Cause Exploration Prize: Distribution of Information Among Humans

Tl;dr: the cost of redistributing human knowledge & information has a lower bound of about 0. Information is often distributed suboptimally, both in the EA community and in the world in general, and deliberate adjustments to how information is distributed can often be a positive-sum interaction. I believe it is worth building detailed models of flow and distribution of information in order to identify ways to improve it. I’m surprised by how difficult it is to find sources about this topic. One could consider “improving distribution of information” a cause area, even though the for-profit sector covers a lot of the area in principle, but the assessment of that cause area comes with a large degree of uncertainty.

Extended Summary

People know certain things, and don’t know other things. Knowing about things can be very valuable, both indirectly (e.g. something actionable such as ‘software X is very good for Y’) and directly (e.g. receiving a compliment that makes you feel good). It’s related to the concept of arbitrage: moving a piece of information from A to B is sometimes strictly better for the people involved, often even in a positive-sum way.

A lot of our digital landscape is centered around moving valuable pieces of information to people who could benefit from them - this includes search functions, databases, rankings, recommender systems, newsletters, dating apps and more. Indeed the for-profit sector is quite involved in the “distribution of information” business. My impression is that nonetheless there’s a lot of room for improvement, both within the EA community and in the world in general. Some of the key points in this explorative post are:

• The cost of moving information has a lower bound of about 0, which makes interventions in the area potentially highly cost-effective
• Having a detailed mental model of the nature, flow and distribution of information is most likely helpful in coming up with promising interventions
• It seems to me that it is fairly uncommon for people to have such a mental model
• While the for-profit sector is very involved in this space, it may be insufficient, and may be systematically neglecting aspects of the area that are e.g. not easily monetizable
• It’s risky to leave information distribution to “unaligned” companies
• AI advances can be instrumental in improving information distribution

Introductory Examples

A good way to motivate what “distribution of information” is referring to in the context of this post is to look at a few made-up-yet-realistic examples, with varying degrees of significance:

• Alice loses her lifetime savings to a scammer, because under pressure she follows the person’s orders, not knowing that she is being scammed
• Bob suffers from from chronic back pain and is unaware of a promising treatment option
• Celine and Daniel are romantically interested in each other, but keep it to themselves due to low expectations of success and/or potentially high social cost of opening up
• Elizabeth misses the concert of her favorite band happening in a city nearby, because she didn’t know the concert was happening
• Fátima spends parts of each summer stressed out by wasps, because it’s unclear to her which of the many commonly cited ways of repelling wasps actually work
• Gerald procrastinates starting going to the gym for years, because the fear of not knowing what to do and looking incompetent scares him, and he doesn’t know that many gyms will offer a very easy onboarding process
• Thousands of people in Germany were surprised by historical floods in summer 2021 that left more than a hundred dead, although it was theoretically known days in advance that excessive rainfalls would cause floodings
• A Kenyan farmer is experiencing suboptimal yields because he’s unaware of a few effective and affordable improvements he could make
• A rural hospital is experiencing unnecessarily high mortality of patients due to the staff’s unawareness (or lack of conviction) of basic hygiene measures
• Many people suffer from STDs or unwanted pregnancies due to lack of awareness of the risks and/or protective measures
• A junior developer accidentally corrupts the production database of their company due to lack of knowledge about the consequences of their actions
• A senior developer spends hours trying to understand a piece of code, because the colleague who had originally written it has left the company and didn’t do a good job with the documentation

In all these situations people are missing some small piece of information that would improve their lives quite a bit if they just had access to it. Here it’s not latent knowledge nor complex mental models, but very simple facts that would in principle be easy to convey, and which would make a great difference to the people involved.

The Problem

Human knowledge is distributed in a certain way at any point in time. It affects both individual well being and economic productivity. A lot of marginal improvements in information distribution appear to be positive-sum transactions, hence the state of the world is substantially worse than that of a hypothetical world with “optimally” distributed information would be.

A lot of sub-problems seem to be involved, such as:

• Predicting which piece of information would be valuable to any given person, or group of people, is hard
• Sometimes people are not even aware what information to look for, because they’re ignorant of something being a problem to begin with
• Sometimes people or organizations are incentivized to keep access to valuable information scarce, in order to capture more of the value for themselves, which has negative externalities on the public
• Sometimes people are not asking for something or telling others something due to some accompanying status risk, or other (social) cost
• Some potentially valuable information is just very hard to bring to existence; we can only distribute what’s already there, but uncovering new information is sometimes much more difficult

In the following I’ll very roughly get into the scale, neglectedness and solvability of the issue.

