CEO of Convergence.
David_Kristoffersson
Thanks for your detailed comment, Max!
Relative to my own intuitions, I feel like you underestimate the extent to which your “spine” ideally would be a back-and-forth between its different levels
I agree, the “spine” glosses over a lot of the important dynamics.
I think I would find it easier to understand to what extent I agree with your recommendations if you gave specific examples of (i) what you consider to be valuable past examples of strategy research, and (ii) how you’re planning to do strategy research going forward (or what methods you’d recommend to others).
Very good points. Both would indeed be highly valuable to the argument. As follow up posts, I’m considering writing up (1) concrete projects in strategy research that seem valuable, and (2) a research agenda.
While I agree that we face substantial strategic uncertainty, I think I’m significantly less optimistic about the marginal tractability of strategy research than you seem to be.
Yeah, we’re more optimistic than you here. I don’t think it’s possible to do useful completely “tactics and data free” strategy research. But I do think there is highly valuable strategy research to do that can be grounded with a smaller amount of tactics and data gathering.
What tactics research and data gathering is key? I think this is a strategic question and I think we’re currently just scratching the surface.For example, while I tend to be excited about work that, say, immediately helps Open Phil to determine their funding allocation, I tend to be quite pessimistic about external researchers sitting at their desks and considering questions such as “how to best allocate resources between reducing various existential risks” in the abstract.
I agree that it seems like that could easily be a bad use of time for “external researchers” to do that. I’m somewhat optimistic about these researchers examining sub-questions that would inform how to do the allocation.
Very loosely, I expect marginal activities that effectively reduce strategic uncertainty to look more like executives debating their companies strategy in a meeting rather than, say, Newton coming up with his theory of mechanics. I’m therefore reluctant to call them “research”.
I think the idea cluster of existential risk reduction was formed through something I’d call “research”. I think, in a certain way, we need more work of this type. But it also needs to be different in some important way in order to create new valuable knowledge. We hope to do work of this nature.
Indeed! We hope we can deliver that sooner rather than later. Though foundational research may need time to properly come to fruition.
AI and X-risk Strategy Unconference at EA Hotel in November
Excellent analysis, thank you! The issue definitely needs a more nuanced discussion. The increasing automation of weaponry (and other technology) won’t be stopped globally and pervasively, so we should endeavor to shape how it is developed and applied in a more positive direction.
The ‘far future’ is not just the far future
Thank you for your thoughtful comment!
All work is future oriented Indeed. You don’t tend to employ the word ‘future’ or emphasize it for most work though.
One alternative could be ‘full future’, signifying that it encompasses both the near and long term.
I think there should be space for new and more specific terms. ‘Long term’ has strengths, but it’s overloaded with many meanings. ‘Existential risk reduction’ is specific but quite a mouthful; something shorter would be great. I’m working on another article where I will offer one new alternative.
Good point, ‘x-risk’ is short and ‘reduction’ should be or should become implicit after some short steps of thinking. It will work well in many circumstances. For example, in “I work with x-risk”, just as “I work with/in global poverty” works. Though some interjections that occur to me in the moment are: “the cause of x-risk” feels clumsy, “letter, dash, and then a word” feels like an odd construct, and it’s a bit negatively oriented.
I’m sympathetic to many of the points, but I’m somewhat puzzled by the framing that you chose in this letter.
Why AI risk might be solved without additional intervention from longtermist
Sends me the message that longtermists should care less about AI risk.
Though, the people in the “conversations” all support AI safety research. And, from Rohin’s own words:
Overall, it feels like there’s around 90% chance that AI would not cause x-risk without additional intervention by longtermists.
10% chance of existential risk from AI sounds like a problem of catastrophic proportions to me. It implies that we need many more resources spent on existential risk reduction. Though perhaps not strictly on technical AI safety. Perhaps more marginal resources should be directed to strategy-oriented research instead.
I’m not arguing “AI will definitely go well by default, so no one should work on it”. I’m arguing “Longtermists currently overestimate the magnitude of AI risk”.
Thanks for the clarification Rohin!
I also agree overall with reallyeli.
The long term future is especially popular among EAs living in Oxford, not surprising given the focus of the Global Priorities Institute on longtermism
Even more than that, The Future of Humanity Institute has been in Oxford since 2005!
