I learnt about wait calculations today. Or rather, I learnt that’s what they’re called. I have been doing them my entire life, but we’ll come back to that later.
What is a wait calculation?
This paper describes an incentive trap of growth that shows that civilisations may delay interstellar exploration as long as voyagers have the reasonable expectation that whenever they set out growth will continue to progress and find quicker means of travel, overtaking them to reach and colonise the destination before they do. This paper analyses the voyagers’ wait
https://gwern.net/doc/statistics/decision/2006-kennedy.pdf
calculation, using the example of a trip to Barnard’s Star, and finds a surprising minimum to time to destination at a given rate of growth that affects the expansion of all civilisations. Using simple equations of growth, it can be shown that there is a time where the negative incentive to travel turns positive and where departures will beat departures made at all other times. Waiting for fear future technology will make a journey redundant is irrational since it can be shown that if growth rates alter then leaving earlier may be a better option. It considers that while growth is resilient and may follow surprising avenues, a future discovery producing a quantum leap in travel technology that justifies waiting is unlikely
If you wanted to travel to the star nearest to us, should you leave today or not? You might think the correct answer is obviously yes, you should leave today. But ask yourself this (ridiculous but illuminating) question: should you have decided to start swim to the United States from India the month before we invented ships capable of such crossings?
Makes rather more sense to just wait for a month and jump onto the ship, no?
Ethan Mollick asks if we should write a book / design a presentation / discover a new element / <insert task that can be done by AI here>, or wait until AI becomes good enough to do this task for us instead.
Which brings us to AI. AI has all the characteristics of a technology that requires a Wait Calculation. It is growing in capabilities at a better-than-exponential pace (though the pace of AI remains hard to measure), and it is capable of doing, or helping with, a wide variety of tasks. If you are planning on writing a novel, or building an AI software solution at your business, or writing an academic paper, or launching a startup should you just… wait?
https://www.oneusefulthing.org/p/the-lazy-tyranny-of-the-wait-calculation
He gives two examples from his own professional life where he thinks he should have waited, because what took a lot of blood sweat and tears (or significant effort, at any rate) took AI not all that much time, not really.
And that, of course, is true for a lot of us, across a lot of tasks that we do in our daily life (present activities included for both you and I, by the way). AI can, and does, do the task of reading and writing blog posts, so why should I bother writing this, and why should you bother reading it? We could have waited! Or rather, we could have asked AI to do these tasks for us.
So which tasks should we do, and which tasks should we wait upon, or delegate? Ethan Mollick says that the answer to this question in turn depends upon two other questions: “how good?”, and “how fast?”. How good is (or will be) the AI, and how long will it take for the AI to become that good?
The troublesome bit is that we just don’t know the answers to these questions, because of how rapidly AI is developing. Ethan Mollick develops a better, richer framework in his post, and as always, you should read the whole thing.
But of all of his excellent questions in his framework, my favorite one was this one:
Does it create a learning trap?
That is, choosing to let AI do something for you robs you of the opportunity to learn how to do it yourself. And in the world of learning (which is the corner of the internet where this blog locates itself), why would you want to give up on the opportunity to learn?
So if it is your 5000th presentation, or your 20th book, or your 400th academic paper, well ok, you may want to let AI write it for you. But if it is your among your first efforts in the field, maybe give it a shot yourself?
As with everything else in life, there’s lots of asterisks and conditions and what have you’s. But in my limited experience of having tried to get AI to do things, having tried it yourself first is the best way to write better prompts. Skin something something game and all that.
So yes, absolutely, waiting probably makes sense in some cases. And as the lifetime president of the Procrastination Society, I don’t have much moral standing to say what I’m about to – but the best way to learn is to try and do it yourself first!
EconForEverybody 