The ITN (Impact, Tractability, Neglectedness) framework is widely used in EA to do cause prioritisation . Which could be other metrics that we might be missing out on? Is the ITN framework exhaustive?
Hide table of contents
Comment Permalink
Reactions
New Answer
New Comment
6 Answers sorted by
INT = good per dollar by definition (when used in the quantitative rather than heuristic way), so in that sense it's exhaustive, though in practice, people often miss some factors that are not as naturally captured by the framework:
- People often assess 'importance' just based on one yardstick (e.g. QALYs) when there are effects on other relevant metrics (e.g. x-risk reduction, economic growth) (and the choice of which yardsticks to focus on in the first place is where a lot of the action is).
- Value of information - can be included in either I or N, but often missed.
- Coordination considerations e.g. portfolio approach, comparative advantage, trade with people with other values.
- Cross-cutting epistemic considerations such as regression to the mean & epistemic humility & how to deal with unmeasured factors – partially covered in my recent podcast. People often only report their 'unadjusted' estimates and don't account for these.
- Movement building effects e.g. promoting one cause might bring people into others.
- Funging e.g. if you solve one problem, it might free up resources to work on another.
I find it useful to try to make a 'direct' estimate using INT, and then to have a separate 'all considered' estimate that aims to take account of all the above.
The application of INT also gets more complicated depending on whether you're interested in the resources spent in a certain year or over all time. Likewise, there are issues like complementarities between different forms of resources (e.g. funding vs. labour) that can mean the analysis is different for different resources.
You could also have another category of timing considerations, such as these. Toby's soon, sharp, sudden framework helps to capture some of timing factors as well, or you could think of them as guides to what's most important.
As an alternative, I think it's also useful to think of cost-effectiveness analysis of specific interventions as a separate framework that provides a different perspective.
INT is also only about how pressing causes are in general. In practice if you're making a real decision, you also need to consider your personal fit, career capital, the quality of the specific opportunity etc. as well as other moral considerations besides good done.
I personally don't find the ITN framework useful and agree with most of John Halstead's criticism of the framework here. Cost-effectiveness seems better if you want to make something numerical, within a particular cause area.
In fact, for me, I think both cost-effectiveness and ITN as intellectual frameworks fall down because they mask what I see as a fundamentally philosophical set of questions, if you're trying to do something like compare mental health with health interventions with animal welfare with more longtermist interventions.
If there's an underlying question: how does one prioritise causes? From my own perspective, I just gravitated towards a more longtermist cause prioritisation over time as I found the arguments quite convincing, but I agree that many other areas are extremely worthwhile.
Other factors
If you are thinking about what cause you should work on, you may also consider personal fit. 80,000 hours explain why they didn’t include it as a factor here. Also, none of us is 100% altruistic, so you may also want to consider the personal benefit of working on the cause, although I guess that can go under personal fit as well.
Is it exhaustive?
You could say that there are two ITN frameworks: informal and quantitative. It’s easier to talk about the quantitative framework, so that’s what I will talk about, even though people usually use the informal one.
The quantitative framework cancels out to Good done / extra person or $. If you are a pure consequentialist, I think that this is exhaustive by definition. It doesn’t capture non-consequentialist concerns. E.g., maybe making cost-effective progress on this cause would involve morally questionable means like lying or blackmail. However, you can incorporate these by redefining tractability to something like “tractable with only using means I am comfortable with”. Or you can just assume that in the long run, using these means is bad from a consequentialist point of view anyway, which usually seems to be the case.
Also, it’s exhaustive only if Good done includes all things you intrinsically value like personal benefit, equity, etc. Usually when people use the framework, they assume pure utilitarianism and don’t include these.
I feel I should also mention that I personally find the framework unnecessary and limiting. My opinion is that we don't need any framework here. I find that it's easier and more productive to simply think about what actions I can take and what consequences those actions will lead to. But this is a bit off-topic and I will explain my view in full another time.
Toby Ord writes about prioritizing risks that are soon, sudden and sharp in the Precipice.
Also see a criticism of the current ITN framework and a proposal for a new one by John Halstead.
