neumannOneProcess2020

Activity Space

Modularization enables pieces to be used in many different technologies

The tech tree is fractal

Technological roadmapping can give an idea of where technological evolution can go and how to get there

A heuretic’s impact is not endogenous to the technology but depends on the environment

The tech tree is a scaffolding inside the knowledge frontier

Technodiversity is important

The creation of a heuretic has two steps - conception and implementation

Top of mind

We don’t have good institutional ways to create multi-point changes

Phenomena are things that happen without a human consciously making them happen

Impact in innovation is analogous to fitness in evolution

The tech tree model of heuretics is a directed acyclic graph

Heuretic is a better word than innovation

The difference between specialists and generalists is fractal in nature

Science and technology advance through the same fundamental mechanism

Interfacing between closer neighbors on the tech tree is easier than between nodes that are farther apart

There is a higher probability of accidentally unlocking a closely related node on the tech tree

Untapped technical people near the end of their careers may have more potential than young people

Charlie Songhurst conversation 31 Aug 2021

Dispersion is a more appropriate word than dissemination or diffusion for heuretics

Uptimal solutions in science and technology are rarer than you think

The affordance of a heuretic dictates how it can connect with another node on the tech tree

Breakthrough fundamental science

The difference between specialists and generalists is not clean but depends on which nodes they can extend

Ring Species are situations where group A can mate with group B and B can mate with C but C cannot mate with A

The tech tree model suggests that more work on more root nodes is important

How do you incentivize people to work on nodes farther down the tree?

Each heuretic by default inherits the constraints, affordances, and assumptions of its parents

Observational technological phenomena

Adam Marblestone feedback on design

Science is discovering and exploring phenomena while technology is using a principle to exploit phenomena towards a use

 from airplane design to drug design

What are the different moves that can happen with a heuretic?

# §Tech Tree Model of Heuretics
Imagine a tree growing “up” with “root nodes” “lower” or “deeper” in the tree. The height is some monotonic function of time, but it definitely isn’t 1:1. 
- [[The tech tree is fractal]]
- [[The tech tree model of heuretics is a directed acyclic graph]]
	- [[Each node on the tech tree can recursively be decomposed into more nodes]]
-  [[The tech tree is a scaffolding inside the knowledge frontier]]
- [[Connections create new knowledge]]
- [[The tech tree creates new nodes either through idea sex or learning by doing]]
	- [[The upshot of learning by doing being a first class process is that the implementation is as important as where the idea came from]]
- [[Tech trees first showed up in the Civilization boardgame in 1980]] but [[There do not appear to be any intellectual connections between video game tech trees and history+philosophy of science+technology]]. [[Tech trees are intuitive]].

### Heuretics
- [[As a platonic ideal, research is an unscheduled quest for new knowledge and inventions with hard-to-predict outcomes]]
- [[Heuretics are either inventions or discoveries]]
- [[Heuretic is a better word than innovation]]
- [[Heuretics come out of people’s brains (TECHNOLOGY IS PEOPLE)]]
	- [[Heuretics cannot exist separate from humans*]]
	- [[New ideas need to come out of a single mind]]
	- [[Brains are wombs for heuretics]]
- [[The creation of a heuretic has two steps - conception and implementation]]
	- [[Conception is the imaginative design of a heuretic]]
	- [[Implementation is the actualization of the heuretic in the real world]]
- [[The process of creating a new heuretic node is like building something out of legos]]
- [[Science and technology advance through the same fundamental mechanism]]
- [[Modularization enables pieces to be used in many different technologies]]
- [[Modularization enables the higher level technology to improve continuously even if pieces improve discretely]]
-  [[The affordance of a heuretic dictates how it can connect with another node on the tech tree]]
-  [[Each heuretic by default inherits the constraints, affordances, and assumptions of its parents]]
-  [[Interfacing between closer neighbors on the tech tree is easier than between nodes that are farther apart]]
- [[There is a higher probability of accidentally unlocking a closely related node on the tech tree]]
- [[Uptimal solutions in science and technology are rarer than you think]]