Scale

On some level, every single human and every organization is affected by suboptimal distribution of information, some obviously more so than others. It appears almost self-evident to me that a hypothetical world with “optimal” information distribution^[1] would look vastly different (and in many ways better) than our current one:

• People would waste a lot less resources on things that don’t work
• General friction in organizations would be substantially reduced
• People would make fewer costly mistakes (both in work and in life)
• Romantically frustrated people would likely have a better shot at finding partners (although this one may be less obviously true, since distribution of romantic partners is closer to zero-sum than other points mentioned)
• A lot of problems people are experiencing in their lives would be easily solved or improved
• People would feel better due to giving and receiving more compliments
• People would have a much better shot at mitigating some personal risks

Access to relevant information is of course far from the only important bottleneck, and there are naturally many other reasons for suboptimal quality of life or inefficient organizations, from material constraints to bad incentives to limits in rationality. Still, the way information is distributed seems to be at least one important bottleneck, and one that if it was “solved” would have a major positive impact on the world.

Ultimately the key uncertainty with regards to how promising this area is seems to lie less the scale, which is quite large, than in the neglectedness and solvability.

Still, to provide at least one (as “back of the envelope” as it gets) estimate: if we assume that

• The subjective life experience of a life under optimal information distribution would be on average 10% better than the current default
• A typical human experiences ~0.5 QALYs per year
• An organization with processes allowing for optimal information distribution would be on average 10% more productive than the current default

then we obtain a total improvement of about

• 400 million QALYs per year, which is in the order of ten million lives saved (which is on a similar level as ending world hunger) - note that this would even be significant underestimate, because it takes into account the improvement of quality of life, but not to what degree access to information would extend lives and counterfactually prevent deaths
• An increase in GWP of over $8 trillion

Additionally, better distribution of information would also be instrumental in solving many other problems, which is harder to quantify. For instance, in a world of optimal information distribution:

• People with an inclination to be interested in effective giving or effective altruism in general would be exposed to the idea
• People interested in e.g. AI Safety would have an easier time identifying what others are working on, what research questions are open, and where they might provide value themselves
• The best arguments for and against important controversial questions would be easily available to everybody and allow people to build a much more robust and accurate model of the world

Neglectedness

There are two sides to the neglectedness of the area of improving the distribution of information:

• On the one hand, it’s explicitly talked about very little; from my own limited experience most people (mainly EAs) are surprised when I bring up the topic, and have a fairly limited mental model of information flows and ways to improve them in robustly positive ways
• On the other hand, there’s a lot of companies in the for-profit world working on products that one some level redistribute information in valuable ways

To name just a few examples for the second point:

• Google (with its search engine) helps people efficiently find answers to many relevant questions
• Stack overflow and quora facilitate curation of frequently asked questions and identifying the most useful answers
• Dating-sites make information about available singles available, often provide information on the expected compatibility of two people, and (to a degree) make people aware of others who are interested in them
• Many recommender systems, while not necessarily aligned with the customers’ values, aim at delivering highly relevant information to people
• Reddit allows users to share and rate links as well as original posts, and to access them based on overall community ratings, which lets content get more exposure the more broadly popular it is
• Photofeeler enables people to get honest feedback about pictures of themselves

A lot of organizations in the nonprofit world can also be considered as working on the problem of improving distribution of information:

• Givewell makes information about the most effective nonprofits in global health and development easily accessible
• Development Media International saves lives by encouraging positive behavior change through media campaigns
• Wikipedia makes factual and conceptual information freely available to the world
• The 80,000 Hours job board is a central place for high impact career opportunities

Certainly parts of the problem are already being tackled by a variety of organizations. An open question is if there are subsets of the area that are systematically underexplored by the existing solutions. Possible reasons for such neglect could be

• Misaligned values of for-profit solutions - e.g. recommender systems as well as dating sites are often providing a lot less value to people than they could, as they’re optimized much more for benefiting the organization behind it than the users)
• Lack of promising monetization options - e.g. DMI’s work is very unlikely to be covered by the for-profit sector and thus more likely to be neglected
• Suboptimal implementation - e.g. a few attempts at solving a particular sub-problem have been made, but have not yet succeeded in being widely adopted

Ultimately neglectedness is hard to estimate for the area as a whole. If this topic is researched further, one of the most promising paths seems to be to identify particular sub-problems of information distribution that are important yet also systematically neglected (and solvable).

