Long Term World Improvement

2017 · ~4,726 words

The LTWI founding manifesto (May 2017)

There's a lot to like about Effective Altruism, but ultimately, I fundamentally disagree with a core assumption EA makes. I'd like to explain why, what my plan is to improve on EA, and why I think it'll be more effective at shaping the world than EA.

Effective altruism, like most groups, is based on synergy: a whole greater than the sum of its parts. You form a group because you can do more with teamwork than by yourself. You start a company because if you pool everyone's money, you can make more products. You put the aluminium smelter next to the dam, because one feeds into the other. You win by taking advantage of positive externalities, division of labor, and splitting large fixed costs. However, no matter how much business schools use the word, not everything has synergy. You don't put a pizza restaurant and the aircraft warehouse in the same building. You wouldn't put physics and dance students in the same classroom. There's no point. They'd just get in each others' way.

A central belief of EA is that there's synergy from all different ways of doing good. The EA message is that people donating to fund insecticide nets, campaigning for veganism, and doing research on unfriendly AI can all benefit from each other. They can all be more effective if they share websites, go to the same events, plan strategy together, fundraise together, cooperate in recruiting new members, and so on. And to some extent, I agree with that. Even projects that sound very different can, sometimes, benefit from teaming up.

However, I think there's very little synergy between doing good on a short timescale (say, one year or less), and a long timescale (five years or more). Doing good in the long term means you must consider how systems evolve over time, and that's a very different mode of thinking. If a charity buys cancer drugs for patients, they need to know how cancer treatment works. But if a pharma company develops a new cancer drug, they need to analyze not just how things work now, but all the ways they might work, five or ten or fifteen years later. A competitor could make a better drug before they release theirs. The FDA might change its rules for drug approval. The economy could crash, and make banks pull their funding. Insurers could get together, and force companies to lower prices. Courts could change how they enforce patents, the AMA might recommend more or less aggressive treatment, the tools scientists use for drug discovery could be replaced, it goes on and on and on. If anything big happens, the charity's donors can just give elsewhere; the pharma company's investors are mostly stuck.

And that's a tame example. When there's even more change, you have to make plans with no obvious connection between cause and effect. For example, suppose someone said they wanted to replace Boeing in the airplane industry, and their strategy was to sell Boeing fuel pumps. At first glance, that makes little sense. But it's essentially what Microsoft did with computers. Sell an apparently minor, but critical, component to the industry's leader, and expand on that component until the original leader is irrelevant. This type of indirect maneuvering has almost no connection to, say, doing a controlled study of vegan diets or polio vaccines; the time horizons, and the uniqueness of each situation, make statistical research impossible.

If short-term and long-term plans have little synergy, that implies each group ought to pick one, and focus on that. I'm picking long term. Partly because it plays to the strengths of me and my friends, who are mostly younger and more intellectual. And partly because I have long time horizons; I don't put a huge discount on the world in ten or twenty years, compared to the world now. This preference seems to be pretty rare, and supply and demand mean it's easier to achieve rare goals than common ones. Since many EAs share these traits, I think a lot of them would benefit from picking long-term along with me.

Right now, as I'm writing this, there are also two extra advantages to long-term. The first is that it's been in a slump recently, so a small number of people can have a bigger impact. (Eg., to my knowledge, the only serious attempt to forecast AI trajectories in the last three years was by AI Impacts, and it's unfortunately still unpublished.) The second is that, compared to EA, it makes developing new ideas much easier. The defining EA question for any new plan is, "does it do the most good?". But that's very hard to answer. Partly because "the most good" really depends on who you are, and what you value. And partly because, to say something is "the most good", you also have to know how much good everything else is, which is a lot of work. It's not a coincidence that the three biggest EA causes - global poverty, factory farming, and AI risk - are all ideas that came from elsewhere (and were brought into EA afterwards), not invented from a first principle of "do the most good". Of course, effectiveness and benefit-per-dollar are still important. But if the first question asked is "how long does this plan take?", it's much easier to explore the space.

