Abstract
Conventionally, transactions of goods over the internet is performed by the use of a trusted intermediary that holds the payment in escrow and settles disputes. They are assumed to behave honestly because of reputation effects; if they misbehave, its users will move to another market. This effect underlies the success of various commerce platforms that have seen a surge in profits over the past decades. However, these systems inherently constitute a privacy risk, in addition to other problems resulting from their market dominance.
In this thesis, we propose a system for fully decentralized commerce involving rational agents that interact using a blockchain. At its core, the system consists of an escrow mechanism that enables both the buyer and the seller to wager money to threaten to invoke a dispute resolution system that determines who were honest. Intuitively, an agent will wager money to settle the dispute only if they think they will win the dispute. This deters an agent from misbehaving, as they can infer that the other agent would threaten them with invoking the dispute resolution system. The dispute resolution system is in turn implemented as a decentralized jury system. Here, the main challenge is to ensure the agents exert an effort to assess the evidence and vote in favor of the honest agent. In a decentralized and anonymous setting, there is no way to hold an agent accountable to their vote, and worse yet, the true state of the dispute is unobservable to the mechanism. We analyze a class of mechanisms that use the wager from the losing agent to compensate only those jurors that made the majority decision. We show that the mechanism is likely to produce good adjudication outcomes under reasonable assumptions. While variations of these ideas are already proposed and deployed in practice, to the best of our knowledge, they lack a thorough and rigorous analysis, making it unclear under which conditions these markets can be assumed secure and/or comply with laws and regulations. By contrast, our system is secure under rather minimal assumptions. Importantly, our system can be combined with recent advances in cryptographic identity management to strike a reasonable balance between privacy and compliance with laws and regulations.
In this thesis, we propose a system for fully decentralized commerce involving rational agents that interact using a blockchain. At its core, the system consists of an escrow mechanism that enables both the buyer and the seller to wager money to threaten to invoke a dispute resolution system that determines who were honest. Intuitively, an agent will wager money to settle the dispute only if they think they will win the dispute. This deters an agent from misbehaving, as they can infer that the other agent would threaten them with invoking the dispute resolution system. The dispute resolution system is in turn implemented as a decentralized jury system. Here, the main challenge is to ensure the agents exert an effort to assess the evidence and vote in favor of the honest agent. In a decentralized and anonymous setting, there is no way to hold an agent accountable to their vote, and worse yet, the true state of the dispute is unobservable to the mechanism. We analyze a class of mechanisms that use the wager from the losing agent to compensate only those jurors that made the majority decision. We show that the mechanism is likely to produce good adjudication outcomes under reasonable assumptions. While variations of these ideas are already proposed and deployed in practice, to the best of our knowledge, they lack a thorough and rigorous analysis, making it unclear under which conditions these markets can be assumed secure and/or comply with laws and regulations. By contrast, our system is secure under rather minimal assumptions. Importantly, our system can be combined with recent advances in cryptographic identity management to strike a reasonable balance between privacy and compliance with laws and regulations.
| Original language | Danish |
|---|
| Publisher | Århus Universitet |
|---|---|
| Publication status | Published - Sept 2023 |