Thanks Tobias, I think you make a really good point! You’re definitely right that there are some in the cause area who don’t think the technological transformation is likely.
I don’t think you’ve established that the ‘technological transformation’ is essential.
What I wanted to say with this post is that it’s essential to the view of a large majority in the cause area. The article is not really meant to do a good job at arguing that it should be essential to peoples’ views.
It’s possible I’m wrong about the size of the majority; but this was definitely my impression.
You may believe that shaping AI / the technological transformation would offer far more leverage than other interventions, but some will disagree with that, which is a strong reason to not include this in the definition.
I personally believe in something like a mix of shaping technology with other interventions, such as the ones you mentioned (“moral circle expansion, improving international cooperation, improving political processes (e.g. trying to empower future people, voting reform, reducing polarisation”).
State Space of X-Risk Trajectories
I think that if I could unilaterally and definitively decide on the terms, I’d go with “differential technological development” (so keep that one the same), “differential intellectual development”, and “differential development”. I.e., I’d skip the word “progress”, because we’re really talking about something more like “lasting changes”, without the positive connotations.
I agree, “development” seems like a superior word to reduce ambiguities. But as you say, this is a summary post, so it might not the best place to suggest switching up terms.
Here’s two long form alternatives to “differential progress”/”differential development”: differential societal development, differential civilizational development.
But beyond the trajectories (and maybe specific distances), are you planning on representing the other elements you mention? Like the uncertainty or the speed along trajectories?
Thanks for your comment. Yes; the other elements, like uncertainty, would definitely be part of further work on the trajectories model.
Happy to see you found it useful, Adam! Yes, general technological development corresponding to scaling of the vector is exactly the kind of intuition it’s meant to carry.
This kind of complexity tells me that we should talk more often of risk %’s in terms of the different scenarios they are associated with. E.g., the form of current trajectory Ord is using, and also possibly better (if society would act further more wisely) and possible worse trajectories (society makes major mistakes), and what the probabilities are under these.
We can’t disentangle talking about future risks and possibilities entirely from the different possible choices of society since these choices are what shapes the future. What we do affect these choices.
(Also, maybe you should edit the original post to include the quote you included here or parts of it.)
I think this is an excellent initiative, thank you, Michael! (Disclaimer: Michael and I work together on Convergence.)
An assortment of thoughts:
More and more studious estimates of x-risks seem clearly very high value to me due to how much the likelihood of risks and events affect priorities and how the quality of the estimates affect our communication about these matters.
More estimates should generally should increase our common knowledge of the risks, and individually, if people think about how to make these estimates, they will reach a deeper understanding of the questions.
Breaking down the causes of one’s estimates is generally valuable. It allows one to improve one’s estimates, understanding of causation, and to discuss them in more detail.
More estimates can be bad if low quality estimates swamp out better quality ones somehow.
Estimates building on new (compared to earlier estimates) sources of information are especially interesting. Independent data sources increase our overall knowledge.
I see space for someone writing an intro post on how to do estimates of this type better. (Scott Alexander’s old posts here might be interesting.)
Thank you for this article, Michael! I like seeing the different mainline definitions of existential risk and catastrophe alongside each other, and having some common misunderstandings clarified.
Just a minor comment:
That said, at least to me, it seems that “destruction of humanity’s longterm potential” could be read as meaning the complete destruction. So I’d personally be inclined to tweak Ord’s definitions to:
An existential catastrophe is the destruction of the vast majority of humanity’s long-term potential.
An existential risk is a risk that threatens the destruction of the vast majority of humanity’s long-term potential.[4]
Ord was presumably going for brevity in his book, and I think his definition succeeds quite well! I don’t think generally adding 4 words to Ord’s short nice definition would be worth it. There’s other details that could be expanded on as well (like how we can mostly consider the definition in Bostrom 2012 to be a more expanded one). Expanding helps with discussing a particular point, though.
Vision of Earth fellows Kyle Laskowski and Ben Harack had a poster session on this topic at EA Global San Francisco 2019: https://www.visionofearth.org/wp-content/uploads/2019/07/Vision-of-Earth-Asteroid-Manipulation-Poster.pdf
They were also working on a paper on the topic.
Good points.
Perhaps funding organizations would like better ways of figuring out the risks of supporting new projects? I think valuable work could be done here.
Justin Shovelain came up with that. (Justin and I were both on the strategy team of AISC 1.)