effective altruism – Andrew Dornon

Is Effective Altruism Funding Constrained? A Case for Fundraising

Correction: I wasn’t 100% clear on the definition of funding constrained. I stand by my claim that EA orgs should at the margin think about fund raising more.

Is Effective Altruism Funding Constrained?

I posed this question to few folks at EA Global 2018 and got almost exclusively “No’s” in response. Many claimed that charities that work on AI risk are talent constrained— for example they weren’t sure MIRI could hire many more people if their funding 5X’d. I made the counterclaim that surely GiveWell was funding constrained, and was told that they weren’t because of their relationship with Good Ventures and OpenPhil. I’m not sure this is the case.

Good Ventures started with an endowment of $8.3 billion, which is no doubt a lot of money. Annual giving to US charities alone is approximately $400 billion. So even if GV/OpenPhil/GiveWell gave away all of that money tomorrow, EA giving would amount to less than 5% of annual giving by US charities.

Perhaps taking this relative approach isn’t exactly what EAs normally mean when they say funding constrained. But considering that the United Way—the largest charity in the US—gives away $3.5 billion per year, it’s probably time to start considering how to take share from organizations with good intentions but bad execution.

Fundraising likely receives too little attention from EAs because many are more technically oriented and fundraising feels at best fluffy and at worst like sales and marketing. As anyone who has worked in an engineering culture knows, sales is not respectable, at best it’s a necessary evil. Marketing is where buzzwords come from. Given these sorts of aesthetic predispositions, it’s not that surprising EAs have neglected trying to take share from less effective organizations.

But that doesn’t mean it’s the optimal strategy. Especially considering the low hanging fruit of simply copying the most effective fundraising strategies that look a lot like B2B enterprise sales motions. There are strategies that look more like marketing than sales which I may explore separately, but given the cost of sales is predominately time this is a place where EAs with sales aptitude could put in direct work and generate big results.

What Effective Fundraising Would Look Like

Direct Corporate Social Responsibility

Many Fortune 500 companies give away lots of money to causes that are somehow connected to their business and the communities they work in. These large causes have to be approved by very senior executives who are used to making decisions based on businesses cases that show ROI. Winning over CHROs and then the rest of the executive team would by no means be easy. Based on experience I’d guess it would take over a year of conversations.

The initial step would be to have folks with sales aptitude to try to book appointments with junior folks in a few target firms’ Corporate Social Responsibility departments and have big names from EA organizations give presentations.

Indirect CSR

Most F500s have some list of approved charities that they’ll match donations to. The list was created so that there would never be a PR scandal about an organization accidentally matching a donation to a terrorist group or similar. The strategy here would be to first get on those lists, but then to also try to shape criteria going forward for other orgs to get on the list (and perhaps for staying on the list). Those criteria would include effectiveness metrics to benefit EA orgs and steal share.

Government Grants

Getting government grants is a sales process I’m mostly unfamiliar with, but lots of people are and USAID gives away lots of money that EAs aren’t accessing right now.

Why Take Market Share?

EAs may generally be underconfident about their ability to deploy more money effectively, but surely we should be more confident that we can deploy a large share of global charitable giving more effectively than the least effective decile of charitable dollars which are actively harming people and communities. Additionally, creating a fundraising side of EA gives folks with sales and marketing aptitude a role in the community outside of earning to give. On the whole, taking market share from ineffective organizations would at least reduce harm caused, create new opportunities for EA orgs to deliver today, and more optimally utilize the human capital of the EA community.

The Things People Do For Meaning Now

I started this blog because I wanted to investigate what people will do for money and meaning in the near and far future. As I approached it obliquely, it may have gotten a bit muddled, but it remains my main concern. To understand where we are going, I needed to construct a model of where we are. I haven’t yet done that, but I have observations on what people like me and unlike me do for meaning today..

Travel

There are two types of meaning being derived from travel. Business travelers often complain about the demands of their job, and they likely genuinely mean it. But I witness and have succumbed to a small delight in the implied status of the fact that “I, personally, am physically needed somewhere else”. The other type of meaning is the one that certain people get from “seeing the world” or interacting with other cultures. This meaning is harder to describe and prove exists, but it’s become such a generic value that I suggest you try telling people at a social gathering that you “don’t like to travel”. You’ll like be looked at askance and interrogated. And will likely lose status in the average person’s eyes. You will have challenged a common mechanism of generating meaning.

