Micromorts! Children get sick! Deep learning made waves!
A very quick look at the effect of deep learning since 2012, some plots on how often children get sick (and whether they acquire meaningful immunity from being pre-exposed to disease), and a note about measuring and comparing risks.
Sorry about the clickbait. Today's issue is accidentally themed "bad things children expose themselves to".
New articles
What Does an AI Revolution Look Like?
You don't need to click this link. It's just a bunch of plots showing how deep learning was successfully tried once in 2012 and then immediately took over in 2013 and has been used ever since.
Full article (0–2 minute read): What Does an AI Revolution Look Like?
How Often Does a Child Get Sick?
Our first child was born a few weeks before the pandemic came to our corner of the world, so for the first year of his life, he was not sick even once. Then he started daycare, and you can guess what happened. The second child did not suffer from early life lockdowns – can you guess whether they have been sick more or less after starting daycare?
Full article (1–4 minute read): How Often Does a Child Get Sick?
Flashcard of the week
I realised during the summer that with children, one sometimes has to quite explicitly trade time or money to reduce risk of death, but I also realised that I don't have a good mental language to use when reasoning about the economics of these tradeoffs.[1] I spent a few minutes flashcarding various risk levels in micromorts. One of the questions is
How many micromorts is a day in the life of a 90-year old?
In case it's not clear, a micromort is a millionth chance of death. They aren't really additive, but as long as we are working with small numbers, we can pretend they are and get a fair approximation. The answer is:
500 micromorts
We can arrive at this by working backwards from an actuarial table. Of the men alive at 90, only 84.3 % are still alive at 91. The 365.2425th root of 84.3 % (the daily probability of survival) is something like 99.953 %, which is 467 micromorts risk of dying.
If we had instead pretended micromorts were additive and divided the 16 % chance of dying by 365, we would have gotten 400 micromorts – close enough. (Especially considering women at 90 are exposed to a daily risk of 403 micromorts. But some numbers are easier to remember than others so my flashcard says 500.)
The reason this is useful to know is that if we find an activity (say BASE jumping) with a risk level of 500 micromorts, we can imagine that as being instantly transformed to a 90-year old for a day and then back again, to calibrate our sense of the risk. I.e. non-trivial, but not catastrophic.
[1]: I hear someone in the back shouting "you cannot use economics to reason about the risk of death of your children!" It's as distasteful to me as to you, but what other framework should replace it? I want to spend my time and money in a way that maximises their safety, and economics is the study of using existing resources to maximise the desirable.
Your opinions
As always, I cannot improve without feedback. Reply to this email to share your thoughts on any of the topics above, or anything else!