### Dispersion
- [[Dispersion is the process of a heuretic impacting the world]] ::needs work::
- [[Dispersion does not happen instantly and often requires a lot of work]] ::needs work::
	- [[List of things that affect dispersion]]
- [[Constraints limit the impact and dispersion speed of a heuretic]] ::needs work::
- [[Dispersion is a more appropriate word than dissemination or diffusion for heuretics]]

### Specialists and Generalists
-  [[The difference between specialists and generalists is fractal in nature]]
- [[Modularizing ideas and specialization are the same thing]]
- [[The difference between specialists and generalists is not clean but depends on which nodes they can extend]]
- [[Fast moving fields have lots of unconnected nodes]]

### Other Phenomena
- [[General purpose technologies are technologies that are lower on the tree]]
- [[Macroinventions are the tail end of the impact power law distribution clashing with our intuition]]
- [[The phenomenon of seemingly trivial technologies becoming incredibly important happens because there is fog of war on the tech tree]]

## Tech Tree is analogous to evolutionary tree
- [[Impact in innovation is analogous to fitness in evolution]]
-  [[A heuretic’s impact is not endogenous to the technology but depends on the environment]]
- [[There are many heuretic environments]]
- [[Ring Species are situations where group A can mate with group B and B can mate with C but C cannot mate with A]]
- [[There are scientific and technological problems that make other technology advances irrelevant by altering the environment]]
- [[In the same way that species don’t suddenly replace each other, technologies and ideas don’t suddenly replace each other]]

## Upshots
-  [[Technodiversity is important]]
-  [[The tech tree model suggests that more work on more root nodes is important]]
- [[How do you incentivize people to work on nodes farther down the tree?]]
- [[People are incentivized to work on nodes higher up the tech tree]]
- [[Systems must take performance hits to get out of local optima]]
- [[The tech tree model suggests that explicitly drawing out tech trees is important]]
	- [[The structure of the tech tree can inform actions that are short term suboptimal but long term useful]]
	- [[Roadmapping could be thought of as backwards-chaining together a series of conception-only heuretics]]
- [[What does technological genetic engineering look like?]]
- [[What does technological husbandry look like?]]
- [[What does technological mutation breeding look like?]]

### References
- [[neumannOneProcess2020]]
-  [[Breakthrough fundamental science]]
- [[Stagnation and Scientific Incentives]]
- [[arthurNatureTechnologyWhat2009]]
- [[narayanamurtiCyclesInventionDiscovery2016]]
- [[Road-mapping Biology - Article]]
- [[Adam Marblestone Podcast 2020-03-31]]
- 

[Web URL for this note](http://notes.benjaminreinhardt.com/§Tech_Tree_Model_of_Heuretics)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/§Tech_Tree_Model_of_Heuretics)






§Academia Constraints

Peer review has failed as an institution

Peer review doesn’t catch duplications

Human capital theory of stagnation

People are incentivized to work on nodes higher up the tech tree

It is easier to have a multipoint mutation in technology than in evolution

Peer review gives peers veto power over each other

Startups and Academia are bad at addressing structural constraints

# Academics are incentivized to do peer-reviewed publishable work
Peer review published papers both determine an academics livelihood and status. Therefore, they are incentivized to spend their time doing work they think can be published. 

On top of this [[Academic credit is a first-to-the post system]]

[[People and organizations are all playing some game that has different ways of gaining status and power]] In Academia there are a couple of long term games:
- Gain [[Kleos]] through contribution to the [[Knowledge frontier]]
- Just get to scratch your itch to investigate certain things
- ….

[[People create proxy games for long term or hard-to-measure games]] and in academia the entwined proxy games are:
- Peer reviewed Published papers
- Citations on those papers
- Grants (which are usually based on papers) 
- Awards (which are usually based on who knows you which comes from papers)

It’s a safer strategy to publish lots of papers than to try to write one super-cited paper because it’s very hard to know which papers will be highly cited. Additionally everybody is impressed when you say “I’ve published a bajillion papers” while only a subgroup might know your super-cited paper by title unless you’ve [[hit the top of the columnade]]. 

