Why I used to prefer permissive licenses and now favor copyleft
2025 Jul 07
See all posts
Why I used to prefer permissive licenses and now favor copyleft
Within free
open source software (and free content more
generally), there are two major categories of copyright licenses:
- If content is published with a permissive license
(eg. CC0,
MIT), anyone can take
it and use it and redistribute it for any purpose with no restrictions,
perhaps with minimal rules requiring attribution.
- If content is published with a copyleft license
(eg. CC-BY-SA,
GPL), anyone
can take it and use it and redistribute copies with no restrictions, but
if you create and distribute a derivative work by modifying it
or combining it with other work, the new work must also be released
under the same license. Additionally, GPL requires any derivative work
to openly publish its source code, in addition to a
few other requirements.
In summary: permissive licenses freely share with everyone,
copyleft licenses freely share only with those who are also willing to
freely share.
I have been a fan of, and developer of, free open source software and
free content ever since I've been old enough to understand what these
things are and build things that I thought other people might find
useful. Historically, I was a fan of the permissive approach (eg. my
blog is under the WTFPL). More
recently, I am warming up to the copyleft approach. This post explains
my reasons why.
One style of software freedom, promoted by the WTFPL. But not the only style.
Why I was
historically a fan of permissive licenses
First, I wanted to maximize adoption and
distribution of my work, and releasing it under permissive
licenses facilitates that, by making it clear that there is
nothing anyone needs to worry about if they want to build off
of something I make. Enterprises are often unwilling to release their
projects freely, and given that I did not see myself having any ability
to nudge them to fully join the free software side, I wanted to avoid
being needlessly incompatible with the approach they already had and
would not give up.
Second, I generally philosophically dislike
copyright (and patents). I dislike the idea that two people
privately sharing bits of data between each other can be perceived as
committing a crime against a third party whom they are not touching or
even communicating with and are not taking anything away from (no, "not
paying" is NOT the same as "stealing"). Explicitly releasing to public
domain is legally
complicated for various reasons, and so a permissive license is the
cleanest and safest way to get as close as possible to not copyrighting
your works.
I do appreciate the copyleft idea of "using copyright against
itself" - it's a beautiful legal hack. In some ways it's
similar what I always found philosophically beautiful about
libertarianism. As a political philosophy, it's often described as
banishing the use of violent force except for one application: to
protect people from other violent force. As a social philosophy, I
sometimes see it as a way of taming the harmful effects of the human
disgust reflex by making freedom itself a sacred
thing that we find it disgusting to defile: even if you think two other
people having an unusual consensual sexual relationship is disgusting,
you can't go after them, because interfering in the private lives of
free human beings is itself disgusting. So in principle, there are
historical precedents to show that disliking copyright is compatible
with using copyright against itself.
However, while copyleft of written work fits into this
definition, GPL-style copyright of code oversteps
beyond a minimalistic notion of "using copyright against itself",
because it offensively uses copyright for a different purpose: mandating
publication of source code. This is a public-spirited purpose,
and not a selfish purpose of collecting licensing fees, but it is
nevertheless an offensive use of copyright. This becomes even more true
for stricter licenses like the AGPL, which
require publication of source code of derivative works even if you never
publish them and only make them available via software-as-a-service.
Different types of software licenses, with different sets
of conditions under which someone making a derivative work is required
to share the source code. Some of them require publishing source code
under a wide variety of situations.
Why I am warmer to copyleft
today
My switch from favoring permissive to favoring copyleft is motivated
by two world events and one philosophical shift.
First, open source has become mainstream, and nudging
enterprises toward it is much more practical. Plenty of
companies in all kinds of industries are embracing open source.
Companies like Google, Microsoft and Huawei are embracing
open source, and even building major software packages open. New
industries, including AI and of course crypto, are heavier on open
source than previous industries ever were.
Second, the crypto space in particular has become more
competitive and mercenary, and we are less able than before to
count on people open-sourcing their work purely out of niceness. Hence,
the argument for open source cannot just rely on "please"; it must also
be accompanied by the "hard power" of giving access to some code only to
those who open up theirs.
One way to visualize how both pressures increase the relative value
of copyleft is a graph like this:
Incentivizing open source is most valuable in situations
where it's neither unrealistic, nor guaranteed. Today, both mainstream
enterprise and crypto are in that situation. This makes the value of
incentivizing open source via copyleft high.
Third, Glen Weyl-style economic arguments have
convinced me that, in the presence of superlinear returns to scale,
the optimal policy is actually NOT Rothbard/Mises-style strict property
rights. Rather, the optimal policy does involve some nonzero amount of
more actively pushing projects to be more open than they otherwise would
be.
Fundamentally, if you assume economies of scale, then by simple
mathematical reasoning, nonzero openness is the only way that the world
does not eventually converge to one actor controlling everything.
