Review of ITN Critiques
After 5000 years of delay, I have finally taken 10 seconds to dump this post in a community editable Google Doc. Thanks to Vaidehi for her important, tractable and long-neglected suggestion. Please edit and improve here, and I will periodically update this post with changes. It will automatically send me a notification if you edit it.
Note: Although this review draws together reviews of ITN and mentions prominent EA organizations, it takes no position on the validity of these critiques. The purpose is merely to collate, summarize and categorize them for ease of reference. For this review, only the main body of linked posts was used. Brief citations are given at the end. This is a work in progress, and any additional corrections or references are appreciated.
Introduction
The Important, Tractable, Neglected framework is a heuristic or framework for estimating the marginal utility of additional resources applied to an altruistic cause. Personal fit is sometimes added as an additional criterion, depending on the purpose of the cause prioritization effort. It is heavily featured in writing within the Effective Altruist movement, including in the introduction to the movement on EffectiveAltruism.org, the Key Ideas article on 80000Hours.org, and the Focus Areas page of Open Philanthropy. The importance of cause prioritization is a common theme of the EA movement, and the ITN framework is one of the most prominent frameworks for carrying out this analysis.
An array of critiques of the ITN framework has emerged, as well as responses to them, which are categorized and summarized here. Some critiques are theoretical and others are of its application. Theoretical critiques generally examine how ITN diverges from an ideal model of cause prioritization. Critiques of application focus on how specific individuals or organizations use ITN, either for cause prioritization or as a rhetorical device. In reviewing the use of ITN or any other cause prioritization approach, it is useful to keep the organizational or social context in mind.
There are many critiques of cause prioritization generally that are not specific to ITN. Alternatives methods for estimating cost-effectiveness exist [12] [13]. Problems of cause prioritization generally are beyond the scope of this review. When they are mentioned, it is in the context of how ITN leads to blindness or mistaken assumptions around those deeper issues.
The following is a list of critiques of ITN.
Neglectedness improperly assumes diminishing marginal returns
Neglectedness assumes that increased investment in a cause area can suffer from diminishing marginal returns (DMR) to investment [2] [3] [4]. DMR could be represented as a graph showing the relationship of past investment to expected marginal returns to additional investment. Neglectedness as a criterion supposes that additional investment will necessarily decrease the marginal utility of additional investment. While this may be true beyond some threshold specific to a given cause, other effects may be more relevant in context such as:
Setup costs (high expense and little-no value at the beginning) [1]
Payoffs being clustered at the end of an effort [2] [3]. In clustered value causes, it might make more sense to evaluate average marginal returns [2].
Economies of scale (cost of productive units may decrease with investment, even if value produced by each unit decreases) [2] [3]
Ability to take on high risk/high reward projects increasing with investment [2]
Technological development or political and economic change providing additional opportunities for productive investment over time [2]
Economic theory that justifies assumptions of DMR in a profit-maximizing market context may not apply in charitable contexts [8].
Neglectedness prioritizes change over magnitude of marginal returns
One cause may face diminishing marginal returns (MR), yet have high absolute MR. Another might have low absolute MR but enjoy increasing MR, perhaps due to high setup costs. Neglectedness favors the latter cause over the former, which may not be appropriate [2] [3]. This may result in prioritization of smaller causes over larger ones, as small causes may have less to do and therefore may attract less investment [3]. This might be accounted for in measures of Importance, but given the informal heuristic nature of ITN this is not guaranteed.
Neglectedness may mislead our intuitions about what counts as an “investment”
Once an appropriate DMR graph shape and scale is selected, the analyst must decide how to measure the level of investment thus far. Present investments seem clearly to count. Past efforts may have picked the low-hanging fruit, leaving little for further present efforts to accomplish [3]. Yet economic, political, and technological dynamism may reveal opportunities unavailable to past efforts as described above [2] [3]. Some causes may be more or less likely to be accomplished in the future even if they are neglected now, so anticipated future efforts may also count [3] [11].
In some areas such as biological research, it is my view that natural selection should be counted as a “past investment.” While natural selection is not a teleological force, its effect is to enhance the fitness of a species through biochemical change. This overlaps the work of medical research to some extent. Other indirect work or natural forces may be relevant to other cause areas. It is also unclear how we should count directly contradictory efforts, such as pro- and anti-nuclear climate change prevention advocacy [2], overlapping cause areas, or general competition in moral advocacy [3].