Tenure is based on the number of papers you’ve published and the amount of grant money you’ve brought in.  Without tenure, academia is extremely rough. People who don’t think they can get tenure leave universities. [[National labs and other government research centers are a research alternative to universities]].

[[The peer review and citation system incentivizes people to work on things that other people think is interesting]]

### Related
- [[§Academia Constraints]]



[Web URL for this note](http://notes.benjaminreinhardt.com/Academics_are_incentivized_to_do_peer-reviewed_publishable_work)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Academics_are_incentivized_to_do_peer-reviewed_publishable_work)




Shadows of the Future

Changing paradigms is hard

People giving out grants try to derisk them as much as possible

DARPA Program managers pull control and risk away from both researchers and directors

How does money work in PARPA?

Pieces of the Action Forward Intro Notes

Reaper functions beat culture decks

§Corporate R+D Constraints

People are more averse to short term losses than they are attracted to long term gain

NSF Program officers can have multi-decade tenures

DARPA’s aversion to people with a web presence may be how they avoid asymmetric career risk

The idea that academia is for crazy exploration and companies are for low risk exploration is bullshit

Lots of innovation is stifled because of the overhead of going off of the beaten path rather than direct forbiddance

The only time you can get philanthropies to do interesting things is at the beginning

People are hesitant to try things when they don’t know where the line is and the consequences are large

DARPA’s aversion to people with a web presence hints at a way that the internet has eroded institutions

How do you have an institution that avoids asymmetric career risk and going rogue in a world where everybody is a free agent?

Asymmetric career risk impacts corporate R+D

Often spending money is the primary goal over the output of that money

There are no perceptions of outsized returns to philanthropy

The problems we associate with government orgs is not caused by government per se but by low accountability vs high accountability institutions

Nadia Eghbal conversation 7 Jul 2021

Glossary

Career hysteresis

David Lang conversation 20 Apr 2021

# Asymmetric career risk
You can’t get fired for failing in the conventional way, especially early in your career. You can get fired or have you career hampered by failing in a weird way.

On the flip-side, in most career situations you don’t capture a lot of the value of the wild success that can come from taking weird risks. 
This is one of the [[§Corporate R&D Constraints]]

![](../BearImages/72D8815F-0DBD-4198-90E8-48026CE139D3-621-000434AF61C86F43/9E902B34-82EF-4CCE-BD2C-5FB18A6C9537.png)

### Related
- [[Shadows of the Future]]
- [[We do not give enough credit to good philanthropic funders]]



[Web URL for this note](http://notes.benjaminreinhardt.com/Asymmetric_career_risk)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Asymmetric_career_risk)




§Tech Tree Model of Heuretics

§Specialists vs. Generalists

Different research fields are different and those differences change over time

Many approaches to augmented knowledge generation ignore humans

# Fast moving fields have lots of unconnected nodes
You could think of fast moving fields at evolving more like viruses and slow moving fields like mammals. Mammals are extremely complex. Viruses just depend on replication errors to do crazy shit.

### Related
- [[Connections create new knowledge]]
- [[Specialists create knowledge within fields]]

### References
[[Creativity at the Knowledge Frontier: The Impact of Specialization in Fast and Slow-paced Domains]]



[Web URL for this note](http://notes.benjaminreinhardt.com/Fast_moving_fields_have_lots_of_unconnected_nodes)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Fast_moving_fields_have_lots_of_unconnected_nodes)




The number of researchers has exploded

Successful professors are not necessarily the best researchers

Universities have been pushed more towards applied research since the end of WWII

In defense of academia outline

How and Why Government, Universities, and Industry Create Domestic Labor Shortages of Scientists and High-Tech Workers

Mark Johnson the Elder Conversation 4 Jan 2021

Things that are not paper-worthy enough for academia and not product-focused enough for a startup

Nobody will admit that a field is stuck

The university is being unbundled but nobody is taking on the research part

What is the history of Universities with respect to research?

