Types of planners and plans for EAs

by Diego_Caleiro 10th Oct 20145 comments

5


What would it look like if we were winning? 
What would it look like getting from here to there?

This post:
Tales of people within effective altruism who succeeded because they had plans
Secret planning
Plan design

Next post:
Have an idealized plan
Murphy-jitsu your plans
Compare with other people's plans last


Tales of people who succeeded because they had plans

On a foggy Wednesday afternoon we were scheduled to talk for an hour. All the people I had met in that building were interesting, smart, and quirky, so I expected that, no less. Compared to the other visitors to the appropriately named Haste street apartments, however, he seemed more focused and organized, which was not saying much. Within maybe fifteen minutes of conversation we mutually perceived a distinctive trait we shared, but others didn't. Everyone in that building was smart, everyone was altruistic, everyone wanted the world to improve, but both of us had a plan. Three hours later we shook hands and decided to work together.

My plan and Geoff Anders' plan were at different levels of analysis, using different methodologies, different levels of abstraction. That was before Leverage Research or IERFH (Institute Ethics Rationality, and Future of Humanity) had their respective names, or even existed at all. Those plans, mine written, his in flowchart form, were the seeds of two institutions.

Geoff used to say that the way he got people to work with him was basically by impressing The Plan upon them. It worked for me; I decided to dedicate 15+ hours of my week to what would become Leverage Research, even though my main quest was still to pursue my relative advantage - returning to Brazil to create my own institution there - It also worked for many of Leverage's initial members.

Not long after that Wednesday, another planner arrived at the Haste apartments. Luke Muehlhauser, who would later become the CEO of MIRI. Though many different things have contributed to Luke's position, in particular blogging at commonsenseatheism.com, I perceive his continuous insistence on a yearly plan for MIRI as pivotal to his ascendance to CEO.

Those were important days at the Haste apartments and they were important days because people were making plans. Plans that start with a box saying "we are here" and end with a box saying "the World is significantly improved".

Another case study of having plans is that of Matt Fallshaw, a successful effective altruist who has created several successful companies, giving him substantial resources to direct to the causes he finds most important. Fallshaw is remarkable for his productivity. He sleeps for four hours a day, fasts daily and exercises by sprinting repetitively up a hill. When asked about productivity techniques, Fallshaw described basically techniques for making good failsafe accurate plans. His advice is: 1) Use a randomized spaced repetition system like anki to make sure you are successfully installing the habits that will cause you to execute your plan. 2) Make sure you have above average energy when making and reviewing plans. 3) If a goal seems impossible or intractable, use a 5-Minute timer (i.e. CFAR's Focused Grit) to try to a) solve the problem, or, failing that b) Generate a set of action sequence steps, each less than 5 minutes long, that will take you 80 or 100% of the way towards the goal. 4) Use a system for freeing attention, such as Getting Things Done. 5) Apply the principles of Theory of Constraints to large plans.

Finally, another successful founder of EA projects, Peter Thiel, has recently spend considerable effort in giving presentations about and writing a book chapter on the idea that You are not a lottery ticket. By that he means that trying to find a reference class to which you belong, and evaluating your chances of obtaining a particular outcome can only go so far. It is not just chance that brings new ideas - zero to one - to the world. Having a probabilistic view of the world, an indefinite view of the future, is dangerous - yes, this applies to you too, Bayesian startup creators. Let me illustrate this with some examples of actions you'd take:

If you see the future as indefinite and optimistic, if whether you succeed or not depends on chance, then the best thing you can do is improve your chances. Since chances are imagined relative to a reference class, then you will feel incentivized to be in the right reference class. So entering Harvard will seem like a good place to seed a successful startup. Once you get in, the grass will be greener if you drop out. No reference class beats Harvard drop-outs right? If you are an investor, having a broad portfolio of stocks will also feel like a sound investment decision, since you "know" it’s unwise to choose any definite thing, including stocks. That is not how the world works, according to your indefinite model. In an indefinite world, everyone outsources everything, no one takes responsibility and makes specific decisions, things stop being built, everyone fights for reference class hopping, and hopes that over the long run the dice will turn in their favor.

Thiel contrasts this with a definite optimistic view of the future, which can be seen in Marx, Hegel, US policy from 1950 to 1960, and I'd argue also in his investment decisions, in planners like Anders, Luke Muehlhauser, Toby Ord, Nate Soares, Thiel himself, and a minority of effective altruists. According to the definite view, it is worthwhile to have ideas, to seek opportunities, to find market failures, and to act on them. It is desirable to conceive of specific ways the world could be, and design plans to move us there. Thiel says this approach was more common in the 1940’s:

Bold plans were not reserved just for political leaders or government scientists. In the late 1940's a Californian named John Reber set out to reinvent the physical geography of the whole San Francisco Bay Area. Reber was a schoolteacher, an amateur theater producer, and a self-taught engineer. Undaunted by his lack of credentials, he publicly proposed to build two huge dams in the Bay, construct massive freshwater lakes for drinking water and irrigation, and reclaim 20,000 acres of land for development. Reber though he had no personal authority, people took the Reber Plan seriously. It was endorsed by newspaper editorial boards across California. The US congress held hearings on its feasibility. [...]

