🎓 a16z Crypto School – #7 Incentive Engineering

Pretty high level, nothing ground breaking… But worth it just to learn about the Duplessis Orphans as an example of mechanism design gone wrong, and incentives trumping morality…

(Youtube video link)

Speaker

  • Sam Williams is the founder of the Arweave blockchain, which is a pretty awesome project. The goal of this chain is to incentivise its actors to store data forever.
  • Sam speaks eloquently about the mission to build a “digital library of Alexandria”, and the importance of being able to save data, eg to notice bad actors rewrite past events, as has happened throughout history. They are working with the Internet Archive – good interview with them here.
  • They have been gaining traction this year to store the files that NFTs point to (an NFT on Ethereum is just a token ID and a pointer to a web address, which now is often an Arweave address).
  • We are happily using it for the GitNFT project
  • Disclosure, I own some $AR, the utility token of the Arweave blockchain

What is Mechanism Design?

  • “programming human behaviour through the careful design of incentives”
  • Incentives are produced by economic games that can be crafted and engineered
  • Love it or hate it money is the unifying sub-goal for almost all larger goals in life
  • Examples of mechanism designs
    • Picking up empty bottles, in Berlin you get 30c per bottle return. Gives people who don’t have it jobs, cleaner streets 
    • Scooter cos incentivising people to charge and distribute scooters around the city 
  • Many of the mechanism designs push to society the question of how to play the game that you’ve created most effectively. Decentralised by nature.
  • Example: Bittorrent uses optimistic tit for tat
    • “I will give you data if you will give me data” (I only seed with you if you seed)
    • This works… at one point it was 1/3 to 1/2 of all the internet traffic
    • Requires 0 global sync. Only need to have the clients remember who they are ok to seed to

Pitfalls

  • Incentives are often more powerful than most moral frameworks of players in economic games
  • Maurice Duplessis, Quebec governor in province in Canada. Noticed that federal government gave him $2/day for keeping children homes, and $7/day for keeping someone in psychiatric ward
    • Caused 20k children to be misdiagnosed to mental health issues that they didn’t have 
    • They were forced to take meds… look up the “Duplessis orphans”…
  • At least in Canada case they could notice the problem and change it. In decentralised network need enough miners to agree, and they may not 
  • eg Block size debate. Miners actually didn’t have an incentive to increase the 1MB block limit, they were getting well compensated by tx fees people would have to pay for their tx to go through. Fork, fragmented the community
  • eg PoW. Environmental cost. We could have picked another consensus model that produced useful work, for instance protein folding.
  • eg AI safety. How do you make sure you can turn it off? Last thing you would do is put it on a decentralised network.

Mechanism Design Examples 

  • Bitcoin
    • Goal: maximise network security to avoid double spends
    • Reward mechanism: tokens distributed relative to contribution of each miner
    • Reward function: proof of work, block production acceptance, fork avoidance
      • Incentive to use the longest chain 
  • Arweave
    • Goal: securely store data permanently
    • Reward mechanism: users contribute funds to an endowment when they add data, miners are rewarded from the endowment over time
      • You pay for 200 years worth of storage upfront. < 0,01$ / MB
      • Over the last 50 years cost of storage has declined by 30.5% per year on average
    • Reward function: proof of access. check the data in the network at random
      • so there is no better way to play the game than replicating as much of the data as you can

(I took a while to wrap my head around how Arweave’s incentive structure works. Basically what happens is you grossly overpay for the storage of your data upfront, in comparison to what you would have to pay if you were storing it at the time when the miners have to prove they own it, which will be way in the future. Sorry if I confused you further)

Misc

  • “Wisdom of the crowd could get centralised again with AI”
  • Not all mechanisms are designed. Some (most?) are emergent, eg panic buying. I need to front-run your panic buying
  • Network builds itself on people verifying the permaweb source code. 

Trent McConaghy

Recommended article from Sam, by Trent McConaghy at Ocean Protocol.

  • Token engineering could become a trade, like mechanical, electrical, software engineering.
  • Should include ethics education. Was fun to see The Tacoma Narrows Bridge which is basically the only thing I remember from engineering ethics course
  • “How often does an academic economist (or anyone, really) get a chance to deploy an economy? Such an opportunity happens each time a cryptonetwork is launched, as it spawns a new economy with a native unit of account and supply schedule to incentivize unique behavior— a responsibility we can’t take lightly.”
  • Token design is especially like evolutionary algorithms (EAs), where there are many agents “searching” at once and there is no top-down control of what each agent does. Agents live and die by their block reward or fitness.

Good memories of the Introduction to Complexity course – must write notes!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s