I’m building tools to make sense of the future at Sage. Currently building AI Digest, Fatebook and Quantified Intuitions.
Previously I was doing a PhD in HCI at St Andrews, and worked at Clearer Thinking.
Website: https://binksmith.com
Tweeting, sometimes about EA: https://twitter.com/adambinksmith
Hi Tom! I think this idea of giving based on the signalling value is an interesting one.
One idea—I wonder if you could capture a lot of the signalling value while only moving a small part of your donation budget to non-xrisk causes?
How that would work: when you’re talking to people about your GWWC donations, if you think they’d be more receptive to global health/animal ideas you can tell them about your giving to those charities. And then (if you think they’d be receptive) you can go on to say that ultimately you think the most pressing problems are xrisks, and therefore you allocate most of your donations to building humanity’s capacity to prevent them.
In other words, is the signalling value scale-insensitive (compared to the real-world impact of your donations)?
Quick meta note to say I really enjoyed the length of this post, exploring one idea in enough detail to spark thoughts but high readable. Thank you!
You might be aware of this but for others reading - there’s a calculator to help you work out the value of your time.
I think it’s worth doing once (and repeating when your circumstances change, e.g. new job), then just using that as a general heuristic to make time-money tradeoffs, rather than deliberating every time.
If I was an EA grantmaker, I’d want to start small by maybe hiring an educational-youtube-video personality (like John Green’s “Crash Course”) to make an Effective Altruism series.
I think this is in the works! Kurtzegat got a $2.8m grant from Open Phil.
See also A Happier World and Rational Animations.
Great post, thank you! This is useful as a guide to what to try and add in to intro fellowships, in particular:
There are a lot of real professional people in EA, and those people are influencing things in the real world – EA is by no means just a philosophy discussion club, even if your local EA club is one (and it does not have to be one forever!)
I think this is a really important realisation to have as someone doing an intro fellowship/getting into EA. My guess is that realising this makes it a lot easier to think seriously about making career choices based on ideas/methods from EA.
So, how can we help new people realise this sooner?
A quick brainstorm:
Include some readings/podcasts in intro fellowships where people talk in the first person about their EA-aligned work
Encourage new members to attend EAG(x)
Have talks/Q&As with people currently doing EA-aligned work
Include a few bios of individuals and their stories of getting into this kind of work
Chat to new members about what previous members of your group have gone on to do (if your group is mature enough)
I think it’d be ideal if, people understand that EA is not just a philosophy discussion group, but a thing that they could shape their career around, from when they first learn about it.
Thanks!
Hi! It’s a while since you posted this, I was curious if you did see an effect on fellowship signups as a result?
I agree—and if the multiplier numbers are lower, then some claims are don’t hold, e.g.:
To get more than 50% of her maximum possible impact, Ana must hit every single multiplier.
This doesn’t hold if the set of multipliers includes 1.5x, for example.
Instead we might want to talk about the importance of hitting as many big multipliers as possible. And being willing to spend more effort on these over the smaller (e.g. 1.1x) ones.
(But want to add that I think the post in general is great! Thanks for writing this up!)
I just came across a paper which mentions a loosely related method—pairwise rating for model elicitation. See p13 of this PDF (or ctrl-F for pairwise), might be of interest:
Thanks, I’m using Chrome on Windows 10.
Thanks for the writeup! I tried to copy and paste the simple model into Squiggle but it gave an error, couldn’t immediately spot what the cause is:
Prestigious forecasting tournaments for students
Epistemic institutions, empowering exceptional people
To scale up forecasting efforts, we will need a large body of excellent forecasters to recruit from. Forecasting is a skill that improves over time, and it takes time to build a track record to distinguish excellent forecasters from the rest—particularly on long-term questions. Additionally, forecasting builds generally useful research and rationality skills, and supports model-building and detailed understanding of question topics. Therefore, getting students to forecast high-impact questions might be particularly useful for both students own development and the development of the forecasting community.
While existing forecasting platforms allow students to participate, the prestige and compensation offered by success is limited, especially outside of the narrow forecasting community.