Solvability

The previous section contained many examples of organizations and products that already effectively solve sub-problems of information distribution. This can be considered good evidence that parts of the area are highly tractable. Technology, particularly powered by the internet, but also recent (as well as expected future-) AI advances, make many solutions possible and scalable that weren’t possible just a few years ago.

The dawn of new types of solutions in this space that weren’t possible before also adds an element of urgency for value-aligned organizations^[2] to get involved, in order to prevent first-mover and second-mover advantages to be taken fully by for-profit organizations, which may be less well aligned with the values and preferences of the public.

Similar to the notes on neglectedness earlier, solvability also depends a lot on which sub-problem is being considered. For instance a few sub-problems seem inherently difficult:

• Lack of trust can add a lot of friction to information flows
• Turning actionable information into actualized value may require a lot of individual discipline

At the same time other sub-problems and interventions seem quite solvable, especially in the face of technological advances:

• Improving predictions about which information is valuable to any given person
• Requiring less human attention in the identification of relevant information, e.g. by having personal AI assistant scrape the web

One aspect that adds to the suspected good solvability of some of these sub-problems is that information processing is inherently cheap and scalable, with a lower bound of marginal costs of practically 0. In practice of course, much of the valuable information that exists is actionable information, which hence requires the beneficiary to bear some cost (such as time, money or willpower) in order to obtain the value. Still, the fact that in many cases a central entity may be able to facilitate the sharing of valuable information practically free of any cost, makes it a very promising area for technological solutions.

It’s possible that there are sub-problems that are not very solvable on a level that scales globally, but still for particular communities, which might still be a worthwhile endeavor for nonprofits to investigate. For example, high-trust environments (such as the EA community) can allow people to share exploitable information that enables more effective cooperation. A somewhat trivial way in which we’ve used that in our local group in the past is that we asked people to honestly share how excited they would be about attending a particular valuable event with limited capacity, in order to help us decide who to invite.

In conclusion, “distribution of information” may be too wide a field to conclusively judge with regards to scale, neglectedness and tractability, but it seems promising to invest some effort into building better models of the whole area and identifying more tangible sub-problems.

Some Hints at Possible Interventions

This post is of course primarily meant to hint at the problem itself and has no ambitions of solving (parts of) it to any significant degree. But hinting at a few potentially promising interventions probably doesn’t hurt. Some of these things are already practiced by quite a few people, particularly within the EA & rationality movement, but may still be underutilized.

Interpersonal 

• Create spaces for sharing honest feedback (e.g. via anonymous forms)
• Deliberately try to identify “information arbitrage” opportunities in conversations with people (e.g. be transparent about your practical problems and other struggles)
• Be transparent about your general interests and projects, to allow others to identify and share valuable opportunities with you
• Give sincere compliments (while they’re often not very actionable, they can have tremendous direct value for the recipient)
• Spend in-person time with others and be exposed to each other’s everyday struggles, habits and life hacks (e.g. as a programmer it can be really useful to observe other programmers using their IDE / dev tools)
• Whenever encountering something particularly useful to you, take a moment to ask yourself who else could benefit from this

Groups & Public

• Run sessions focused on finding and sharing the most valuable information that exists in this particular group of people
• Make relevant information easy to find / navigate; e.g. for a local group, consider having a single google doc collecting all the relevant links for that group, that people can always fall back on to find anything else; and/or cross-link a lot between your central documents
• When writing posts, provide a title & short summary that allows people to very quickly make a good prediction about whether the full post will be valuable to them
• Optimize (semi-)publicly accessible materials for being search-friendly

Scalable

• Personal AI assistants that has knowledge about their “owner’s” interests and is able to find and summarize the most relevant sources from the web on an ongoing basis
• Capable and “value-aligned” recommender systems in practically every area
• Building a networking app with an actually useful search feature (looking at you Swapcard)
• When building any kind of reasonably complex software, give users the opportunity to stumble upon non-obvious useful features, e.g. by providing context-sensitive hints, non-intrusive “did you know?” messages, or at least have them prominently placed, with useful tooltips

