Early 2021 proof of memory experiment on 1729

In April 2021, I launched a low-stakes “proof of memory” experiment with Balaji Srinivasan, on his bounty-centric blog, 1729, based on the ideas described in The mnemonic medium can surface “proof of memory” social signals. We offered $10 each to 100 people who read QCVC and collected all the prompts. Then we gave $100 to 10 people who actually kept up with the reviews for a month (defined as completing each review session within 72 hours after notification for a month).

There are lots of problems with a structure like this. Does this incentivize people to read, but not to pay attention? I think this approach certainly causes some people to read the material who aren’t especially interested in it, and there’s a moral hazard in that. But I was curious what would happen, and it was a simple thing to try, so here we are.

338 people submitted their Quantum Country account information (with some dupes—==TODO==). Of these, 157 had “legitimately” completed QCVC. The rest either didn’t collect all the cards, or had timing / accuracy data which indicated that they probably cheated. I was surprised that so many people tried to cheat. I guess it doesn’t cost them anything to try. I noticed that the initial glut of entrants were much more likely to cheat than later entrants.

But this experience suggests that any systems which want to experiment with “proof of memory” will probably need to design their mechanics carefully to tolerate a high rate of attempted “cheating.”

I was surprised to see that 58 of these entrants (or 37% of the valid entrants) continued to study for a month. That’s a higher retention rate than I’d expect, particularly since I’d imagined many (most?) of the readers wouldn’t really care about the material.

Next, I’d like to email some of the entrants to learn more about their motivations and experiences.

Last updated 2023-07-13.