Second Lieutenants at AFRL are like graduate students in universities

Research scout network

Another downside to importing graduate students to work in labs is that it makes papers less intelligible

Michael Filler Conversation 25 Aug 2022

There is a group of people who are capable interested in science and the current institutions are not working for them

# Graduate students are the labor in academic science and engineering
Modern science needs a lot of skilled labor to do not particularly exciting jobs. Pipetting, building devices, coding, running tests. Arguably they are creating a lot of value for the PIs they work for: progressing their careers, bringing grant money to the lab and the university, creating  knowledge for society. 

Grad students provide this labor much more cheaply than the value they create. Graduate students are paid around $40k a year but it’s much much less than the same smart ambitious people would be making in a real job. The explicit deal is that you get non-monetary compensation. You do get a credential at the end of grad school. The assumption is that this credential is incredibly valuable but that deal has broken down.

This system is problematic for several reasons. 
- Imagine trying to run a company building a problem that demands a ton of [[Tacit Knowledge]], but you could only hire people with almost no experience who would never stay more than five or six years. This constant retraining is an unescapable tension if your labor is students.
- The amount of money you need to support a grad student is actually quite similar to the amount of money you need to support a high-skilled employee at a company, but about half of it goes to the university for administrative overhead and student fees. The university is actually getting a student’s worth of money from grad students to “teach” them but it comes out of grants instead of undergrads pockets. 
- Most graduate students’ incentive is to publish as many high impact papers as possible and generally do work that sounds good to a faculty hiring committee. These incentives are often aligned with their professor and lab. However, if a project requires many people’s worth of work a grad student either won’t make it onto the paper or will be last in a massive list of authors which nobody wants because a finite amount of credit is distributed among all the authors. It’s also not appealing for a grad student to work on a project that won’t produce paper-worthy results on a timescale that aligns with graduating. There’s the obvious case where the timescale is longer than the grad student expects to be in grad school. The more subtle case is that the project takes a few years but is risky enough that it’s not clear that it will produce a paper-worthy result, possibly leaving a PhD student high and dry several years in. This incentive selects against long-term engineering-heavy projects. 
- Hiring committees care about potential professors being able to do novel work. [[The word ‘novel’ is used as an idea bludgeon in academia]]. This criteria means that grad students need to start doing work that stands out from work that has been done before in the lab if they want to get a faculty job. This incentive puts pressure on projects that need a lot of focus over a long time.  [[Academia incentivizes novelty, not focus]]
- Graduate students are primarily funded through grants. Grants are typically renewed on a yearly basis (with exceptions.) So even regardless of what the grad student is ok with working on, there’s a tight coupling between grad students and doing work that can hit paper-worthy milestones on a yearly basis. Additionally, funding grad students through grants effectively ties them to a specific project. Project-coupled grad students make it near impossible to shift effort away from projects that are going poorly towards projects that are going well. Imagine if a startup couldn’t abandon things that weren’t working in favor of things that were.  [[Most grant money is tagged so unrestricted money is like gold for professors]]. 
- You can’t really fire grad students or hire them whenever you need them. For the most part, new graduate students join labs at the beginning of the academic year so if a project needs a lot of labor halfway through the year you either need to plan that out and have a grad student underutilized for half a year or you need to postpone that project until the beginning of the next year. Once you’ve taken on a grad student you’re stuck with them until they graduate or give up - this is a terrible situation for both sides.

There are several theories about how this system came about. Like any [[Complex System]], the real answer is probably “a bit of each of them.”
- It is easier to get grants (especially government ones) that are primarily earmarked for grad students as opposed to lab technicians or labor-saving automation. Why? Some combination of tradition, incentives on the part of the university because they get more fees from grad students than other spending, and the idea that money towards grad students is building up the country’s STEM workforce. 
- [[How and Why Government, Universities, and Industry Create Domestic Labor Shortages of Scientists and High-Tech Workers]] actually argues that the system was set up on purpose to keep wages for high skilled scientific workers low. 
	- Universities get more money when professors bring in larger grants. Because its easier to get big grants for grad-student-heavy projects universities are incentivized to hire professors who do more grad-student-heavy projects, perpetuating the cycle. [[In academia more grant money leads to more power]]. 