My investigations

I flew more in 2017 than I ever have, including 3 trips to Europe and 26 flights total. I explored a fair number of European cities. It’s hard to estimate how much meaning these travels generated, but based on how frequently I talk about it (even if negatively), I would say some. I’ll now introduce a meaning scale from 1-10 of how much meaning it generated for me. This, of course, is approximate at best and will vary considerably by individual.

Meaning rating: 3

Exercise

Thinking about exercise as meaning producing activity is strange. But it’s hard to argue that Crossfitters, yogis and SoulCycle-ists don’t seem to form a part of their identity around exercise. Those specific activities are easy to pick on because they’re group activities that come with obnoxious life advice—but that’s the point, they’re exercise as meaning creation taken to its logical end—not bastardizations. Marathon training is a solitary activity, but anything you do for that many hours a week must be throwing off a lot of meaning.

My investigations

Over the last several years, I’ve run 3 miles a few times a week fairly consistently, but if I felt like I could get away with stopping, I would. On top of that I’m often invited/dragged to yoga, bootcamp style classes, and SoulCycle. I’ve always liked group exercise, but the group and meaning part always made me cringe.

Meaning rating: 1

Art

Strangely, the most predictable and most ancient “hobbyist” way of creating meaning—the production and consumption of art—seems to be on the wane. Or rather, new arts emerged and have taken the place of old arts. Whatever bizarre number of hours Americans watch tv/surf the internet surely count as arts consumption—and truly this is the golden age of TV and hot takes. I’m just not sure how much identity and meaning is drawn from those things compared to going to a metal show or sculpting things with your hands. It’s possible I’m just a Luddite on this topic—no doubt Youtube video hobbyists and internet bloggers would take issue with it.

My investigations

Over the past year, I’ve listened to more albums intentionally, read more fiction and way more non-fiction than any year prior. My film consumption declined. I went to some really amazing concerts. I’ve shared more book and music recommendations than in the past. Art consumption has always been a big identity driver for me, and I think that it grew this year. Art production has been mixed. I took fewer (zero) art classes this year. Conversely, I started writing, which has been ok.

Meaning rating: 6

Altruism

Doing good things for other people generates a warm glow and a sense of purpose. Donating money to earthquake victims, volunteering at a soup kitchen—these things make you feel meaningful. Unfortunately, generating meaning isn’t the same as generating utility most efficiently. Oftentimes meaning and effectiveness are at odds. People who derive a lot of meaning from altruism have been those who reject effective altruism the most in my experience. Their emotional investments are so high that they can’t accept that their work/time/money has been less meaningful than they feel it is.

My investigations

I volunteered at the Austin homeless shelter by running the computer lab. This mostly consisted of sitting around while the clients did what everyone does on the internet—read email, watched Youtube videos, scrolled through Facebook and occasionally applied for government assistance or jobs. I liked doing this, it generated meaning, it felt good to tell people about doing it. But it wasn’t very effective. I also have gotten very interested in effective altruism and have put some time into how best to implement effective giving into my life. The ideas, conversation and people around EA are very interesting and feel meaningful. Giving away 30% of my income anonymously on the internet…doesn’t feel meaningful, even if it would generate the most utility, which is why I give away dramatically less than that to GiveWell. The meaning-utility gap is likely the biggest problem EAs have to overcome.

Meaning rating: 2

Tribe

Tribe is a catchall for both a community you are in dialogue with (think a church congregation) and an identity group that you consider yourself a part of (the Democratic Party). People clearly generate a lot of meaning from both, but I would argue that communities generate a lot more. This is likely because you are a participant rather than an audience member. I see a lot of shifts from community to identity group that go unnoticed because both generate a tribal feeling that occludes your changing role in the tribe.

My investigations

I struggle with being a member of a tribe, so this was really challenging to investigate. This attitude is probably what makes it easier for everyone to join identity groups—I always vote straight ticket Democrat (unless, this being Texas, the real race is between the GOP and Libertarians), but I doubt I could be in dialogue with the party if I tried. The only big community investigation was my attendance of a Rationalist MegaMeetup in New York. It was essentially a bunch of folks interested in rationality staying in a giant house and talking about stuff. Oh, and Scott Alexander was there. I went with some friends from Austin, so that made it reasonably easy. I really enjoyed meeting a lot of smart, interesting folks doing cool work, but for the most part, I didn’t like the community aspect. I know the whole point is to create an in-group, but that was exactly what I didn’t like.

Meaning rating: 2

Work

It is perhaps unfair to split out work from tribe. At least for people, like me, who work for tribal organizations. What most people think of as meaning being generated from work is the productive use of time to make the world a better place (for some business or person). In companies, this means taking your individual skills and combining them with those of other talented individuals and doing something neither of you could do alone. This productive part of work has not started to generate more meaning than it has in the past. It has just always been a huge meaning machine.