Economies of scale means that if I have 2x the resources that you do, I
will be able to make more than 2x the progress. Hence, next year, I will
have eg. 2.02x the resources that you do. Hence...
Left: proportional growth. Small differences at the start
become small differences at the end. Right: growth with economies of
scale. Small differences at the start become very large differences over
time.
A key pressure that has prevented this dynamic from getting
out of hand historically is the fact that we are not able to opt out of
diffusion of progress. People move between companies and
between countries and take their ideas and talents with them. Poorer
countries are able to trade with richer countries and get catch-up
growth. Industrial espionage happens everywhere. Innovations get
reverse-engineered.
More recently, however, several trends threaten this balance, and at
the same time threaten other factors that have kept unbalanced growth in
check:
- Rapid technological progress, which allows
super-exponential curves to be much faster than before.
- Greater political instability both within and
between countries. If you are confident that your rights will be
protected, then someone else getting stronger without touching you does
not hurt you. But in a world where coercion is more feasible and
unpredictable, someone becoming overly powerful compared to others is
more of a risk. At the same time, within countries, governments are less
willing to restrain monopolies than before.
- The modern ability to make proprietary software and hardware
products that distribute ability to use without diffusing ability to
modify and control. Historically, giving a product to a
consumer (whether within a country or between countries) inevitably
implied opening it up to inspection and reverse-engineering. Today, this
is no longer the case.
- Limits
to economies of scale, historically a key limiter
of runaway growth, are weakening. Historically, larger entities
have had disproportionately higher monitoring costs, and difficulty
satisfying local needs. More recently, digital technology again makes
much larger-scale structures of control and monitoring possible.
This all increases the possibility of persistent, and even
self-reinforcing and growing, power imbalances between companies and
between countries.
For this reason, I am increasingly okay with stronger efforts to make
diffusion of progress something that is more actively incentivized or
mandatory.
Some recent policies made by governments can be interpreted as being
about attempting to actively mandate higher levels of diffusion:
- EU standardization mandates (eg. most
recently USB-C), which make it harder for build proprietary
ecosystems that do not play nicely with other technology
- Forced
technology transfer rules in China
- USA banning
non-compete agreements, which I support on the grounds that they
force the "tacit knowledge" inside of companies to be partially open
source, so once an employee leaves one company they can apply skills
learned there to benefit others. Non-disclosure agreements limit this,
but are fortunately very porous in practice.
In my view, the downsides of policies like these tend to come from
their nature of being coercive policies of a government, which leads to
them preferentially incentivizing types of diffusion that are heavily
tilted toward local political and business interests. But the upside of
policies like this is that they, well, incentivize higher levels of
diffusion.
Copyleft creates a large pool of code (or other creative products)
that you can only legally use if you are willing to share the source
code of anything you build on it. Hence, copyleft can be viewed
as a very broad-based and neutral way of incentivizing more diffusion,
getting the benefits of policies like the above without many of their
downsides. This is because copyleft does not favor specific
actors and does not create roles for active parameter setting by central
planners.
These arguments are not absolute; in some cases, maximizing the
chance that something gets adopted by truly everyone is worth licensing
it permissively. However, on the whole, the benefits of copyleft are
much greater today than they were 15 years ago, and projects that would
have gone permissive 15 years ago should at least think about adopting
copyleft today.
Today, this sign unfortunately means something
totally unrelated. But in the future, maybe we can have open-source
cars. And perhaps copyleft hardware can help make that happen.
Why I used to prefer permissive licenses and now favor copyleft
2025 Jul 07 See all postsWithin free open source software (and free content more generally), there are two major categories of copyright licenses:
In summary: permissive licenses freely share with everyone, copyleft licenses freely share only with those who are also willing to freely share.
I have been a fan of, and developer of, free open source software and free content ever since I've been old enough to understand what these things are and build things that I thought other people might find useful. Historically, I was a fan of the permissive approach (eg. my blog is under the WTFPL). More recently, I am warming up to the copyleft approach. This post explains my reasons why.
One style of software freedom, promoted by the WTFPL. But not the only style.
Why I was historically a fan of permissive licenses
First, I wanted to maximize adoption and distribution of my work, and releasing it under permissive licenses facilitates that, by making it clear that there is nothing anyone needs to worry about if they want to build off of something I make. Enterprises are often unwilling to release their projects freely, and given that I did not see myself having any ability to nudge them to fully join the free software side, I wanted to avoid being needlessly incompatible with the approach they already had and would not give up.
Second, I generally philosophically dislike copyright (and patents). I dislike the idea that two people privately sharing bits of data between each other can be perceived as committing a crime against a third party whom they are not touching or even communicating with and are not taking anything away from (no, "not paying" is NOT the same as "stealing"). Explicitly releasing to public domain is legally complicated for various reasons, and so a permissive license is the cleanest and safest way to get as close as possible to not copyrighting your works.