If I can be forgiven for tooting my own horn, I also wrote a forum post about the framework around the same time as John posted his. EAs have often talked about "cause prioritisation" as being distinct from "intervention evaluation": the former is done in terms of ITN, the latter in term of cost-effectiveness. I agree with Ben Todd's suggestion the best way to understand ITN is as three factors that combine to a calculation of cost-effectiveness (aka "good done per dollar"). One result of this that I think it's conf...
Given a set of values, I see there as being multiple layers of heuristics, which are all useful to consider and make comparisons based on:
- Yardsticks (e.g. x-risk, qualys)
- Causes (e.g. AI alignment)
- Interventions (e.g. research into the deployment problem)
- Specific jobs /orgs (e.g. working at FHI)
Comparisons at all levels are all ultimately about finding proxies for expected value relative to your values.
The cause level abstraction seems to be especially useful for career planning (and grantmaking) since it helps you get career capital that builds up in a useful area. Intervention selection usually seems too brittle. Yardsticks are too broad. This post is pretty old but tries to give some more detail: https://80000hours.org/2013/12/why-pick-a-cause/
4
I think it's probably the case that good heuristics for making career decisions are different than good heuristics for making donation decisions. We shouldn't necessarily expect a framework (ITN or otherwise) to be ideal for both.
If someone today decides to work on a certain cause, they strengthen the pipeline of good funding opportunities in that cause. But there's a time lag. Pivoting to work on biosecurity might be a great career decision right now. However funding a person to do that work might not be a great donation until a few years down the road, when they've gained the skills and credentials needed to make an impact.
Other perspectives that are arguably missing or extensions that can be done are:
- Side effect analysis and modelling the entire future trajectory (generalizations of these can be created):
- https://www.lesswrong.com/posts/BfKQGYJBwdHfik4Kd/fai-research-constraints-and-agi-side-effects
- https://forum.effectivealtruism.org/posts/TCxik4KvTgGzMowP9/state-space-of-x-risk-trajectories
- Refining our estimates of value given uncertainties about the future:
- Specialization for a particular problem, with ensuing accuracy improvements can be done by causal analysis, Fermi modelling, Bayesian modelling, and inspired by https://forum.effectivealtruism.org/posts/zdAst6ezi45cChRi6/list-of-ways-in-which-cost-effectiveness-estimates-can-be
- The value to action chain idea can be used to substantiate value more clearly: https://forum.effectivealtruism.org/posts/Ekzvat8FbHRiPLn9Z/the-values-to-actions-decision-chain-a-lens-for-improving
- One can even try to derive things directly from expected utility all the way to actions but the mathematics gets pretty complex quickly (I did some work on this in 2015 and am sitting on a draft paper that someday somewhere I'll publish. It is beyond my current ability to efficiently explain clearly to a large audience.)
Here also is an additional post analyzing the ITN framework: https://forum.effectivealtruism.org/posts/fR55cjoph2wwiSk8R/formalizing-the-cause-prioritization-framework
Comments1
I'm not sure if it directly answers your question, but this question did finally lead me to write the post about the stock issues framework (which seems to be listed in the pingbacks). I hope that is relevant to your question!
Curated and popular this week
·
Once we expand to other star systems, we may begin a self-propagating expansion of human civilisation throughout the galaxy. However, there are existential risks potentially capable of destroying a galactic civilisation, like self-replicating machines, strange matter, and vacuum decay. Without an extremely widespread and effective governance system, the eventual creation of a galaxy-ending x-risk seems almost inevitable due to cumulative chances of initiation over time across numerous independent actors. So galactic x-risks may severely limit the total potential value that human civilisation can attain in the long-term future. The requirements for a governance system to prevent galactic x-risks are extremely demanding, and they need it needs to be in place before interstellar colonisation is initiated.
Introduction
I recently came across a series of posts from nearly a decade ago, starting with a post by George Dvorsky in io9 called “12 Ways Humanity Could Destroy the Entire Solar System”. It’s a fun post discussing stellar engineering disasters, the potential dangers of warp drives and wormholes, and the delicacy of orbital dynamics.