Conversely, as other communities have shifted into identity groups, (and perhaps as art has declined in its generic meaning rating) the relative value of productive work has risen. And organizations have responded to the generic decline in meaning by increasing the community, and thus tribal, aspects of their workplaces. I can’t say if this is good or bad.

My investigations

For the past 4 years, I’ve worked for a strange, Swedish consulting firm that helps Fortune 500 companies do something really hard. We help people change how they think and act at work. On average, I think BTS helps organizations and increases global GDP at the margin. The productive work I do is to help explain this work to clients and shape new offerings. But it’s the community aspects of the company that drive such high meaning.

Meaning rating: 6

Family

Family is difficult because almost everyone has one and almost everyone would say they derive a lot of meaning from it. So I will try to look at what people really do. A large majority of people have children, which is a very concrete action and from which most people say they derive a lot of meaning. Given the time and resource expenditures, they seem to be telling the truth. Conversely, very few people live in multigenerational households caring for aging parents or other relatives. Part of this is increase in healthy lifespan, but part of it must be that they deem it to not be worth it. Perhaps people derive more meaning from family they produce than other family.

My investigations

I did not have a child. But I am quite close to my mother and siblings—we all talk on the phone at least once a week. I didn’t change anything this year that would generate meaning.

Meaning rating: 8 (see the first sentence of family section)

Meaning and Yield Curves

I’ve begun to think about meaning as a sort of capital stock that an individual can grow or shrink via investing time in different meaning generating activities. For instance, my returns to time from art would be higher than my returns to time spent exercising, and thus my stock of meaning would go up.

A potential flaw in this logic is you seem to attribute more meaning to things as the time spent on that thing increases. It’s likely that there is a class of meaning activities that have positive yields that increase as you spend time on them, and others that hit diminishing margin returns (exercise, for me).

There are, of course, also activities that have neutral or even negative yields depending on duration.

An Open Source Plan To Rid The World of Landmines

Note: I did not put 80% of these pieces together, nor am I really an expert. I was exposed to most of the ideas by Rick Dygert at the 2017 East Coast Rationalist MegaMeetup.

First things, first—this is going to sound crazy. When I and about 40 other people first heard this idea being presented, everyone gasped or laughed. It took a while for people to take it seriously. I know that it sounds crazy, but a review of the literature makes it seem…not crazy. Maybe even plausible. Worth trying.

Skip to 2 Facts and 1 Early Result: if you already think landmines are a serious but not cost effective problem to solve because it takes too long.

Some facts about landmines:

“It is estimated that there are 110 million land mines in the ground right now. An equal amount is in stockpiles waiting to be planted or destroyed.
The cost of removing all existing mines would be $50- to $100-billion.
According to the ‘International Campaign to Ban Landmines network’, more than 4,200 people, of whom 42% are children, have been falling victim to landmines and ERWs annually in many of the countries affected by war or in post-conflict situations around the world.”

Source: Minesweepers

Some facts about demining: it’s slow, boring and dangerous work according to the military. And lots of people think it isn’t really worth it (and they’ve done the NPV calculations).

Interestingly, what causes it to be not cost effective is that it’s slow. It takes humans, dogs or machines (or all three) a really long time to find each individual landmine. And yes, they are essentially walking through minefields with metal detectors and probes, wearing Kevlar vests, helmets and visors. Since there are normally lots of landmines and you really can’t miss any before you declare an area “landmine free”, it costs a ton to clear an area even using low wage human labor—a deminer in Afghanistan only makes $145/month. And again, it costs a lot even though labor is cheap because it takes a long time, a study in Mozambique looked at a $40m project to clear 20 sq. miles OVER 10 YEARS.

So this leaves the average person (and certainly effective altruist) in the position of essentially thinking landmine clearance isn’t worth it. And they’re right.

Unless you can reduce the cost of clearance by 10,000 times.

And we can.

(This is where everyone’s jaws drop.)

With bees.

(And this where everyone laughs.)

But to convince you, I only need to you to accept 2 facts and that 1 early research effort is worth pursuing further.

2 Facts and 1 Early Result

Fact 1: Bees accurately communicate where sources of nectar are through something called the Waggle Dance.