### Related
- [[§Academia Constraints]]
- 
- [[Stagnation]]
- [[bhattacharyaStagnationScientificIncentives2020]]
- [[The NSF earmarks a lot of money for training]]



[Web URL for this note](http://notes.benjaminreinhardt.com/Graduate_students_are_the_labor_in_academic_science_and_engineering)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Graduate_students_are_the_labor_in_academic_science_and_engineering)




Nodes deeper in the tech tree take more time and work to unlock

# Interfacing between closer neighbors on the tech tree is easier than between nodes that are farther apart
For example, it’s much easier to get neural networks to play well with parallel processing (and many awesome things have come out of it) than it is for neural networks to play well with rocket manufacturing.  

In other words, it’s easier to interface between technologies in the same ‘domain’ in the sense of  clusters of heuristics arising from the same phenomenon (see [[arthurNatureTechnologyWhat2009]].) Domains tend to emerge around the same time, so in addition to sharing phenomena, they often share assumptions, mental models, and interfaces that make it easier to combine them. ([[Connections create new knowledge]].)

In another evolutionary analogy - it is easier to interface between heuretics that are ‘cousins’ on the tree in the same way that lions and tigers can breed or horses and donkeys can breed. 

[[Leonardo Ideas]] often come from interfacing heuretics that are farther apart on the tech tree. For example, reusable rockets resulting from combining modern control systems, rocket technology, and software engineering which aren’t particularly close. 

### Related
- [[§Tech Tree Model of Heuretics]]
- [[Creating legible interfaces is expensive]]



[Web URL for this note](http://notes.benjaminreinhardt.com/Interfacing_between_closer_neighbors_on_the_tech_tree_is_easier_than_between_nodes_that_are_farther_apart)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Interfacing_between_closer_neighbors_on_the_tech_tree_is_easier_than_between_nodes_that_are_farther_apart)




Long-term projects are inherently highly levered up

# Nodes deeper in the tech tree take more time and work to unlock
[[Each future node in the tech tree model has some function that maps effort in with a success probability distribution over time]]. There is an intuitive sense that discovering nodes that are ‘deeper’ in the tech tree is harder. It is harder to come up with a new explanatory physics theory than it is to create a novel mobile app. 

Why is this? 

Nodes deeper in the tree have less nearby neighbors so there are less easily interfaceable things. ([[Interfacing between closer neighbors on the tech tree is easier than between nodes that are farther apart]].) 

Nodes deeper in the tree are by definition building off of other heretics that have been around longer, so all the ‘low hanging fruit’ around them has been picked.  This explanation is unsatisfactory because it is almost ‘by definition.’

*Hunch: this is the real explantion*
Deeper nodes need to quickly become load-bearing. If it is a discovery-type heuretic it needs to explain all the things that neighboring discovery-type heuretics can and more. If it is an invention-type heuretic, it needs to find a niche where it is more impactful than other inventions with similar characteristics. However, since it is farther down the tree, other ‘families’ of heuristics with similar characteristics have had more time to grow so the chances that it has a higher fitness is low. 

Contrast this to nodes that start off as leaves. Leaf nodes have no expectations so they can more freely see which environment they can interact with to possibly find a large impact. ([[There are many heuretic environments]].) 

### Related
This is one of the assumptions in [[The Management of Innovation]]. The authors don’t even try to justify it. 



[Web URL for this note](http://notes.benjaminreinhardt.com/Nodes_deeper_in_the_tech_tree_take_more_time_and_work_to_unlock)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Nodes_deeper_in_the_tech_tree_take_more_time_and_work_to_unlock)




# People are incentivized to work on nodes higher up the tech tree
 People are incentivized to work on nodes higher in the tree because they have a higher probability of success, that success happens more quickly on average. These two assertions are just the contrapositive* of [[Nodes deeper in the tech tree take more time and work to unlock]]. 

