CS Colloquium: Murat Kararoman
Speaker: Murat Karaorman, Adjunct Professor of Computer Science, College of Creative Studies, UC Santa Barbara
Title: Design by Elicited Consent: From Contracts to Consent in the Age of AI-Built Software
Date: Wednesday, October 21, 2026
Time: 3:30 p.m–4:30 p.m.
Location: 1132 Harold Frank Hall
Host: Ömer Eğecioğlu
Abstract: AI agents now write working software from a conversation. The challenge is guiding exactly what should be built, and understanding what was built. These two gaps, one of specification and one of comprehension, are where AI-built software quietly fails, and producing more code faster does not close them.
Design by Elicited Consent (DbEC) is a novel software engineering method in which a human owner and AI agents reach a recorded agreement before any coding activity. Agents distill intent into plain-language invariants and frame every open question as an explicit decision with options and trade-offs; an independent reviewer audits each round before the human rules. Only then do the agreements compile into executable contracts, checked on every operation of the running system, with each refusal citing the human decision behind it. Charter, the accompanying harness assists in generating the contracts and keeps the whole record in an append-only repository, so any rule can be traced from the running system back to the sentence a person approved.
The ideas have deep roots in this building. Design by Contract and Eiffel language took shape when Bertrand Meyer was on the UCSB faculty. Murat's UCSB Ph.D. research extended Eiffel with concurrency; designed and open-sourced jContractor which brought contracts to Java, and went on to be taught in this department's graduate courses. jMonitor, carried instrumented runtime checking from single calls to sequences of events, an idea Freeman later built into Cydia Substrate, used on millions of jailbroken iPhones. DbEC is the third turn of the original idea, specifications that execute, now governing not only programs but the human and AI process that produces them. This work has just been presented at Meyer's own VerifAI and LASER/FISEE workshops this Fall.
The talk includes a live walk through the record of a real system, and closes with what this approach could mean for how we teach software engineering, for accountable AI, and for emerging regulation that asks AI systems to keep a record of why they act as they do.
Speaker Bio: Murat Karaorman has continuously taught Computer Science in the College of Creative Studies at UC Santa Barbara since 1996. His Ph.D. research in Computer Science at UCSB introduced concurrency to Eiffel, Design by Contract, Active Objects, and Aspect Oriented Programming to Java. In parallel, he has spent three decades building industrial systems: speech recognition and translation at Panasonic; OMAP platform software shipped in billions of mobile devices, as Lead Software Architect at Texas Instruments; Power over Ethernet firmware at Analog Devices; an FDA-approved voice and vision control system for surgical robotics at True Digital Surgery; and affordable 3D surgical imaging as CTO of SB3D Technologies. He holds patents in speech translation, hardware abstraction, and surgical AI.