As a first step, I'm creating a new private Facebook group, called "Long Term World Improvement". Like the plans it will discuss, it's an experiment; it may fizzle, or transform into something very different. Topics might include, but are not limited to:

While these projects are very different, they all take place in a single, shared future. And their success all depends on understanding that future. This links them all together, in a way that's impossible for "doing good" as a pure abstraction.

Off-topic items include, but aren't limited to:

Hacker News moderation rules will apply. (For non-Hacker News users, check out https://news.ycombinator.com/newsguidelines.html; you can also create an account, turn on "showdead", and browse https://news.ycombinator.com/threads?id=dang to get a feel for how it works.) To join, message me on Facebook, email me at alyssamvance@gmail.com, or chat me on Signal ([phone number redacted]). If you don't know me, please include a short summary of who you are, and what you're interested in. As moderator, I reserve the right to not add people, or to remove existing people if they break the rules. Feel free to message me with questions. Good luck, and may tomorrow be brighter than today.

What went wrong at the World Transhumanist Association (2017)

I was executive director of the World Transhumanist Association, now Humanity+, back in 2011. Although Humanity+ technically still exists, it has few active members, and does very little. I haven't discussed what happened in a while, so I thought I'd write a post-mortem to help inform other futurist projects. In my view, here's what went wrong:

  1. The organization was run by a Board of Directors, who were elected by the membership. Each director was elected individually. The membership was not a well-defined group, but varied widely from year to year, since anyone could join by paying a small fee. Therefore, the organization had very little unity and coherence of purpose. The directors didn't have to agree on anything, and were constantly fighting. The extremely broad mission statement made this problem worse - at one point, several directors fought to make the org's main mission repealing the American embargo of Cuba (seriously, that wasn't a joke). Some directors essentially did nothing, while others continuously caused trouble, and "firing" a bad director was almost impossible.
  2. There was almost no connection between Humanity+, and the people building the technologies we ostensibly focused on. The original goal of transhumanism was the enhancement of human bodies and human minds through technology. But that requires knowledge of certain fields, notably biology, which basically none of us had (including me). Even information-wise, most of us heard very little about advances in biology as they were announced. I made some attempt to fix this, by connecting with the DIY Bio crowd, but I got very little support and the effort flopped.
  3. The organization was very inwardly-focused. The most common topics of conversation were things like who should be in charge of what, who was rude to whom and needed to apologize, what the website design should be, who was running in the next election, who was a dirty rapscallion who shouldn't be trusted, what chapter should get official status, how the student guide should be worded, and so forth. Not much attention was given to our real goals, whatever those were (see point #1).
  4. Except for people we already knew well, there was little real engagement between us and our members. I religiously kept records of how many people joined each quarter, and what percentage of readers opened each monthly newsletter, but the lists of names might as well have been warehouse inventories. At one point, I did try to connect everyone with other transhumanists near them, but us and the members remained two distinct groups rather than a unified whole.

The Singularity Summit, and the 2011 program (2017)

The Singularity Summit was an annual event from 2006 through 2012; I helped run parts of it during its last four years. At the end of 2012, it was bought by Singularity University as part of their branding deal with MIRI (at that time, called the Singularity Institute). Nothing ever really replaced it, which I find very sad. Singularity University has some conferences now, but their main goal is to wring money out of Fortune 500s by selling tickets to VPs of Bus Dev, so they're both extremely expensive and full of MBA-speak.

Here's the program for Singularity Summit 2011, which was held in Manhattan at the 92nd Street Y:

