cause-prioritisation/interests! Project 1 may go on longer, it's part of a broader cause-prioritisation fascination of mine; would be great to have a longer-term team member to tackle this puzzle with.
  • Little AI use/ writing skills: It's much better for writing to be humble/less jargon-y than convoluted and repeating pre-existing confusion. In my experience, AI just perpetuates unclear claims. Regular writing skills e.g. topic sentence/concluding sentence, starting summary, final conclusion are important. Critical thinking: generating your own critiques and not just reporting abstract figures is very useful too. The "expert assessment"/conference outlets may sound daunting but this actually calls for more humility and relaxed writing; these fields have enough flashy jargon. If you're intrinsically motivated all of this would show in writing anyway!
  • Remote work etiquette: Important to be mindful of the challenges involved with asynchronous remote work (see Support I can offer mentees).

Questions for applicants

  1. Which timezone are you in?
  2. More for yourself to prevent any future stress: is there any likelihood you may not have time during the project? (e.g. ongoing previous time commitments etc. For myself: I'll be moving some time between Sept-beginning Nov. which may take me out for a week or so). 

Support offered

  • Very versed in AI-assisted systematic review and have transferable skills people can greatly benefit from
  • Interviews: transferable skills from my Psych degree and outreach/part. recruitment strategy
  • Psychological hurdles of remote work and impostor syndrome: I've been working fully-remote and in teams for the past 2 years so have a good ecosystem for this kind of work. I personally love to always be on Gather Town when working with team members. If you hadn't heard of it, Gather is a sort of "Metaverse" like that of Facebook but just not in VR form. You get your own virtual office and can walk around in it, when you approach people both individuals' webcams turn on so it simulates real-life. I like to work on there as one can brainstorm, rant, rejoice... all the things which are otherwise bottled up when working in isolation. Am good with meeting notes and responsiveness on Slack (using emojis, setting statuses.. anything which creates a sense of team, community and life online.
format
Research & writing · Project scoping
topic
Artificial sentience · Longtermism · Other · Macrostrategy · Near-term impact · Post-AGI transition
Max Miller

Max Miller

Independent

  1. AI X-risk "scout mindset": have been interested in verifying the assumptions underpinning as-yet-speculative AI X-risk threat models (through my MRes thesis and a 200pp expert assessment for Karlsruhe University for which I was PI as part of a team).
  2. In parallel: flipping this anthropocentric "AI X-risk" frame, I am concerned about potential AI welfare considerations. Have been doing a systematic review and interviews regarding the evidentiary thresholds for AI moral consideration with a previous SF team. We've applied to the Models of Consciousness conference in Copenhagen and would love to have some help continuing that work.