From Wikipedia:

“Waggle dance is a term used in beekeeping and ethology for a particular figure-eight dance of the honey bee. By performing this dance, successful foragers can share, with other members of the colony, information about the direction and distance to patches of flowers yielding nectar and pollen, to water sources, or to new nest-site locations.^[1]^[2]…Austrian ethologist and Nobel laureate Karl von Frisch was one of the first who translated the meaning of the waggle dance.^[5]

…

The direction and duration of waggle runs are closely correlated with the direction and distance of the resource being advertised by the dancing bee. The resource can include the location of a food source or a potential nesting site.^[7] For cavity-nesting honey bees, like Apis mellifera or Apis nigrocincta, flowers that are located directly in line with the sun are represented by waggle runs in an upward direction on the vertical combs, and any angle to the right or left of the sun is coded by a corresponding angle to the right or left of the upward direction. The distance between hive and recruitment target is encoded in the duration of the waggle runs.^[1]^[8] The farther the target, the longer the waggle phase. The more excited the bee is about the location, the more rapidly it will waggle, so it will grab the attention of the observing bees, and try to convince them.”

Bees essentially communicate a direction and distance based on the location of the sun and the hive! Cool, interesting, but not unbelievable. As yet unclear how it connects to landmines.

Fact 2: Humans can hijack the waggle dance and train bees to detect things that aren’t nectar using their insanely good sense of smell.

From Wikipedia:

“Sniffer bees or sniffer wasps are insects in the order Hymenoptera that can be trained to perform a variety of tasks to detect substances such as explosive materials or illegal drugs, as well as some human and plant diseases. The sensitivity of the olfactory senses of bees and wasps in particular have been shown to rival the abilities of sniffer dogs, though they can only be trained to detect a single scent each.

Bees and wasps are trained using classical conditioning, being exposed to a particular odour and then rewarded with a sugar solution.^[3] Within five minutes they learn to associate the smell with an impending supply of food and this triggers the proboscis extension reflex (sticking out their tongues).^[4]

Trained hymenopterans have been shown to successfully detect explosive materials including TNT, Semtex, and C-4 as well as gunpowder and propellants.^[3]^[5] . ”

So bees can be trained to identify the scent of explosives. Now we’re getting somewhere. Oh, and good news, here’s a prototype of a machine that can train bees at scale. The company it was designed for went out of business, but it’s plausible their business model didn’t work, not that their product (tubes of bomb sniffing bees for airports) didn’t work.

At this point, we know that we can train bees to find mines and they can tell each other and highly trained humans where they are with a high degree of accuracy. That’s good and all, but watching bees do the waggle dance is labor intensive and requires highly skilled humans. We haven’t solved the problem yet. Other methods tried involve using thermal imaging to follow bees in coordination with other high tech mechanisms.

An Early Research Effort Worth Pursuing: We can use computer vision to decode the waggle dance.

From The MIT Technology Review:

“Enter Tim Landgraf and pals at the Free University of Berlin in Germany. These guys have developed a neural network that can automatically decode honeybee waggle dances. “We have developed a system capable of automatically detecting, decoding, and mapping communication dances in real time,” they say. The new method has the potential revolutionize the study of honeybee foraging.

The waggle dance decoding system is simple in principle. It consists of a video camera that records the movement of bees on a honeycomb. This may look like a seething mass of random motion but there is considerable order here.

…

In tests on bees trained to visit a known food source some 300 meters from a hive, Landgraf and co say their system accurately identified the position of the food source over 90 percent of the time. That’s just as good as human observers. But crucially the machine vison system can work on much larger groups of bees over much longer time scales.”

To clarify what these folks have done: they used a $10 Playstation3 Eye camera to watch 2000 bees do the waggle dance and their machine vision software output this map that corresponded with 90% accuracy to the location of artificial feeders:

The full paper including setup and source code are here.

A Quick and Dirty Plan:

If you bring all these things together, then it looks to me like you have a possible plan to map minefields and demine them with blistering speed compared to the status quo.

Get some test mines.
Have a third party hide them in an area.
Buy the bee training machine.
Build waggle dance decoding rig.
Rent bees from a nearby apiary.
Train bees to find landmines.
Install waggle dance decoding rig in hive.
Set bees loose.
Wait 3 days.
Look at decoding map.
Check for accuracy.
Improve on machine vision algorithm as needed.
When 98% accurate, proceed.

How To Go From Testing To Demining The Whole World

I am offering to fund the first steps of any serious effort with up to $10,000USD and support further fundraising with additional capital and time/effort.
Create a self-sustaining business model: use the above plan, but for truffles.
Create a map of truffles in the Perigord region in early fall.
Show proof that it’s real by finding one.
Sell it to the highest bidding trufflieres.
Fund large scale trials.
If it’s effective, get OpenPhil and GiveWell involved.
If it’s not effective enough, simply self-fund by finding commercial applications.
Demine the whole world.

