Published on April 10, 2025 2:43 PM GMT
Essentially all solutions in life are conditional: you apply them in the right context, in the right conditions to achieve a good outcome. Obviously banging a hammer on your desk
does probably no good, while banging a hammer on a nail you want to use to hang that nice painting on the wall may be a great idea :)
I want to talk about gradual change, and it doesn't apply all the time. Sometimes if you are certain enough of an outcome and things are urgent gradual change is unnecessary and even harmful.
I think the context in which gradual change is great is when there is significant conflict and controversy. Or when discussions and ideas are turning unstable (violently swinging
between extremes), when there is uncertainty.
I'd say this applies to a lot of current politics and technology (as of 2025). There is talk of radical change, often in opposing directions. This can risk fracturing and destabilizing societies, compromising the basis that has been carefully built upon for decades or centuries.
Here are some closely related arguments for (and against) gradual change:
(1) (For) If a lot of something is good, usually a little of something is good too.
Effects being linear is a pretty reasonable default assumption unless there are reasons for otherwise. And if not linear, usually approximately monotonic holds. Which is to say, usually a little medicine is better than no medicine (in case a large dose is good), a little salt is better than no salt (in the case a lot of salt is good), and so on.
So there's usually no harm, compared to the baseline, of trying a little change. If said change has the desired positive effect, do more of it, repeatedly, until you're doing it enough.
An example of a non-monotonic system to change might be a program. People don't change programs by changing single characters and recompiling. You need a minimum viable change. A too small change (say if you were restricted to only be able to change a few characters of a source code each deployment) is probably going to affect the system negatively (making code more difficult to understand and not fully able to enact a desired change). But usually fairly small (a few lines) changes, which are then tested and possibly deployed, are good from an engineering and reliability standpoint.
(2) (For) The risk of a little of something is much smaller
By a similar argument, if that thing is bad, usually a little of that is much less bad.
This allows de-risking the point under contention. Is <change> good? Try a little bit of
it, and let's sincerely analyse the results. Then go for the best for everyone.
(3) (Against) Too little of something is hard to evaluate
The only way to make sense out of change is to plunge into it, move with it, and join the dance.
-- Alan W. Watts
Most changes in real life will have all sorts of random effects, side effects and other unrelated changes going on all the time. Those effects can be modeled as noise: although something on average may yield a good result, it may randomly vary because of all other things going on in the world. A good change may yield bad results on the short term because of other unrelated happenings, and vice versa a bad change may yield good results.
So we should try changes at least for (a) a long enough time/ repeat the experiment enough, and/or (b) with enough intensity, so that we will likely be able to conclusively evaluate the change.
(4) (For) Too much of almost anything can be bad
This one is a well known wisdom. Even drinking too much water, which is thought as an essential element and almost universal good, can be not just harmful but deadly. The dose for that to happen is very high (wikipedia reports drinking 6 liters in 3 hours has caused death).
This is less an argument for gradualism, but an argument against extremism. It's unlikely going extremely in one direction will yield good results, at least in usual cases (that may not be true in a specific case of course!).
(5) (For) Almost every significant change can be made gradually
This one can be considered obvious or extremely counter-intuitive. Solutions often seem to come to us (or at least to me, in some cases) like a flash. All at once. Revolutions appear to happen all at once. Systems go online. On and off, and so on.
But indeed reality itself appears to be continuous. So in a certain (physical) sense every change (with enormous variation in speed of course) happens continuously, going from one state to the other in a smooth fashion with smaller change for smaller time intervals under consideration.
Of course the great example nowadays is in Large Neural Networks. We are training those models by variations of gradient descent, which means they make tiny changes toward a little improvement. Over time those accumulate, and the network acquires extremely complex behaviors over trying small changes, re-evaluating, changing again, re-evaluating, and so on. Most optimization problems (which is essentially what we're discussing, even if the subject is the satisfaction of society) can be solved gradually (if only by invoking the trick of making a discrete A/B variable probabilistic, having P(A)=1-P(B)).
