BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260523T215633EDT-1156vRPiUz@132.216.98.100
DTSTAMP:20260524T015633Z
DESCRIPTION:Abstract\n\nAI systems of varying autonomy are increasingly emb
 edded in how we communicate\, create\, and make decisions across everyday 
 and professional domains. While these technologies offer transformative po
 tential\, they also introduce novel sociotechnical harms\, posing safety r
 isks that differ from those in traditional safety-critical domains. Buildi
 ng on insights from the field of safety engineering\, human-computer inter
 action\, and science and technology studies\, this thesis argues that such
  harms cannot be fully understood or mitigated through model-level interve
 ntions alone. They must be addressed at the system level by examining how 
 AI systems are developed and integrated within broader sociotechnical cont
 exts. System safety\, as part of a new generation of safety engineering ap
 proaches\, offers a valuable lens for understanding safety as an emergent 
 property of complex sociotechnical systems. However\, its approaches have 
 yet to be meaningfully adapted to the context of AI\, where new forms of c
 omplexity and risk demand both translation and expansion. This thesis addr
 esses that gap by operationalizing system safety for AI\, developing actio
 nable approaches for responsible development\, evaluation\, and use of AI 
 systems across their lifecycle.\n\n \n\nUsing a mixed-methods approach\, t
 his thesis examines three critical dimensions for establishing AI system s
 afety practices. First\, at the organizational and compliance level\, we i
 dentify that existing practices for managing risks from AI systems are ad 
 hoc and fragmented. To address this\, we translate hazard analysis framewo
 rks from system safety\, particularly STPA\, to support practitioners in i
 dentifying and mitigating potential hazards early in the AI system develop
 ment process. We demonstrate that our adaptation of STPA offers key afford
 ances for risk management by tracing how sociotechnical harms can emerge f
 rom the interactions among technical components\, organizational processes
 \, and institutional decision-making. Notably\, mitigating these hazards a
 nd harms requires adequate evaluation and monitoring. Second\, at the deve
 loper and evaluation level\, we find that current measurements of ethical 
 AI principles tend to focus on model-level metrics and a narrow subset of 
 harms. Using a system safety lens\, we trace links between these measures\
 , the attributes they assess\, and the types of hazards and harms they sig
 nal. This mapping clarifies which hazards and harms are addressed or overl
 ooked\, supporting the design of evaluation measures that serve as more in
 tentional feedback mechanisms for system-level safety. Regardless of all m
 itigation strategies\, AI systems will have erroneous and inappropriate ou
 tputs. To understand how users respond when this happens\, the third dimen
 sion focuses on the user level. In a controlled experiment involving AI-ba
 sed writing assistants\, we investigate the relationship between different
  types of mental models and user control over AI systems. We find that whi
 le users with more in-depth mental models—those involving how the system w
 orks—find the tool easier to use\, they do not necessarily demonstrate mor
 e effective control over the system\, particularly when the AI generates e
 rroneous suggestions. This highlights the importance of designing human-AI
  interactions that not only support user understanding but also enable use
 rs to translate that understanding into effective responses when failures 
 arise.\n\n \n\nTogether\, these contributions establish a foundation for o
 perationalizing AI system safety across key points of intervention\, hazar
 d mapping\, system evaluations\, and end-user interaction\, and offer a fo
 rward-looking lens for unpacking what safety will require as we move towar
 ds increasingly autonomous AI systems embedded in society.\n
DTSTART:20250812T140000Z
DTEND:20250812T160000Z
LOCATION:Room 603\, McConnell Engineering Building\, CA\, QC\, Montreal\, H
 3A 0E9\, 3480 rue University
SUMMARY:PhD defence of Shalaleh Rismani – Built In\, Not Bolted On: Operati
 onalizing System Safety for AI
URL:/ece/channels/event/phd-defence-shalaleh-rismani-b
 uilt-not-bolted-operationalizing-system-safety-ai-366243
END:VEVENT
END:VCALENDAR