A Portfolio Theory For Effective Altruism

It concerns me that many in the effective altruism world focus on optimizing for “the best” use of time and resources and often direct people to donate to one non-profit. This strikes me like trying to choose the best investment. Altruism directed toward one cause seems to suffer from the same flaws as investing in a single asset (or even asset class). The flaw essentially being: you might be wrong about the best use of your marginal dollar.

This essay lays out the underpinnings of my portfolio theory of altruism and a rough structure of how I will place bets. I may develop a clearer model for portfolio development, but that is not my intent here.

What Are The Right Causes

This question can feel almost settled by places like GiveWell and Giving What We Can. However, there’s a lot variation in the effective altruism community in terms of what people actually work on and donate to. There are those who care deeply about the suffering of wild animals or those who worry that suffering is a fundamental aspect of the physical universe and so want to destroy it (I don’t fully understand this idea and could be misconstruing it). Largely this variation is created by:

The moral status of non-humans
Moral uncertainty
How much to value the future (human or otherwise)
Nonlinearity

What Is The Moral Status of Non-humans?

Animals likely have non-zero moral status. This is the overwhelming consensus of humans since we tend not to inflict pain on animals directly, and most will go out of their way to avoid directly killing them inadvertently. The fight here is over how much moral status.

If you were to give them 10% as much moral status as a human, it strikes me as quite strange that you would be willing to support the murder of an animal for food. I am personally a vegetarian and fair-weather vegan, and 1/10^th of a human is probably the amount of moral status I ascribe to animals. (Others seem to assign less status, but at what point does something go from non-murderable to murderable?)

In my model, 10% of moral status indicates that I would have to have run out of almost all opportunities to improve human utility before I began to shift resources toward animals. Vegetarianism will be the exception because the negligible effort is non-transitive and meat-eating is itself actually detrimental to human welfare. I won’t put any transferable effort or resources into animal causes—unless there are clear benefits to humans as a by-product.

The Role of Moral Uncertainty

Moral uncertainty asks us to make decisions on issues where we are uncertain. Pascal’s Wager is the classic example of moral uncertainty. Say you thought there was a 1% chance of Pascal’s God existing, but believing led to infinite utility. The expected value of believing is infinite.

Moral uncertainty plays a more interesting role in finite situations. For instance, say you think there is a .001% chance that human fetuses have human-equivalent moral status. This is a very small chance, but the expected harm of 664,000 abortions per year is 6.64 human deaths. In my personal model, the utility derived from those abortions far outweighs that expected disutility. Uncertainty in the correctness of your moral stances is rarely factored into decisions where you feel pretty sure, but given how frequently the moral consensus shifts over time, ever being 100% certain of something feels like a mistake.

In my model, I’ll attempt to avoid supporting or opposing issues where I feel morally uncertain (beyond the standard passage of time could prove me wrong uncertainty) on the issue itself.

What Is The Right Social Discount Rate

The social discount rate can be thought of akin to a financial discount rate. The lower the discount rate, the more we should value the world and people in the future. The higher the discount rate, the more we should care about people alive today.

I tend to side with Parfit and Tyler Cowen, that the appropriate discount rate is close to zero, if not actually zero. I’m not convinced that people alive today have much more moral status than people who could potentially exist. But to presuppose the argument that we’re valuing non-existent (and potentially never existing) people, I’ll frame my time horizon as close to infinite in order to create the best circumstances for human flourishing.

Of course, a discount rate of near zero does not imply a clear course of action.

Oh, and the Universe is Literally Unpredictable Because It’s A Nonlinear System

Not only do our moral choices have to contend with how to value the far future, they have to at least acknowledge that chaos theory means small changes in initial conditions can cause large changes in future states.

A way to potentially combat this is to create utility in the present, so that at least the future starts from a higher base. Another is to make choices that don’t require you to be right about what happens in the future.

Where I’ll Place My Bets

In my model, I will face up to the discount rate and nonlinearity by trying to clearly maximize utility in the initial conditions (today) and as the time horizon extends make increasingly diverse bets with low probabilities, but given population and economic growth, potentially very large payoffs. Because of how these bets work, I can direct 70% of my resources to clear benefits for people and the world today, 20% toward organizations working to reduce the risks of reasonably foreseeable existential threats, and 10% toward speculative bets for the far future and expect similar utility payoffs.

A portfolio could be:

Today

Carbon offsets 2X my current consumption: 150k kT
GiveWell to do with as they see fit

100 Years

1000 Years