If you needed more incentive to work on leaf nodes than the fact that you are more likely to win and that victory will come relatively quickly, there are two other points:
- [[There is a higher probability of accidentally unlocking a closely related node on the tech tree]]
- [[Interfacing between closer neighbors on the tech tree is easier than between nodes that are farther apart]]

How does this play out in the real world?
- [[Asymmetric career risk]] encourages people to both work on and support low-risk work regardless of outcome and therefore pushes more support to the leaves of the tree. 
- [[Systems must take performance hits to get out of local optima]].  Getting out of local optima requires people to move *away* from leaves and build back up again, so the same forces that incentivize working on higher-up nodes *disincentivize* the moves necessary to get out of a local optima.
- Because [[Graduate students are the labor in academic science and engineering]], the majority of people working in academia want to spend time on problems they can make meaningful movement on in a few years. The need to work on problems with relatively short time-horizons makes leafier nodes more appealing. 
- Almost by definition there is more “low hanging fruit” and the ends of the branches ( [[Fast moving fields have lots of unconnected nodes]] .)  [[Academics are incentivized to do peer-reviewed publishable work]] and a big part of peer-reviewed publishable work is “novelty” ([[The word ‘novel’ is used as an idea bludgeon in academia]].) Novelty is less about importance than newness, so peer-review pushes people towards leaves. Additionally, there are rewards to working in hot, fast moving areas which tend to correlate with unconnected, leafy areas. 

### References
- [[bhattacharyaStagnationScientificIncentives2020]]

[Web URL for this note](http://notes.benjaminreinhardt.com/People_are_incentivized_to_work_on_nodes_higher_up_the_tech_tree)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/People_are_incentivized_to_work_on_nodes_higher_up_the_tech_tree)






Definiteness means that you have a long-term plan

Program design is like creating the lego instruction book

Serious Context of Use

The first step of roadmapping is frontier mapping

Long term plans often involve decisions that look stupid over shorter time horizons

Systems research requires a lot of work that is expensive and uninteresting

Startups are good at point changes

In order to turn space into a true frontier we need many technological revolutions

Systemic decay theory of stagnation

Whence systems research?

Why do we seem to be in a world where humanity is retreating into a world of digital imagination?

Getting to new optima is hard

The structure of the tech tree can inform actions that are short term suboptimal but long term useful

# Systems must take performance hits to get out of local optima
Lock-in at [[Local Optima]]  is often caused by [[Structural Constraints]]. There are many systems where interlocking dependencies mean that you can’t improve the system much by changing one thing at a time (point fixes) but instead you need to ~simultaneously change multiple things, in the same way that you can only pull off certain moves by hitting multiple buttons at the same time in a video game. 

Even if you are able to change multiple pieces of the system simultaneously, it usually results in an initial performance hit. Most people will not accept performance hits for the sake of long term benefits. [[People are more averse to short term losses than they are attracted to long term gain]]. 

The cost in money and time of getting out of a local optimum combined with the inevitable performance hit is a major reason why [[Getting to new optima is hard]]. 

So how do you break this chicken and egg situation? 

1. Heuretics can develop in niches where the performance of the current system is not as good.  A niche can sustain a lower-performing but higher potential technology while it grows into a form that can take on the current paradigm. This is why [[Frontier technologies need to start in niche markets where they are especially valuable]]. This niche-and-expand is how disruptive innovation happens - see the classic examples from [[Clayton Christensen]]: [[PCs vs Mainframes]], [[Minimills vs Steam Mills]]. The niche-and-expand method of development works when there is a fog of war around a heuretic’s potential. [[The phenomenon of seemingly trivial technologies becoming incredibly important happens because there is fog of war on the tech tree]]. 
2. Heuretics can develop while decoupled from the the market system. There are many ways this can happen - slow tinkering in an academic lab, as the secret side project in a company, as a rich person’s company ([[Blue Origin]]), etc. Every successful example of this strategy eventually circles back to something that the market actually likes. This strategy often leads to making something that nobody wants ([[Magic Leap]]) but at the same time can enable things that don’t have an immediate niche. In a way keeping promising technologies alive independent of the market is part of [[Publishing feels more like hardware development than software development]]’s role. [[DARPA provides a derisking role for people in other organizations]] - UAVs and PC technology, for example. If you’re going to decouple from market discipline  you better have a plan. [[Technological roadmapping can give an idea of where technological evolution can go and how to get there]].  [[Decoupling from market discipline is like cave diving]]. 

