[scottrogowski]

I think this is a more compelling idea than people in the comments are giving it credit for

> The problem is that you can easily come up with 100 different hypotheses for what’s going on. Ok, so you run 100 different studies to test each one. But studies take a long time to run — let’s say 6 months per study. Congratulations, you’ve just locked yourself into 50 years

This is a major problem with science whenever you have less of a theoretical foundation. Compared to physics or chemistry, we know very little about nutrition or sports science. Because of this, the search space is very large. One could argue that given the number of surprising results (and difficulty reproducing those results), medicine and psychology also fall into this category.

> A riff trial takes advantage of the power of parallel search. Some riffs will work better than others (or at least differently), and parallel search helps you find these differences faster, especially if the differences are big.

What if we did more to encourage people to track and report their personal experiments? If even 10% of everyone on a diet (any diet) just tracked what they ate, what exercise they did, and how much weight they lost, and reported it to a centralized database, scientists could then look for patterns in that data and do formal studies based on suspected patterns.

We could do similar things with longevity/happiness. Look at the "Harvard Study of Adult Development" but imagine it was spread out over 10s of thousands of diverse people instead of just 300 upper-class American men? The data quality wouldn't matter much if all you are doing is searching for patterns to do follow-up studies.

[axiomaticdoubts]

The main issue with this riff trial is that it doesn't test the most likely reason the potato diet works: that it's a very restrictive diet. Testing several different hypotheses barely helps at all if you don't test the overwhelmingly most likely one.