- Suppose you had a black box that could attribute a new idea to a bundle of old ideas, and identify the novel contribution.
- Can we use this to set up a system where authors get paid based on anything that ever depends on their novel contribution?
- I'm an author. I've taken some other people's ideas, mixed them together, and come up with something new.
- I announce to the world that I have had a new idea. I can't publish the idea openly, because then no-one'd pay me for it! Instead I publish some tags describing the idea so that potential purchasers can come talk to me.
- A potential purchaser shows up. They send me a gogol* of themselves, a little AI approximation of them.
- I put the purchaser's gogol in a sandbox with my idea. It evaluates the idea and proposes a price. The sandbox is then destroyed.
- If I'm happy with the price, I send the purchaser the idea and they send me the cash. This transaction is public.
- Now a colleague shows up and says she thinks the idea I just sold might have depended on an idea I got from her a while back, and she deserves some of the cash from the transaction.
- So we put her idea and my idea in a sandbox, along with a trusted third-party gogol. The third party gogol compares my idea to the colleagues and says that my idea is 50% hers.
- So I pay 50% of the revenue from my sale to her.
- Now another colleague shows up and says his idea is also similar to mine. But I suspect his idea is also a derivative of the one I just paid for, and I don't want to hand out the same fraction of my revenue twice!
- So this time into the box goes my idea, my second colleague's idea, my first colleague's idea.
- The third-party gogol figures out the Venn diagram between the three ideas, and says while 60% of my idea is the same as his, 50% of that commonality is derivative.
- So I pay out 10% of the revenue to my second colleague.
- But what about my two colleagues - which of them is due the 50%?
- Let's reward ideas based on precedence. Whoever historically published a hash of their idea first is the one that gets paid.
- This rewards participation in the system, and encourages people to find novel ideas.
- In this case, my first colleague published a hash before my second colleague. So my second colleague has no claim to the 50% of derivative commonality.
- In fact, my first colleague - now aware that my second colleague has been trying to get paid for something that's largely her idea - can go through my second colleague's public record and see if there's anything else she should have been paid for.
- hoo boy there are a lot of issues with this, but the first big ones to mind are
- Gogols leak information. Fix is either something differential privacy-y, or sending the gogols to a trusted third party.
- I've said 'trusted third party' several times here. If you want a trusted third party database and compute infrastructure in 2021, b-l-o-c-k-c-h-a-i-n. But I think that's orthogonal to the other ideas here.
- The precedence rule means whoever adds all historical ideas to the system is due a lot of cash. That's good incentives for early adopters, but not aligned with the social welfare.
- There's a weak incentive not to openly publish the ideas you've gotten from other people because it'll reduce the value of your own, but is there any way to make it stronger?
- Is a setup like this actually a good idea? Open-source has demonstrated that society can earn a lot of value when individuals give up their stakes so as to reduce friction. While this all feels like lower-friction than copyright-as-is, plausibly lowering those frictions will increase participation rate among people who'd otherwise have given their work to everyone for free.
*Credit to Hannu Rajaniemi's Quantum Thief for the word