The Mad Scientists Theory of Governance is that most organizations would be more effective if they empowered employees with the freedom to make bold, apparently reckless decisions if they take responsibility for their decision.
The name comes from common lore about Manifold itself, that we operate like mad scientists (or, "crazed scientist barons"), first articulated by this comment from Odocare:
We liked this comment so much we printed it on a poster that hangs in our office.
When someone has the freedom to pick what they think is most important to work on, without restriction, you get more creative and more impactful output. In particular, you get more bets that pay off a huge amount, and those more than outweigh the failures.
When applied across an organization, a lot more gets done. Experimenting leads to a lot of collective learning and new sources of value are unlocked. Moreover, employees enjoy working in this environment because they have more control over their work and can see clearly the difference they are making.
It might even appear chaotic. Some customers might get upset when things are changed on them or seemingly anything that could go wrong does go wrong.
But there's a utilitarian calculation that justifies it. You are not optimizing to serve one customer — not even all your current customers. You are optimizing to serve current and future customers. Taking risks is commonly worth it in service of building an exponentially increasing business.
Dominic Cummings on No. 10
The direct inspiration for writing up this piece comes from an interview with Dominic Cummings on Dwarkesh's podcast where he diagnoses the failures of the UK government to effectively govern.
One problem is that no part of the bureaucracy is empowered to change anything or get anything new done, except for the Prime Minister. The PM alone is able to bless a new initiative and waive all the red tape of regulations, including from HR, the Treasury, and other departments. All progress is thus bottlenecked on one person (who is mostly busy dealing with the media and various crises).
It occurred to me that the right way to govern even bigger and more serious organizations, like the UK government, is the same mad scientists model that our small startup operates on.
Hire talented people and let them be live players and make mistakes. When the stakes are high, our natural inclination is to slow down and be more careful, but I think that is wrong.
The stakes are higher, so not experimenting with a land tax could mean lots of inefficiency in land use for years to come. In other words, the upsides are also larger when the stakes are higher, so the underlying utilitarian calculation doesn't change.
I submit that the best executed governance of the UK (or any country) would look fairly chaotic: lots of new initiatives with others being phased out; young bright people taking over huge departments and changing their mission; lots of hiring and firing; private data being leaked; media storms about controversial comments made by government employees; a few people dying in the construction of mega-projects or from trials of experimental drugs; high stakes negotiations with foreign powers including some unprecedented deals and failures; upset citizens who are protesting a change that they think devalues their neighborhood; and so on.
Early Google
Or, think about early Google and what made it so cool. They had a bunch of brilliant people that were very free to work on what they want, with a variety of cool initiatives, like Google Books or Gmail. Empowering employees made the place dynamic, with the sense that cool new stuff was being created every day, and also made it a fun place to work. Now, Google is more often trying to protect the businesses that it already built, and even their core search product has degraded.
I think that if Google had kept up the willingness to experiment (perhaps by only hiring people willing to experiment), their search would be better, and they would have higher revenue, in addition to many more successful products.
Proposed structure
In an organization of mad scientists, everyone is dictator. Meaning, they are free to do whatever they want. The only exception is that they are arranged in a hierarchy and must obey their manager, though managers should use their power sparingly. Of course, managers would probably find it impactful to coach their reports on what areas and initiatives to work on. They should encourage experimentation and try not to reject an idea that their report is fired up about.
To pull this off, you'll need to encode experimentation into the core tenets of the organization and speak about it frequently to set the culture and expectations. You should try to hire only independently-minded people who can come up with their own ideas and pursue them.
There's a question of whether this structure can scale — does it only work at small startups?
I think that large organizations are merely correlated with not having a culture of experimentation. I don't think that it is impossible or that it would not be efficient. I think it would be very efficient, actually, because each employee would be more likely to find something impactful to do and efforts compound. Perhaps the only reason it doesn't exist in larger organizations is that no one has seriously tried.
---------------------------------------
Thanks for reading. I will resolve this to YES if someone at least as important as Scott Alexander (according to my judgment) references "The Mad Scientists Theory of Governance" before 2030.
Related questions
Sounds a lot like valve to me. Have a look at their new employee guide.
https://steamcdn-a.akamaihd.net/apps/valve/Valve_Handbook_LowRes.pdf
This is my mad scientist theme song (warning -- classical music):
https://open.spotify.com/track/0xFXH2HRAUAiDrI291SE5g?si=5f9963d13aed4b71
@BillyMcRascal Yes, Will MacAskill counts! A reference includes either the link to this page, or contains "The Mad Scientists Theory of Governance" (or similar) with an attribution to me.
