,
17 tweets,
2 min read
Read on Twitter
what does it mean for something to be “true”? let me try to explain your intuition. (thread)
something is “true” when it is maximally compressed: you cannot make it more predictive without making it significantly longer.
“the sky is blue” is true because you cannot be more predictive (about the sky’s color) with fewer words.
the sky can also be black, gray, red, etc. – but the moderate gains in precision come at the cost of a message many times longer (though still true, of course).
newton’s gravity is true. einstein’s is true too – slightly more predictive, far more complex. qg will be true as well – a tiny sliver more predictive power for an insane amount of added complexity.
the non-obvious conclusion: many *technically contradictory* statements about something can all be true.
so which true statement do you use? depends on how much predictive power you need. engineers already know this.
the poet and the autist value different parts of truth: the poet values its brevity; the autist values its accuracy.
this is why nitpicking aphorisms is stupid – it was never about having 100% accuracy, but instead, useful application at unparalleled brevity.
likewise, no one would say, e.g. a sum-over-histories integral is "too complex". it was discovered optimizing for precision, and succeeded – wildly – on those grounds.
understand truth as a ratio (predictive power / complexity) and many arguments are illuminated: “we need a higher numerator!” vs. “no, we need a lower denominator!”
conservatism is size-optimizing your model of society (“we’re basically just the plurality group”); liberalism is optimizing it for accuracy (cf. 31 legal gender identities in nyc).
any argument about whether an exception should invalidate a model is also of this form. sup thomas kuhn.
smaller models (heuristics, stereotypes) are easier to remember, communicate, and use. but more accurate models yield better results.
in short: when we say “true”, we mean the predictive-power-to-complexity ratio of a statement is a local maximum.
true statements will always have exceptions. that doesn’t make them false; it makes them less precise.
there can be many true statements/models on a topic, so which to use? depends. accuracy is subject to diminishing returns as complexity balloons. choose wisely, kid.