In reality there is a mix between no.1 and no.2.  Many technologies spend a lot of time in development and *then* move into a niche. [[SpaceX took six years to launch its first rocket]]                                                                                                         

There are tons of examples of this in history:
- [[Tea clippers vs Steamships]]
- [[Cars vs Horses]]
- [[Minimills vs Steam Mills]]
- [[PCs vs Mainframes]]
- Early Heliocentric models of the solar system were actually less predictive than geocentric models

### Related
- [[Optimization Landscape]]
- [[adlerFinancingAdvancedManufacturing2019]]
- [[Dispersion is the process of a heuretic impacting the world]]
- [[alexanderStudiesSlack2020]]
- [[Evolutionarily Stable Strategy]]
- [[Slack - concept]]
- The performance hits necessary to get out of local optima are why [[We don’t have good institutional ways to create multi-point changes]]. 
- [[Startups are good at point changes]]
- [[fillerFundamentalManufacturingProcess2020]]



[Web URL for this note](http://notes.benjaminreinhardt.com/Systems_must_take_performance_hits_to_get_out_of_local_optima)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Systems_must_take_performance_hits_to_get_out_of_local_optima)




# The tech tree model suggests that more work on more root nodes is important
In the [[§Tech Tree Model of Heuretics]] , you get the power law impact behavior described in [[neumannOneProcess2020]] only if every node is equally likely.  However to have a truly random discovery distribution you need *more* effort going into any one node deeper in the tree because [[Nodes deeper in the tech tree take more time and work to unlock]]. [[Each future node in the tech tree model has some function that maps effort in with a success probability distribution over time]]. 

Why is it important to make every node equally likely? More unlikely deeper nodes mean a thinner tail of the impact power law. Progress has been driven by this tail: we point to the same maybe 100 heuretics as what separates us from medieval peasants.   

It might not be important to put in the work to make deeper nodes more likely if it’s fine to leave behind deeper nodes if progress is actually driven by leaves turning out to secretly at the base of a tall branch. ([[The phenomenon of seemingly trivial technologies becoming incredibly important happens because there is fog of war on the tech tree]].)  While trivial nodes can absolutely turn out to be much deeper than we expected, they never harness *wildly new phenomena* (to use the language of [[arthurNatureTechnologyWhat2009]].)  Harnessing new phenomena enables divergent ability ([[Technology gives superpowers]]), which drives progress. ::This last sentence is a bit shaky.::

Taking this one step further, is it true that we could create an even fatter impact tail if we *overloaded* the amount of work happening on the lower down nodes? 

It’s not clear to me how much more effort needs to go into deeper nodes compared to leafier nodes to get their resulting probability distributions to look the same over time - is it 10x? 100x? It also raises the question of what does “effort” actually mean? Is it man-hours? Talent-man-hours? (man hours weighted by how smart the person putting the time in.) Is money a proxy for this? (I suspect not - see the old joke about math departments being cheap because you just need to pay for pencils and paper.) 

The need to have more effort going into root nodes stands in contrast to the fact that [[People are incentivized to work on nodes higher up the tech tree]]. 

### Related 
- [[There are scientific and technological problems that make other technology advances irrelevant by altering the environment]]
- [[How do you incentivize people to work on nodes farther down the tree?]]