The "mad scientists theory" sounds like just the default for how anything works; you get N people who all try to somehow work together to do the thing, according to their judgment.
I think the main reason most companies don't work like this is that their labor is 1) unskilled at their job (very broadly speaking) and/or 2) unaligned with the goals of the company. As long as you can hire skilled labor that is aligned with the goals of the company, you can run your company like this, but that's hard to do.
Note that "skilled" in this case doesn't mean just skilled at their most specialized skill. For example, I would not be a good worker in the "mad scientist" paradigm if I were only a good programmer, and I had very bad judgment for other things about the business. I need at least enough judgment to defer to others on tasks which they are better than me at. Otherwise I will go around messing up random things. So it's reasonably challenging to find appropriately skilled labor. I think hiring from the rationalist community is a huge leg up.
So in short, the traditional approach is trying to get value from the giant pool of cheap unskilled and unaligned labor by establishing a structure that can direct it.
@mqp I see, perhaps true.
Maybe it's a recipe for the top layer of an organization in that case. Like, you have a special caste of mad scientists who you trust to be aligned and have good judgment, and they are fully empowered. And then you have the rest of the employees.
This actually kind of mirrors Google's employees vs. temp workers. Temp workers have a different color badge and don't get as many privileges. I understand this is not a popular structure in our day and age.
As a Brit I have to chip in here and say that Dominic Cummings is widely disliked across the political spectrum. For many reasons, but one being that it’s very easy to say “this is terrible, we should tear it all out and start again” and doing so can make you sound very clever.
But, it is very hard to identify real value that is lost when you tear everything out, and if you only ever tear things up and build quick, easy, “common sense” solutions then you can leave deeply flawed messes behind far more difficult to resolve than the thing that was there in the first place.
Also, “a few people dying in the construction of mega-projects” is a fucking awful ivory-tower take coming from someone whose biggest work health risk is heart disease from too many office pizza parties.
@Noit Sure.
You should listen to the Dominic Cummings interview — he sketches a compelling picture of what No. 10 is doing wrong and how to fix it. I don't think he's saying merely "tear everything down"; he's saying to focus on long term reform over reacting to the media every day, relax bureaucratic rules, delegate authority more so that it's not all bottlenecked on the PM, promote younger talent, and remove people who are too invested in the current system.
Also, I intended to ruffle some feathers with the take that the best governance still has a few people dying in accidents. I think that is likely, but obviously that is not the goal. It would be fewer than with the current government. Just improving healthcare with market reforms would save thousands of lives per year.
The stakes of governance are really high. Bad policy of various sorts affect the lives of millions of people and my thesis that the way to get to better governance is to experiment and empower mad scientists.
In other words, this is the path to saving the most lives.
What you've built is unique and awesome - Manifold FTW. This isn't incompatible with what you wrote, but I'd observe that, for plenty of organizations, the role of innovation is far less important than for Manifold / other startups. The low-hanging fruit has been picked for many (most?) organizations and what is primarily needed is task completers vs operators.
But my overall sense is still that many organizations would benefit from providing more mad-scientist autonomy for innovation!
@CarsonGale Nice, I think you're right. Like, if it's an Amazon warehouse, maybe this model wouldn't improve productivity. Or, maybe there's portions of the government that just need to do the grunt work of processing forms. (I wonder if AI will handle these cases soon though haha.)
I think this point is the weakness:
"In an organization of mad scientists, everyone is dictator. Meaning, they are free to do whatever they want. The only exception is that they are arranged in a hierarchy and must obey their manager, though managers should use their power sparingly. Of course, managers would probably find it impactful to coach their reports on what areas and initiatives to work on. They should encourage experimentation and try not to reject an idea that their report is fired up about."
What is the incentive if you get really large for managers not to hold power and to force through their priorities. Alternatively what is the thing that causes people to lose power of they use it badly. This works on a small scale with a close knit team and competent governance. Does it work with scale, entreeism and bad incentives?
I think a lot of what you have is an internal market for status. I agree that this might be better in government, but a lot of people might hate this. After all, they could lose status as easily as gain it,
@NathanpmYoung Great comment!
I generally think that hierarchies work pretty well, so if you can incubate a culture of empowering people to try stuff, that would make it even better. The government often does worse because it's not a real hierarchy when managers cannot fire.
Part of keeping the culture at scale is hiring the right people. And, someone up the chain should be able to observe unproductive behavior and sack the people involved.
Having it turn into an internal market for status is interesting. It's true that some will optimize for the best impression on colleagues rather than creating the most value. But that should still be compatible with getting a lot done. It probably does need some iteration in the real world to figure out failure cases and how to fix them.