Day 1

7:00 AM Registration and Catered Breakfast 8:00 AM Nathan Labenz: Welcome 8:30 AM Ray Kurzweil: "From Eliza to Watson to Passing the Turing Test" 9:30 AM Stephen Badylak: "Regenerative Medicine: Possibilities and Potential" 10:00 AM Sonia Arrison: "100 Plus: How the Coming Age of Longevity Will Change Everything, From Careers and Relationships to Family and Faith" 10:30 AM Coffee Break 10:45 AM Peter Thiel: "Back to the Future" 11:30 AM James McLurkin: "The Future of Robotics is Swarms: Why a Thousand Robots are Better Than One" 12:00 PM Michael Shermer: "Social Singularity: Transitioning from Civilization 1.0 to 2.0" 12:45 PM Catered Lunch by Chef Kwame Onwuachi 2:15 PM Jason Silva: "'The Undivided Mind' — Science and Imagination" 2:45 PM Stephen Wolfram: "Computation and the Future of Mankind" 3:30 PM Dmitry Itskov: "Project 'Immortality 2045' -- Russian Experience" 4:00 PM Christof Koch: "The Neurobiology and Mathematics of Consciousness" 5:00 PM Coffee Break 5:45 PM Eliezer Yudkowsky: "Open Problems in Friendly Artificial Intelligence" 6:15 PM Max Tegmark: "The Future of Life: a Cosmic Perspective" 7:00 PM Closing

Day 2

7:00 AM Catered Breakfast 8:00 AM Alexander Wissner-Gross: "Planetary-Scale Intelligence" 8:30 AM Sharon Bertsch McGrayne: "A History of Bayes' Theorem" 9:00 AM David Brin: "So you want to make gods. Now why would that bother anybody?" 9:30 AM Coffee Break 10:00 AM Tyler Cowen: "The Great Stagnation" 10:45 AM Tyler Cowen & Michael Vassar Debate The Great Stagnation 11:15 AM John Mauldin: "The Endgame Meets The Millennium Wave — Why the Economic Crisis will be History as We Create the Future" 11:45 AM Riley Crane: "Rethinking Communication" 12:30 PM Catered Lunch by Chef Kwame Onwuachi 2:00 PM Dileep George and Scott Brown: "From Planes to Brains: Building AI the Wright Way" 2:30 PM Jaan Tallinn: "Balancing the Trichotomy: Individual vs. Society vs. Universe" 3:00 PM David Ferrucci: "Watson AI Perceptions" 3:30 PM Dan Cerutti: "Commercializing Watson" 4:00 PM Ken Jennings: "The Human Brain in Jeopardy: Computers That 'Think'" 5:00 PM Closing

A retail store for cryonics (2017)

Proposal: Open a retail store for cryonics, somewhere in San Francisco or Berkeley. In the short term, the number of people this helps is relatively small. But in the long term, cryonics could be brought way down in cost through economies of scale, which would in turn make it more attractive, further lowering costs, and so on in a "virtuous spiral". There are four major obstacles to scaling cryonics:

  1. Few people know about cryonics, and of people that do, the information they have is inaccurate
  2. Almost everyone thinks cryonics costs way more than it does
  3. The process of signing up for cryonics is painful, slow, and difficult
  4. Cryonics is "crazy" and "sounds weird"

Of these, #1-3 could be addressed by a physical store space that makes it easy to sign up. #4 is more difficult, but in some ways, it's actually an advantage here. The "weirdness" of cryo makes it easy to attract media attention, which in turn drives customer traffic.

The next question, of course, is financial viability. In the US, there are roughly ~2,000 people signed up for cryonics right now. On the one hand, making it so much easier (and creating affordances for) signing up should greatly boost customer numbers. On the other hand, a lot of potential customers don't live anywhere near the Bay Area, and won't visit anytime soon, which makes them less likely to buy (you can still offer remote signups, but that will re-introduce many trivial inconveniences). On balance, I think a competent store owner could expect somewhere in the range of hundreds of customers in the first year, which means that to make financial sense, revenue per customer would have to be pretty significant. Fortunately, there is a convenient way to do exactly that: life insurance commissions. Because people rarely switch life insurance carriers, the net present value of each insurance sale is in the thousands or tens of thousands of dollars, which means commissions on sales can be correspondingly large. There is already one life insurance agent (Rudi Hoffman) who specializes in cryo signups, which is a significant indication that the economics there make sense.