Although this concern applies to any cause prioritization effort, not just ITN, the term “neglectedness” may lead analysts to focus too narrowly on present and past deliberate human efforts.
Some translations of ITN into equations have mathematical problems
80,000 Hours converts ITN into an equation:
Scale (of the problem we’re trying to solve) = Good done / % of the problem solved
Solvability = % of the problem solved / % increase in resources
Neglectedness = % increase in resources / extra person or $
In this conversion, “% of the problem solved” and “% increase in resources” would algebraically cancel out, simplifying to “Good done / extra person or $” and making Solvability (another term for Tractability) irrelevant [2] [12].
Prior to simplification, if each factor were evaluated independently, Solvability would be undefined and conceptually “unsolvable” if the problem had received zero investment [2].
Personal fit may be undervalued in some analyses [2].
80,000 Hours assigns numerical rankings to their ITN analysis of causes [5]. Their highest ranked cause is “risks from artificial intelligence” at a combined ITN score of 27. “Developing world health” is ranked 21. In their estimation, each point makes a cause three times as pressing. As “risks from artificial intelligence” is ranked 6 points higher than “developing world health,” it is estimated to be roughly 36, or 729 times as important. Other causes ranked at a score of 20 include extreme climate change risks, land use reform, or smoking in the developing world.
They also use a Personal Fit score from 0-4 which is added to each rank [6].
Taken literally, this advice means that 80,000 Hours encourages career changers working in world health or lower-ranked fields to focus on AI safety, even if they are an actively bad fit. 80,000 Hours advises taking the scores with “a big pinch of salt,” they also say that their scores might be off by “a couple of points.” If AI safety was accurately ranked 25 instead of 27, and and the lower-ranked causes were increased by two points to 22 or 23, personal fit would win out over cause prioritization in this hypothetical case.
Neglectedness improperly assumes irrationality or value divergence of other actors
Greater value overlap and rationality among investors in altruistic causes weakens the relationship between Neglectedness and expected MR [7] [11]. Others may have information we lack. Under these conditions, neglectedness is a potential sign of intractability or unimportance that the analyst has missed [7]. Conversely, when there is reason to expect irrationality or value divergence between the analyst and society at large, neglect is less likely a sign of hidden intractability or unimportance [2] [7] [8].
ITN can be simplistic while also muddying the water
ITN can be a quick top-level heuristic, but it may also be described as a core part of a cause prioritization analysis [2]. It can be unclear whether ITN rankings or scores should be understood as the outputs of an unpublished but explicitly specified formal quantitative analysis, or an informed but essentially intuitive judgment. It is sometimes used, contrary to advice, as a means for assessing interventions rather than causes [4]. However, the boundary between a problem, solution, and a cause or focus area is not always clear.
The way ITN scores can reduce complex arguments to a simple score, as well as its sheer prevalence, can also encourage a dismissive attitude and a lack of critical thinking [2] [12].
Multiple competing definitions of the three factors exist, and some are misleading
Some people use “room for more funding” or other terms improperly to refer to neglectedness [2] [7]. Intuitive proxy definitions of neglectedness may require an effortful translation to be accurate [3].
Similar intuitive but poor definitions exist for tractability [12].
It is unclear how to assign relative weights to scale, severity, indirect effects, and repurposable resources once a cause is solved in assessing Importance [10] [11]. In evaluating Importance, it may be best to focus on the scale of the limiting or bottlenecked solution factor rather than the sheer scale of the problem [9]. The learning value and indirect effects may matter, but terms like “Scale,” “Scope,” or “Importance” may guide our intuition to ignore them [1] [4]. More broadly, the cluelessness literature deals with the challenge of accurately counting indirect effects of actions into the long term future [11]
The three factors can overlap, leading to double counting.
“Unfortunately, using the common definitions of these words they blur heavily into one another...
Naturally, a cause that’s highly neglected will score well on neglectedness. It can then go on to score very well on importance because its neglectedness means the problem remains serious. Finally, it can also score well on tractability, because the most promising approaches are not yet being taken, because it’s neglected.
Most of the case in favour of the cause then boils down to it being neglected, rather than just one third, as originally intended.” [10]
Citations
[1] A cause can be too neglected
[2] Against neglectedness
[3] Complications in evaluating neglectedness
[4] Evaluation Frameworks (or: When Importance / Neglectedness / Tractability Doesn’t Apply)
[5] List of the most urgent global issues
[6] How to compare different global problems in terms of impact
[7] Is Neglectedness a Strong Predictor of Marginal Impact?
