# Good science fiction makes second-order predictions and includes nonlinearities Every piece of science fiction has at least one ‘conceit’ - a difference between its world and our world you need to assume is true. In [[SevenEves]] the big conceit (spoiler from first three pages) is that The Moon shatters. These conceits can be more or less believable or surprising but if there is no conceit it is just boring. On the other hand, if there are too many conceits (especially about physics) then it’s just fantasy with space lasers. The tension created by the fact that [[Good science fiction is self-consistent]] keeps conceit explosions in check. Conceits in science fiction are non-linearities. In this context a non-linearity means that something changes in a way that is different than the way things have been changing in the past.[^1] Nonlinearities are non-obvious and part of what makes science fiction a good prediction tool. [[A good prediction enables new thoughts and actions by being non-obvious and creating agency]]. Extrapolating an exponential trend is not a nonlinearity, even though an exponential curve itself is nonlinear. Sometimes non-linearities are [[Unknown Unknowns]], literally just coming out of nowhere but they can also result from complex interactions between different trends. [[Sci-fi conceits are Kuhnsian paradigm shifts]] I’m not going to rigorously define second-order predictions, but it’s some intellectual admixture of second order [[Taylor expansions]] and graphs with a maximum degree of at least two. A second order Taylor expansion extrapolates not just from the current rate of change, but captures the change in change. *waves hands* A prediction with a greater than two maximum degree means that it digs into how the world might be affected by two different interacting trends. The combination of second order predictions and nonlinearities makes good science fiction complex enough that it can’t explore the entire self-consistent world it creates. These gaps give you space to fill with your imagination. [[Self-consistent worlds enable you to draw off of the edge of the map]]. [^1]: More rigorously - there is no variable that you could linearize around ([[Taylor expansions]]) that would give a vaguely accurate approximation of the real function. <!-- #evergreen #needs/examples --> [Web URL for this note](http://notes.benjaminreinhardt.com/Good+science+fiction+makes+second-order+predictions+and+includes+nonlinearities) [Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Good+science+fiction+makes+second-order+predictions+and+includes+nonlinearities)