Online dating as a neglected cause (2017)

I think one of the ignored causes in EA might be online dating sites. That probably sounds crazy, compared to things like war or slavery or extreme poverty. But if you asked random Americans what they really cared about, what mattered the most in their lives, I expect a majority would say "my husband (or wife) and children". A lot of people care about this problem, and it's super important to them, but the organized effort going into it is pretty small because (like Sendwave or newspapers) the ratio of producer surplus to consumer surplus is tiny.

When I think about the user experience for something like OkCupid, which is the current market leader, it looks like:

That's just a terrible UX, by any standard. I think you could do a lot better, especially if your main motivation was altruism rather than maximizing revenue, and especially with recent machine learning techniques that can process and infer information intelligently and automatically.

Konrad Zuse (2017)

Konrad Zuse is one of the people who most inspire me. Few people know of him, but he built the world's first programmable digital computer, and he did it despite obstacles that would make Bill Gates blush. When he was 22, Adolf Hitler came to power in his native Germany; the ensuing turmoil meant that he worked in near-total isolation from Turing, Shannon, and other Anglosphere researchers. At age 25, without any significant funding, he moved in with his parents and built the first real computer prototype in their living room. At age 29, when Germany invaded Poland, he was drafted into the Wehrmacht; he managed to talk the German air force into letting him stay in Germany and giving him funding for his research. At age 31, he built the Z3, the first practical programmable computer, only to see it destroyed two years later by an Allied bombing raid. His parents' apartment was bombed a few months later, along with his earlier prototypes. Undeterred, Zuse started work on the Z4, the world's first commercial computer; he had to move it out of Berlin to avoid the Soviet army, and then when he managed to finish it elsewhere, he immediately had to move it again because of the war. When the Allies won, they set about systematically dismantling German industry, making computer manufacturing impossible; at this time Zuse worked on his Plankalkul, the first high-level programming language, but was unable to publish it. He was eventually able to start computer work again, and sold computers through his company Zuse KG until 1967. Wolfram's idea of the universe as a cellular automaton was stolen from Zuse, who published it forty years earlier in "Calculating Space".

Body cameras and citizen patrols for downtown San Francisco (2017)

I think downtown San Francisco's unpleasantness - the crime, the drug dealing, the vandalism, the street harassment, the open-air defecation - is a significant long-term productivity drag for people living and working here, although I sadly have no hard data. The obvious solution is to call for a police crackdown, as many have. But, aside from that being politically distasteful, the SFPD has major problems of its own.

My guess is we can take a better approach, with a little elbow grease and organizing skill. There's already a San Francisco organization of Citizens On Patrol, neighborhood volunteers who walk the streets and watch for crime (Castro Community On Patrol). Anybody can sign up, and they're currently under-staffed. The obvious problem, of course, is that the police won't respond to reports. But anti-police activists have already shown us the perfect tools for that: body cameras. If everyone wears cameras, then people saying how bad things are won't just be "teenagers trolling on Reddit". When someone gets randomly sucker-punched by a screaming maniac, and the police don't show up for four hours, and then they shrug and don't file a report, everyone on YouTube will be able to see exactly what happened. It would be hard work, but I think if these videos get posted over and over and over, day after day, month after month, with good writing and social media skills, you could exert enough pressure to both make streets safer and reform an inefficient and corrupt police department.

Subway construction costs as an economic bottleneck (2017)

Subway construction may seem boring, but the high cost of it inside the US is a good candidate for a bottleneck on the entire American economy. The two major expenses that haven't improved with technology are housing and health care, and a city's transportation system effectively determines how much housing it can have, since most people will only live where job and social centers are accessible. The topic of tunnel costs is also almost entirely neglected. New York City, with one of the world's largest subway systems, knows so little about why construction is more expensive there than anywhere else that they deferred to Alon Levy, a random math postdoc who blogs about it sometimes:

"How New York City's megaprojects compare in cost to those in similarly developed countries around the world is a question that is, somehow, very rarely studied. [New York City Comptroller] Stringer's spokesman said the comptroller relied for his numbers, in part, on a mathematician named Alon Levy, who's now completing his post-doc at the Royal Institute of Technology, and who notes, in his blog Pedestrian Observations, that, mass transit is a 'side interest' for him and 'entirely unrelated to my work.'" (Politico)

Elon Musk's Boring Company recently published an FAQ on how tunneling costs can be reduced. But it focuses exclusively on technical construction issues, which leaves a mystery the question of why some cities are ten times or more as expensive to build in.

