Seminar: Robust Incentive Mechanisms for Blockchain & AI Systems

Abstract: In this talk I focus on two illustrative settings where a new emerging technology creates a novel mechanism design problem. Blockchains offer a decentralized payments and settlement system, without trusted third parties, and with the promise of reduced fees. However, new challenges, stemming from the permissionless anonymous setting, require a robust approach to fee mechanism design. I cover our work in [Gafni & Yaish EC’24] that resolves [Roughgarden EC’21]’s foundational open question to the field. I discuss the mitigations and tradeoffs it implies, together with some more recent results. As another illustrative case, I show how data-sharing protocol design (where Federated Learning is the archetypal example) matters for the security of cooperative AI, given an attack vector we call ``Exclusivity Attacks’’ by self-interested protocol participants [Gafni & Tennenholtz EC’22]. 

Bio: Yotam Gafni received his PhD in Operations Research from the faculty of Data and Decisions Sciences at the Technion, where he was advised by Profs. Ron Lavi and Moshe Tennenholtz. He holds a BA in Math and Philosophy from the Hebrew University. Yotam was a Research Member at the SLMath (MSRI) Berkeley’s Fall 2023 program for Market and Mechanism Design, a Uniswap Foundation Fellow on MEV Auctions, and is currently a postdoc at the Weizmann institute of Science, hosted by Uri Feige and Shahar Dobzinski.