But would anybody today take such a vision seriously in the first place? In the 1950's, people welcomed big plans and asked whether they would work. Today a grand plan coming from a schoolteacher would be dismissed as crankery, and a long-range vision coming from anyone more powerful would be derided as hubristic.

We can see two messages here:

1) Having a definite view of the future has been a major driving force of few particularly productive people in the current effective altruist movement.

2) Adopting a definite view of the future predicts success, and requires making specific action plans.

 

Secret planning

If you are convinced that having a definite vision of the future is a good idea, then you have two choices: create your own view of the future and make a plan to make it happen; or adopt someone's specific view of the future, and make it happen.

These are the most important actions. Moving from 0 to 1, and from 1 to 2. Not unlike geometry, getting from 2 onwards is not only easier, it what everyone is already doing. If you are determining a line, the third point is one point too many. It is, in a word, redundant.

If you adopt a plan that more than 0 or 1 people already have then you are incurring in a specific failure mode, you are zooming in at the wrong level of granularity. For instance, suppose you consider the creation of Friendly Artificial Intelligence to be the most important task. Well, so do Muehlhauser and Yudkowsky, therefore you should look elsewhere right? No. What you should look for are actions - steps instrumental to the creation of Friendly Artificial Intelligence that others are not looking into. As Thiel contends, this is very hard, it is unnatural to try new things. It is especially unsettling to look for failures in worldscale plans in part because, should you actually find an error, you may inherit responsibility for fixing it, it is now your responsibility to make it happen. You are the only person who knows the secret path, it is a necessary route to arrive at your goal, and if you don't go through it, no one will. Furthermore, others may not only fail to see the path, but be dismissive of it’s importance. This is good indication that your planning lens are zooming in from the right distance - if everyone agrees with you, you are too far, if no one understands you, you are too near - when in between you may still be wrong, but in that zone fly the ideas Peter Thiel calls secrets, ideas that looks clearly bad, but happen to be actually good. That is the group of ideas Thiel seeks to fund, in case you are wondering. When you find a secret, you become irreplaceable.

 

Plan design

Now what would your plan look like?

Your plan may be an emotional grandiose vision of what is to come. Yet that is suboptimal.
It may be a narrative, with describes in linear fashion what needs to happen. Yet that is still suboptimal.
It may be a call to arms, so that people see the importance of what you have noticed. - We are getting hotter, but still not optimal.

Remember Fallshaw's advice that using Theory of Constraints can help in designing your plan, it helps you predict time pressures, releasing valves before they constrain your ability to move. Furthermore, your plan must have parallel paths depending on how reality reacts to your actions, and so that actions can be done in parallel. Finally your plans must be understandable and communicable, which is how Geoff attracted the first researchers to Leverage.

The optimal plan then , is, like Leverage's old plan, a flowchart, or another clear structure (my EA plan) that permits parallel paths, hierarchization with different levels of granularity visible at particular times. A graph. I use Freemind for ease of keyboard shortcuts, then Mindmup for different visualization, though it loses formatting and pictures.

Maybe you think Geoff2011 choose the wrong level of granularity to design a plan. Perhaps, but just having a plan allows much more thorough monitoring of what is going on, tractability of how well we are doing in relation to predictions, and specially pre-conceived reactions in case reality reacts in one or other particular way.

Nick Bostrom's book Superintelligence doesn't create one causal story and run with it. It maps out many different ways in which reality can behave as it approaches an intelligence explosion. In unprecedented detail, it scrutinizes ways to steer the future so that reality's moves are, as much as possible, amenable to some level of control, before, during, and after an intelligence explosion. It is not a complete flowchart, but it is the best set of predictions and contingency plans we have so far regarding this particular cause.

Superintelligence and Leverage's plan are good examples of what a plan should look like. An initial plan to personally act on the secret you discovered, however, will justifiably be simpler. We will return to that on the next post, for now, here is a template of a plan without content so you can use it for your own EA plan. Freemind has faster commands, Mindmup has more visual variety.

Try things.

 


If anyone solved the problem of copying texts from google docs to here without losing some spaces, bold and italics when you clean it let me know

Thanks Ryan Carey for careful commenting and editing, Oliver and Nate for increasing the urges and urgency.