Lesson 2: The ITN framework
The ITN framework evaluates cause areas along three dimensions: importance, tractability, and neglectedness. The idea is simple: work on problems that are big, solvable, and under-resourced. But each dimension is subtler than it first appears.
Importance (scale)
How much suffering, death, or lost potential does this problem cause?
This is the most intuitive dimension. A disease that kills millions is more important than one that kills hundreds—all else equal. But “all else equal” conceals several hard questions:
Who counts? If you only count humans alive today, global health dominates. If you count animals, factory farming involves orders of magnitude more suffering. If you count future people, existential risk dwarfs everything.
How do you measure? Deaths? DALYs? Subjective wellbeing? Economic output? Each metric gives different answers. Malaria kills roughly 600,000 people per year—mostly children. Factory farming confines roughly 80 billion land animals per year. Climate change could displace hundreds of millions. An existential catastrophe could, by definition, eliminate all future value. These numbers aren’t in the same units.
Over what timescale? A problem might cause moderate harm now but catastrophic harm in 30 years. Do you discount the future? By how much? Your answer to this question alone can flip your entire cause ranking.
How likely is it? Some problems are certain—malaria is killing children right now. Others are probabilistic—an engineered pandemic could kill billions, but might never happen. Raw scale is misleading without adjusting for probability. Malaria kills 600,000 people per year with near-certainty. A catastrophic pandemic might kill 100 million, but with perhaps a 5% chance per century. The expected harm (scale × probability) is what matters for comparison, and it can look very different from the headline number. This is where Week 3’s expected value reasoning becomes essential.
Importance is sometimes split into scale (how many are affected) and severity (how bad it is per individual affected). A problem that mildly inconveniences a billion people is different from one that kills a million—even if some aggregate metric makes them “equivalent.”
Tractability
If we threw more resources at this problem, how much progress would we actually make?
A problem can be enormously important but essentially intractable—at least right now, with current tools and knowledge. Tractability asks: is there a plausible path to making progress?
This is where the framework gets genuinely difficult. Tractability depends on:
The state of knowledge. Do we know what works? For malaria, yes—insecticide-treated nets reduce child mortality, full stop. For AI alignment, we’re still trying to define the problem precisely. High certainty about solutions increases tractability.
The nature of the problem. Some problems have clear technical solutions (distribute more nets). Others require coordinating millions of actors (climate change) or solving deep conceptual puzzles (AI alignment). Coordination problems and conceptual problems are generally harder than distribution problems.
Diminishing returns. Tractability isn’t fixed—it changes as more resources flow in. The first million dollars for a neglected problem might be transformative. The billionth dollar for a well-funded one might accomplish very little. This is where tractability and neglectedness interact.
Political and social feasibility. An intervention might work in principle but face insurmountable political barriers. Ending factory farming by banning it outright would be enormously impactful—and is politically unthinkable in most countries right now. Tractability has to account for the world as it is, not as we’d like it to be.
A useful way to think about tractability: imagine a doubling of resources dedicated to this problem. What percentage reduction in the problem would you expect? 1%? 10%? 50%? Your answer is an estimate of tractability.
Neglectedness
How many resources are already going toward this problem?
This dimension captures a basic economic insight: the first euro spent on a problem typically does much more good than the billionth. If a problem already receives billions in funding, the marginal impact of your contribution is lower—not because the problem is less important, but because the low-hanging fruit has already been picked.
Neglectedness is measured relative to the scale of the problem. Global health receives billions in funding annually—but affects billions of people, so the per-person spending is still quite low. AI safety receives tens of millions—but the number of researchers is tiny relative to the difficulty of the problem. Factory farming receives almost nothing—a few hundred million dollars globally against 80 billion animals per year.
Some useful comparisons:
Global health (beyond government): ~$40 billion annually against ~1.2 billion people in extreme poverty
Climate change mitigation: ~$600 billion annually; affects the entire planet
Factory farming (welfare): ~$200 million annually against ~80 billion land animals per year
AI safety research: ~$300 million annually and growing fast; potentially existential
Pandemic preparedness (beyond government): ~$1 billion annually; potentially existential
These are rough figures—meant to illustrate orders of magnitude, not to be cited precisely.
The neglectedness dimension explains why EA community members often end up working on problems that most people haven’t heard of. Popular, emotionally compelling problems attract funding. Neglected ones don’t. But if you’re trying to maximise your marginal impact, the neglected problems are often where you can do the most.
Putting ITN together
The framework’s logic is multiplicative. A cause area scores high when it’s important and tractable and neglected. Being strong on two dimensions but weak on one can be enough to make it a poor priority.
Climate change is enormously important and reasonably tractable—but it’s not neglected. Hundreds of billions are already flowing toward it. Your marginal contribution, relative to what’s already being done, is small.
Factory farming is enormously important (if you count animals), modestly tractable, and extremely neglected. Almost no one is working on it. Even modest contributions could be transformative.
AI safety might be enormously important (if the risk is real), has uncertain tractability (the problem might be solvable or might not), and is only moderately neglected (funding is growing rapidly). The case rests heavily on your beliefs about the probability and severity of the risk.
This isn’t a formula that spits out a number. It’s a structured way to think—and to identify where your view hinges on contested assumptions.