*I may be using contrapositive wrong



[Web URL for this note](http://notes.benjaminreinhardt.com/The_tech_tree_model_suggests_that_more_work_on_more_root_nodes_is_important)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/The_tech_tree_model_suggests_that_more_work_on_more_root_nodes_is_important)




Graduate students are the labor in academic science and engineering

Academia is not a good environment for systems engineering

Academia incentivizes novelty, not focus

Academic credit is a first-to-the post system

We have no flying cars - failures of paradigm-shifting engineering

Academia judges via novelty because there are no metrics but there are no metrics when the thing is still research

Peer review is a great system for preventing waste

Legible novelty is the worst

# The word ‘novel’ is used as an idea bludgeon in academia
“Novelty” is one of the constant criteria by which academic research is judged, both in terms of funding and publication. 

Whether something is novel enough not purely about the magnitude of the contribution (the characterization of reactions that barely differ from other reactions can be considered novel) or the actual newness of the idea (If I argued that the transition between turbulent and laminar flow was mediated by conscious 11-dimensional elves it would be a new idea, but not in the good way.) Instead, the worthy novelty of a contribution revolves around its relationship to existing paradigms.  [[Paradigms outline a set of questions and the rules around answering them]].  

A contribution has massive headwinds if the paradigm judges the question being answered to be already answered or out-of-scope or the answer is not in an approved form (the 11-dimensional elves again.) 

It’s easy to say “argh! The system is full of risk aversion, politics, and narrow mindedness!” But [[Most people are not malicious or stupid]] and that goes for academics as well. Most ideas that answer old questions in new ways or have completely different answers are crackpot ideas (11-dimensional ideas!) so it’s not irrational to default dismiss them given limited time and bandwidth[^1]. 

If a question has existed unanswered for a long time, inevitably many people have tried to answer it. Every failed answer rationally increases the priors that the next answer will also fail. Therefore it’s perfectly reasonable prior that a new answer to an old question will also fail. It’s also technically correct that proposing a new answer to an old question is not “novel” creating a rational excuse for people not to waste their time with crackpot answers. However, the novelty criteria become a cudgel when people use it  to avoid spending the time to actually engaging with an idea. The problem is that it’s not actually clear where to set the filter between “take every crackpot idea seriously” and “reject every answer to long-standing problems.” 

The incentives around new answers to old questions also apply to the creation of new questions. [[Academia pushes against the creation of new questions]]. New questions cannot be measured by their performance on standard metrics or data sets. At least in early 21st century machine learning, it’s easier to get a paper accepted by improving performance on a standard dataset from 98.2% to 9.85% than it is to argue that your system is doing something important independent of an established test or *heaven forbid* has performance that can’t be measured numerically. Again, I suspect this situation at least in part arises from the number of people all clamoring for attention and everyone’s limited bandwidth [[Attention is the ultimate scarce resource]]. New questions can also be completely useless. One out of a thousand times “What if we ran it backwards?” Is the key to saving the universe, but most of the time it’s someone trying to be clever with a bad idea.

These two situations parallel startups heavily. It’s rational to have a prior that a company trying ideas that are littered with the bodies of other companies (memexes, new social networks, elder care, etc.) will fail. However, [[It’s rational to try something that failed in the past as long as you explicitly call out why you will succeed when they failed]] but that requires time and precision from both the founders and appraiser. In parallel with new ideas in science, startups trying to introduce a product that “nobody knew they wanted” usually fail as well.  

The upshot is there are good reasons to dismiss new ideas that are both novel and not novel. Therefore “novelty” is a relatively arbitrary idea discriminator and easy for people to use to dismiss ideas they don’t want to engage with. 

### Relevant
- [[§Academia Constraints]]
- [[Mysteries vs puzzles]]
- [[Peer review has failed as an institution]]
- [[Novelty and Heresy - PG]]

[^1]: The coupling between bandwidth and ability to address crackpot ideas on a case-by-case basis of course plays into my bias that insufficient [[Slack - concept]] is a core source of friction in the system. 



[Web URL for this note](http://notes.benjaminreinhardt.com/The_word_‘novel’_is_used_as_an_idea_bludgeon_in_academia)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/The_word_‘novel’_is_used_as_an_idea_bludgeon_in_academia)




Benʼs working notes

People are incentivized to work on nodes higher up the tech tree

References

Links to this note