Meta Level

There are a few more meta projects in this space that I would be quite excited about:

• A think tank diving deep into how to improve distribution of information (either for the world as a whole, or e.g. for the EA community in particular)
• An interested author writing a pop-science book on the topic, with the most valuable ideas around it and the most useful mental models; there are books like How to Measure Anything, Thinking Fast & Slow, Why We Sleep or Rationality that make a wide and complex topic available and applicable to the reader, and it seems to me that “how humans exchange value via knowledge & information” would be a topic that deserves a similar book (if you happen to know of an existing one that qualifies, please let me know)
• For the EA community in particular, it would be great to have some high-quality resources explicitly on how to improve the flow and distribution of information in (local) groups or sub-communities

Why Information Distribution May not be a Promising Cause Area After All

While I’m obviously of the opinion that this whole topic is worthy of more research, there are possible reasons to assume it’s less promising than I think. I’ve tried to compile the most likely reasons in this section.

1. The information market is sufficiently efficient

One key assumption is that the for-profit sector doesn’t solve the whole problem, and that there are sub-areas of the field of information distribution that are systematically neglected. Maybe this is wrong, and the for-profit world has us covered to a greater extent than I think, and more or less every information distribution problem will over time get solved by market forces.

2. A cause area that contains everything is of little use

Similar to how a theory that can explain anything and thus doesn’t make any falsifiable predictions is mostly useless, maybe a similar point could be made about what I’m calling information distribution: it’s such a wide area that almost any problem can be framed as a problem of how information is distributed. If this is the case, then “improving information distribution” may not really be its own cause area in the first place, but just a different name for a variety of problems that are worked on already under different names.

3. Finding good interventions is very hard

Maybe the reason that many unsolved sub-problems in this space exist is that they’re just very hard to solve. I somewhat assume that people generally either haven’t looked hard enough for solutions, or that solutions were until now impractical due to limited technical capabilities. I think the latter point is still true, and advances in AI will open many doors in this domain. But I may indeed overestimate the degree to which neglect is a relevant cause of suboptimal states in information distribution.

4. Information is distributed close to optimally already

Admittedly I don’t really believe this to be very likely. But I may still overestimate just how suboptimal the current state of information distribution in the world really is, and how much better it could be.

Uncertainties and Open Questions

In this post I’ve made the point that the distribution of knowledge and information among humans is by default suboptimal, and that there may be opportunities to systematically improve it. I’ve claimed that the scale of the problem is large, and that it additionally affects the work done in other cause areas through its effects on efficiency in organizations and research. Still my confidence in the claim that more resources should be spent on improving the distribution of information is limited. This is mainly due to the following remaining uncertainties:

1. It’s a very wide area that contains a lot of very different sub-problems; this lack of specificity may make the concept of “information distribution” less useful than other, more focused cause areas
2. Sub-problems may differ greatly in how neglected and solvable they are, and digging deeper into them will be necessary to get a better grasp on how promising the area is
3. To what degree is the for-profit sector satisfying needs in information distribution? Does it systematically neglect any valuable parts of the issue?
4. As you may have noticed, this post has barely any references, often appeals to intuition, and at the same time makes relatively few really concrete claims or predictions; a follow-up post with a bit more “meat” and surface for attack would probably make sense, and it would be great to find some existing research that is relevant to the subject.
5. I found it difficult so far to find existing resources on the topic and am unsure if it just hasn’t really been covered from the “cause area” and “public goods” angle, or if it has and I just failed to find it
    

In conclusion, a lot of questions and uncertainties remain, and this post has barely scratched the surface of a potential cause area that is still rather vague around the edges. Yet I hope the discussion up to this point is sufficient to spark some further exploration of an interesting topic, and may ideally even lead to some promising projects or interventions in the future.

1. ^{^}

   Whatever “optimal” information distribution may mean in this context - it could e.g. be defined as any pareto optimal state where no further positive-sum information trades are possible; or a state that maximizes some metric such as average well-being of humans currently alive.

2. ^{^}

   “Value-aligned organizations” in the sense of organizations that care about the values and preferences of the public and/or their users, rather than primarily about capturing as much of the value as possible for themselves.