(Note: earlier, shorter take on the theme of the published "NYT cites Alon Levy on subway construction costs" (Dec 2017) — related but not a duplicate.)

The Republic of Minerva, and what it takes to start a country (2017)

Several attempts have been made to start a new country founded on libertarian principles, but none have really succeeded. In most cases, the underlying problem seems to be finding a critical mass of people who care enough. In this case - the short-lived Republic of Minerva - it appears that the Tongan soldiers sent to reclaim the area were equipped with nothing but bolt-action rifles. It would have been easy to fight them and win, but nobody did. While I totally empathize with not wanting to risk death or permanent disability in a shootout, the essential definition of a state is an organization with a monopoly on violence; a group with no one willing to risk violence can't be a state.

Edit: Although "not enough people care enough" was reflected in the Minerva crew not being willing to use violence, a closely related problem was that so few people were willing to join the South Pacific crew at all. The smallest true states today have a population of about 10,000. It's unclear exactly how large the Minerva project was - at a guess maybe a few hundred people, but certainly nothing like 10,000.

The long-term decline in US policymaking capacity (2017)

(Note: Political news is off-topic here, but serious long-term political strategy is not)

At last year's EAG, there were a few people who held mid-level US policy positions, which is an interesting contrast to the typical EA backgrounds of computer science and philosophy. One of them argued for more EA attention to policy, on the grounds that "right now" (in 2016) the government was split between political parties, but "later" this would change, and policy experts could go back to getting more done.

I'm sympathetic to the conclusion, but I think this argument is a common myth, created to justify a wishful-thinking attitude of "business as usual". There seems to be a long-term secular decline in the US's ability to make policy changes, dating back to FDR and WWII. I argue that the historical record supports that this was the cause of recent dysfunction, not just "split parties". Eg., nobody could argue that Reagan was an inconsequential president, but he had a Democratic House for his entire term in office. In fact, every Republican president had a Democratic House from 1952 through 1994. Likewise, eg., the major welfare reform bill signed by Clinton in 1996 was written when Republicans were in the majority in both the House and Senate.

A Zagat for scientific journals (2017)

Meta-science plan I thought of last year, but haven't had time to implement: a lot of scientific journals, especially in "softer" fields, are affected by the replication crisis. We could encourage journals to take action with a rating system for whether each journal followed modern scientific best practices, like Zagat for restaurants. These might include things like:

GiveWell and the long-run supply of surgeons (2017)

GiveWell's focus on the short term causes it to not evaluate many charities, but I think it leads to mistakes even for charities that it does evaluate. Eg., a major argument against donations to fund surgeries was that there was a limited supply of surgeons. At any given moment, this is true. However, economically, increased funding for surgery should cause higher wages for surgeons. Over the long term, this should in turn cause more surgeons to be trained, and more to move to the relevant areas from elsewhere.

Over-the-counter birth control (2017)

A friend told me that the most environmentally friendly thing a man could do was get a vasectomy. I think it'd be much more effective to fight to make birth control available without a prescription - one of those issues where libertarians and leftists can team up. Unplanned pregnancies have very serious consequences; I suspect they're a leading cause of poverty.

On the sepsis trial (2017)

Blog post by Sarah Constantin on a new treatment for sepsis - a very common cause of death - that could be highly impactful if proven in clinical trials. More funding is needed to complete the trials. Disclaimer: I have not personally analyzed this research; a large fraction of "potential medical breakthroughs" (in some fields nearly all) are pure PR, and never had any chance of working. I don't know the researchers at all, and can't make any claims about whether they would use money wisely. I have a very high opinion of Sarah's analytical capabilities and trustworthiness, but would still strongly recommend further investigation before writing anyone a check.

Excluded from this bundle (and why)