We would be excited to fund highly prestigious forecasting tournaments for students, similar to the Maths Olympiad and IGEM in that it would aim to attract top talent, while being focussed on highly impactful questions. A second option is working with universities to give course credit for participation and success in the tournaments. In either case—excellent student forecasters would be rewarded by a prestigious marker on their CV, and fast-tracked application to superforecasting organisations.
Align university careers advising incentives with impact
Effective altruism
Students at top universities often have lots of exposure to a limited set of career paths, such as consulting and finance. Many graduates who would be well-suited to high-impact work don’t consider it because they are just unaware of it. Universities have little incentive to improve this state of affairs, as the eventual social impact of graduates is hard to evaluate and has little effect on their alma mater (with some notable exceptions). We would therefore be excited to fund efforts to more directly align university incentives with supporting their students to enter high-impact careers. We would be interested in work identifying simple heuristic metrics of career impact, and lobbying efforts to have university league tables incorporate these measures into their rankings, rewarding universities who support students in entering impactful work.
EA Founders Camp
Effective altruism, empowering exceptional people
The EA community is scaling up, and funding ambitious new projects. To support continued growth of new organisations and projects, we would be excited to fund an organisation to run EA Founders Camps. These events would provide an exciting, sparky environment for (1) Potential founders to meet co-founders, (2) Founders to hear about and generate great ideas for impactful projects and organisations, (3) Founders to get key training tailored to their project area, (4) Founders to build a support network of other new and existing founders, (5) Founders to connect with funders and advisers.
Find promising candidates for “Cause X” with an iterative forecast-guided thinktank
Epistemic institutions
How likely is it that the EA community is neglecting a cause area that is more pressing than current candidates? We are fairly confident in the importance of the community’s current community areas, but we think it’s still important to keep searching for more candidates.
We’d be excited to fund organisations attacking this problem in a structured, rigorous way, to reduce the chance that the EA community is missing huge opportunities.
We propose an organisation with two streams: generalist research, and superforecasting. The generalist researchers create shallow, exploratory evaluations of many different cause areas. Forecasters then use these evaluations to forecast the likelihood of each cause area being a top cause area recommended (e.g. by 80,000 Hours) in 5 years time. The generalist researchers then perform progressively more in-depth evaluations of the cause areas most favoured by forecasters. Forecasters update their forecasts based on these evaluations. If the forecasted promising-ness exceeds a threshold, the organisation recommends that an EA funder funds in-depth research into the cause area.
Help high impact academics spend more time doing research
Empowering exceptional people
Top academic researchers are key drivers of progress in priority areas like biorisk, global priorities research and AI research. Yet even top academics are often unable to spend as much time as they want to on their research.
We’d be excited to fund an organisation providing centralised services to maximise research time for top academics, while minimising the overheads of setting up these systems for academics. It might focus on:
(1) Funding and negotiating teaching buy-outs,
(2) Providing an efficient shared team of PAs to handle admin, streamline academic service duties, submit papers, scout and screen PhD students, and accelerate literature surveys.
As AI research assistants like Elicit improve, this organisation could scalably offload work to these services.
The GitHub Archive Program probably isn’t quite what you’re looking for, but I think is interesting. It’s not historical, and it does have some short-term effects (especially publicity).
This led me to think about the fact that the description of resilience is itself an estimate/prediction. I wonder how related the skills of giving first-order estimates/predictions and second-order resilience estimates are. In other words, if someone is well-calibrated, can we expect their resilience estimates to also be well-calibrated? Or is an extra skill that would take some learning.
To add to your list—Subjective Logic represents opinions with three values: degree of belief, degree of disbelief, and degree of uncertainty. One interpretation of this is as a form of second-order uncertainty. It’s used for modelling trust. A nice summary here with interactive tools for visualising opinions and a trust network.
Great to see tools like this that make assumptions clear—I think not only useful as a calculator but as a concrete operalisation of your model of AI risk, which is a good starting point for discussion. Thanks for creating!