Anders Sandberg responded to the post on his blog and assessed whether these solar system disasters represented a potential Great Filter to explain the Fermi Paradox, which they did not[1]. However, x-risks to solar system-wide civilisations were certainly possible.
Charlie Stross then made a post where he suggested that some of these x-risks could destroy a galactic civilisation too, most notably griefers (von Neumann probes). The fact that it only takes one colony among many to create griefers means that the dispersion and huge population of galactic civilisations[2] may actually be a disadvantage in x-risk mitigation.
In addition to getting through this current period of high x-risk, we should aim to create a civilisation that is able to withstand x-risks for as long as possible so that as much of the value[3] of the univers
·
Thank you to Arepo and Eli Lifland for looking over this article for errors.
I am sorry that this article is so long. Every time I thought I was done with it I ran into more issues with the model, and I wanted to be as thorough as I could. I’m not going to blame anyone for skimming parts of this article.
Note that the majority of this article was written before Eli’s updated model was released (the site was updated june 8th). His new model improves on some of my objections, but the majority still stand.
Introduction:
AI 2027 is an article written by the “AI futures team”. The primary piece is a short story penned by Scott Alexander, depicting a month by month scenario of a near-future where AI becomes superintelligent in 2027,proceeding to automate the entire economy in only a year or two and then either kills us all or does not kill us all, depending on government policies.
What makes AI 2027 different from other similar short stories is that it is presented as a forecast based on rigorous modelling and data analysis from forecasting experts. It is accompanied by five appendices of “detailed research supporting these predictions” and a codebase for simulations. They state that “hundreds” of people reviewed the text, including AI expert Yoshua Bengio, although some of these reviewers only saw bits of it.
The scenario in the short story is not the median forecast for any AI futures author, and none of the AI2027 authors actually believe that 2027 is the median year for a singularity to happen. But the argument they make is that 2027 is a plausible year, and they back it up with images of sophisticated looking modelling like the following:
This combination of compelling short story and seemingly-rigorous research may have been the secret sauce that let the article to go viral and be treated as a serious project:To quote the authors themselves:
It’s been a crazy few weeks here at the AI Futures Project. Almost a million people visited our webpage; 166,00
·
Note: This post was crossposted from the Open Philanthropy Farm Animal Welfare Research Newsletter by the Forum team, with the author's permission. The author may not see or respond to comments on this post.
----------------------------------------
> Despite setbacks, battery cages are on the retreat
My colleague Emma Buckland contributed (excellent) research to this piece. All opinions and errors are mine alone.
It’s deadline time. Over the last decade, many of the world’s largest food companies — from McDonald’s to Walmart — pledged to stop sourcing eggs from caged hens in at least their biggest markets. All in, over 2,700 companies globally have now pledged to go cage-free.
Good things take time, and companies insisted they needed a lot of it to transition their egg supply chains — most set 2025 deadlines to do so. Over the years, companies reassured anxious advocates that their transitions were on track. But now, with just seven months left, it turns out that many are not.
Walmart backtracked first, blaming both its customers and suppliers, who “have not kept pace with our aspiration to transition to a full cage-free egg supply chain.” Kroger soon followed suit. Others, like Target, waited until the last minute, when they could blame bird flu and high egg prices for their backtracks.
Then there are those who have just gone quiet. Some, like Subway and Best Western, still insist they’ll be 100% cage-free by year’s end, but haven’t shared updates on their progress in years. Others, like Albertsons and Marriott, are sharing their progress, but have quietly removed their pledges to reach 100% cage-free.
Opportunistic politicians are now getting in on the act. Nevada’s Republican governor recently delayed his state’s impending ban on caged eggs by 120 days. Arizona’s Democratic governor then did one better by delaying her state’s ban by seven years. US Secretary of Agriculture Brooke Rollins is trying to outdo them all by pushing Congress to wipe out all stat
I feel I should also mention that I personally find the framework unnecessary and limiting. My opinion is that we don't need any framework here. I find that it's easier and more productive to simply think about what actions I can take and what consequences those actions will lead to. But this is a bit off-topic and I will explain my view in full another time.