I do appreciate the copyleft idea of "using copyright against itself" - it's a beautiful legal hack. In some ways it's similar what I always found philosophically beautiful about libertarianism. As a political philosophy, it's often described as banishing the use of violent force except for one application: to protect people from other violent force. As a social philosophy, I sometimes see it as a way of taming the harmful effects of the human disgust reflex by making freedom itself a sacred thing that we find it disgusting to defile: even if you think two other people having an unusual consensual sexual relationship is disgusting, you can't go after them, because interfering in the private lives of free human beings is itself disgusting. So in principle, there are historical precedents to show that disliking copyright is compatible with using copyright against itself.
However, while copyleft of written work fits into this definition, GPL-style copyright of code oversteps beyond a minimalistic notion of "using copyright against itself", because it offensively uses copyright for a different purpose: mandating publication of source code. This is a public-spirited purpose, and not a selfish purpose of collecting licensing fees, but it is nevertheless an offensive use of copyright. This becomes even more true for stricter licenses like the AGPL, which require publication of source code of derivative works even if you never publish them and only make them available via software-as-a-service.
Different types of software licenses, with different sets of conditions under which someone making a derivative work is required to share the source code. Some of them require publishing source code under a wide variety of situations.
Why I am warmer to copyleft today
My switch from favoring permissive to favoring copyleft is motivated by two world events and one philosophical shift.
First, open source has become mainstream, and nudging enterprises toward it is much more practical. Plenty of companies in all kinds of industries are embracing open source. Companies like Google, Microsoft and Huawei are embracing open source, and even building major software packages open. New industries, including AI and of course crypto, are heavier on open source than previous industries ever were.
Second, the crypto space in particular has become more competitive and mercenary, and we are less able than before to count on people open-sourcing their work purely out of niceness. Hence, the argument for open source cannot just rely on "please"; it must also be accompanied by the "hard power" of giving access to some code only to those who open up theirs.
One way to visualize how both pressures increase the relative value of copyleft is a graph like this:
Incentivizing open source is most valuable in situations where it's neither unrealistic, nor guaranteed. Today, both mainstream enterprise and crypto are in that situation. This makes the value of incentivizing open source via copyleft high.
Third, Glen Weyl-style economic arguments have convinced me that, in the presence of superlinear returns to scale, the optimal policy is actually NOT Rothbard/Mises-style strict property rights. Rather, the optimal policy does involve some nonzero amount of more actively pushing projects to be more open than they otherwise would be.
Fundamentally, if you assume economies of scale, then by simple mathematical reasoning, nonzero openness is the only way that the world does not eventually converge to one actor controlling everything. Economies of scale means that if I have 2x the resources that you do, I will be able to make more than 2x the progress. Hence, next year, I will have eg. 2.02x the resources that you do. Hence...
Left: proportional growth. Small differences at the start become small differences at the end. Right: growth with economies of scale. Small differences at the start become very large differences over time.
A key pressure that has prevented this dynamic from getting out of hand historically is the fact that we are not able to opt out of diffusion of progress. People move between companies and between countries and take their ideas and talents with them. Poorer countries are able to trade with richer countries and get catch-up growth. Industrial espionage happens everywhere. Innovations get reverse-engineered.
More recently, however, several trends threaten this balance, and at the same time threaten other factors that have kept unbalanced growth in check:
This all increases the possibility of persistent, and even self-reinforcing and growing, power imbalances between companies and between countries.
For this reason, I am increasingly okay with stronger efforts to make diffusion of progress something that is more actively incentivized or mandatory.
Some recent policies made by governments can be interpreted as being about attempting to actively mandate higher levels of diffusion:
In my view, the downsides of policies like these tend to come from their nature of being coercive policies of a government, which leads to them preferentially incentivizing types of diffusion that are heavily tilted toward local political and business interests. But the upside of policies like this is that they, well, incentivize higher levels of diffusion.
Copyleft creates a large pool of code (or other creative products) that you can only legally use if you are willing to share the source code of anything you build on it. Hence, copyleft can be viewed as a very broad-based and neutral way of incentivizing more diffusion, getting the benefits of policies like the above without many of their downsides. This is because copyleft does not favor specific actors and does not create roles for active parameter setting by central planners.
These arguments are not absolute; in some cases, maximizing the chance that something gets adopted by truly everyone is worth licensing it permissively. However, on the whole, the benefits of copyleft are much greater today than they were 15 years ago, and projects that would have gone permissive 15 years ago should at least think about adopting copyleft today.
Today, this sign unfortunately means something totally unrelated. But in the future, maybe we can have open-source cars. And perhaps copyleft hardware can help make that happen.