[8] Neglectedness and impact
[9] How scale is often misused as a metric and how to fix it
[10] The Important/Neglected/Tractable framework needs to be applied with care
[11] Why Charities Usually Don’t Differ Astronomically in Expected Cost-Effectiveness
[12] The ITN framework, cost-effectiveness, and cause prioritisation
[13] Cost-effectiveness of research: overview
- Health and happiness research topics—Part 1: Background on QALYs and DALYs by (9 Dec 2020 0:38 UTC; 126 points)
- What posts you are planning on writing? by (24 Jul 2019 5:12 UTC; 60 points)
- EA Updates for October 2019 by (1 Nov 2019 11:47 UTC; 52 points)
- Should you still use the ITN framework? [Red Teaming Contest] by (14 Jul 2022 4:02 UTC; 32 points)
- 's comment on There can be highly neglected solutions to less-neglected problems by (11 Feb 2023 3:50 UTC; 32 points)
- The COILS Framework for Decision Analysis: A Shortened Intro+Pitch by (7 May 2022 19:01 UTC; 16 points)
- [Outdated] Introducing the Stock Issues Framework: The INT Framework’s Cousin and an “Advanced” Cost-Benefit Analysis Framework by (3 Oct 2020 7:18 UTC; 14 points)
- The TUILS/COILS Framework for Improving Pro-Con Analysis by (8 Apr 2021 1:37 UTC; 11 points)
- More to explore on ‘What do you think?’ by (9 Jul 2022 23:00 UTC; 7 points)
- 's comment on Biases in our estimates of Scale, Neglectedness and Solvability? by (6 Mar 2020 7:14 UTC; 3 points)
- 's comment on Marcel D’s Quick takes by (5 Dec 2020 4:10 UTC; 3 points)
- 's comment on Propose and vote on potential EA Wiki entries by (5 Dec 2020 8:42 UTC; 2 points)
- 's comment on Ending The War on Drugs—A New Cause For Effective Altruists? by (14 May 2021 3:42 UTC; 1 point)
Thank you for writing this up—always good to see criticism of key ideas.
I want to contest point 4.
The fact that we can decompose “Good done / extra person or $” into three factors that can be roughly interpreted as Scale, Tractability and Neglectedness is not a problem, but a desirable property.
In ultimate instance, we want to evaluate marginal cost effectiveness ie “Good done / extra person or $”. However this is difficult, so we want to split it up in simpler terms.
The mathematical equation that decomposes the cost serves as a guarantee that by estimating all three factors we will not be leaving anything important behind.
Agreed. However, one of the subcritiques in that point is the divide-by-zero issue that makes issues that have received zero investment “theoretically unsolvable.” This is because a % increase in resources from a starting point of 0 will always yield zero. The critic seems to feel it’s a result of dividing up the issue in this way.
I leave it to the forum to judge!
TIL WIP = Work In Progress.
Upvoted because I like format of collecting sources of critique and splitting them by topic. This post seems conducive to the discussion on the topic.
Thanks for writing this. I want to emphasize a point you make implicitly here, which is that it’s not always clear when ITN is being used as an informal heuristic and when it’s being used for actual or abstract calculation. I think arguments made previously by Rob Wiblin and John Halstead about the conceptual and practical difficulties of this approach make it clear that it is not a suitable method for rigorously ranking causes.
Still, I think it remains a valuable heuristic and a guide for more exhaustive calculations. Though neglectedness may be the wobbliest aspect, it’s a (generally) good approximation of the possibility for additional value when in-depth information on possible marginal returns to a candidate cause area is immediately unavailable.
I also hoped to imply that ITN is more than a heuristic. It also serves a rhetorical purpose.
I worry that its seeming simplicity can belie the complexity of cause prioritization. Calculating an ITN rank or score can be treated as the end, rather than the beginning, of such an effort. The numbers can tug the mind in the direction of arguing with the scores, rather than evaluating the argument used to generate them.
My hope is to encourage people to treat ITN scores just as you say—taking them lightly and setting them aside once they’ve developed a deeper understanding of an issue.
Thanks for reading.
Since this is a WIP, it might be helpful to share a live Google Doc that people can check/comment on/suggest additions to in the future.