(6) (Against) Gradual change leaves benefits at the table
We want change, in the end (at least we should), because we're looking for a better society. If we're too slow, we're lingering in if not categorically "bad" at least non-ideal ways. We're arguably causing harm by failing to do good, in some sense.
That's another argument against gradualism: we don't want changes so slow that humanity goes extinct before anyone can reap the benefits of a better society. But we also don't want things too fast they can risk blowing everything up.
I am not sure what is the correct timeframe to consider positive changes against. Is it that bad that we waited 10 years? 100? 10,000 years? For long enough timeframes, uncertainty starts to creep in and benefits become more questionable. We could be hit by an asteroid, annihilate ourselves in silly wars, die from pandemics, and so on. Also, I believe there are metaphysical reasons to consider early lives more likely in some sense.
This is a difficult calculation I won't give any exact time frame in which future benefits should be discounted (or the exact discount function, which is usually exponential for more or less plausible reasons). I think anywhere between 100 years and 10,000 years (but probably not much more) is plausible for a discount taking most of the aforementioned factors into account.
(7) (For) Longer time horizons than intuitive
Continuing the last item, still 100-10,000 years is much longer timeframe than most people think about! Most people want the positive change to happen in the next 5-10 years and ideal states be reached within maybe 20.
We tend not to have much intuition for several decades, centuries or even millennia. If this is true, we're probably changing a lot of things too fast. We should probably both be a little more experimental (both scientifically and culturally), and also more careful with large scale changes that risk destabilizing society altogether.
(8) (For/Against) Exponentials and plateaus
As a counterargument to this last point, one can point out that gains usually compound.
So if you get 10% better than otherwise you would (by going faster), per year, than
over 20 years you get not 2x better, but 6.7x better, and over 100 years you'd get 13780x
better than the slower scenario.
But this only holds for certain exponential factors under consideration, like technology, maybe some measures like productivity. The exponential factors may include the number of people we can provide good conditions with (can increase very rapidly).
There's however seemingly no direct correspondence to social, political systems and institutions or cultural artifacts. Although technology probably can be said to spillover to produce better culture (more interesting and varied music, more access to culture, new forms of culture like games and some sports), there are some limitations. In other words, we probably won't (and simply can't) be 100000x more happy, more joyous, etc.. To me improving our inner lives is the goal of society.
Moreover, probably most changes will only stay exponential for a while. Even for core technological changes, like computer technology, are stalling, and probably can't continue for very long. Then as potential improvements slow, it makes less sense to take a radical attitude to reap a marginal improvement.
Also, since slowdown is very likely, the total effect of a slower transition tends to be simply a delay of getting to the plateau (which of course can be significant!).
(9) (Against graduality) Rapid change is exciting
I get it, rapid change is fun. But, ...
(10) (For graduality) Many things are exciting
There are so many things to be excited about. Your hobby. Your family. The smile of your co-worker or your loved one. We probably don't need a lot of drama happening in politics for our lives to be meaningful, specially not when that carries a lot of risk for all of us.
A reasonable heuristic?
Again it really depends on the particular situation, but before making a radical change,
maybe consider going 10-20% of the way instead, or some fraction of the way that is low
risk and still informative. If you repeat the same magnitude only 5-10x you'd get the
radical effect you wanted.
Conclusion
As important as any of the previous observations, is that we do the basic job: that we actually put effort and attention into seeing what is working and what isn't. That we're careful when comparing things. If we don't evaluate and plan changes properly we won't be able to make meaningful progress on all the many things we likely can improve in the world. It's important we build enough consensus around changes and address reasonable suggestions and objections.
I hope this provided some good ways to view speed of change and radicalism, and maybe apply some slowdown (or speedup) when necessary.
"Chill, dudes!"
Thanks for reading.
Note: Article still being revised. Suggestions welcome.
Discuss
